netdata/web/api/queries/query.c

// SPDX-License-Identifier: GPL-3.0-or-later

#include "query.h"
#include "web/api/formatters/rrd2json.h"
#include "rrdr.h"

#include "average/average.h"
#include "countif/countif.h"
#include "incremental_sum/incremental_sum.h"
#include "max/max.h"
#include "median/median.h"
#include "min/min.h"
#include "sum/sum.h"
#include "stddev/stddev.h"
#include "ses/ses.h"
#include "des/des.h"
#include "percentile/percentile.h"
#include "trimmed_mean/trimmed_mean.h"

#define POINTS_TO_EXPAND_QUERY 5

// ----------------------------------------------------------------------------

static struct {
    const char *name;
    uint32_t hash;
    RRDR_TIME_GROUPING value;
    RRDR_TIME_GROUPING add_flush;

    // One time initialization for the module.
    // This is called once, when netdata starts.
    void (*init)(void);

    // Allocate all required structures for a query.
    // This is called once for each netdata query.
    void (*create)(struct rrdresult *r, const char *options);

    // Cleanup collected values, but don't destroy the structures.
    // This is called when the query engine switches dimensions,
    // as part of the same query (so same chart, switching metric).
    void (*reset)(struct rrdresult *r);

    // Free all resources allocated for the query.
    void (*free)(struct rrdresult *r);

    // Add a single value into the calculation.
    // The module may decide to cache it, or use it in the fly.
    void (*add)(struct rrdresult *r, NETDATA_DOUBLE value);

    // Generate a single result for the values added so far.
    // More values and points may be requested later.
    // It is up to the module to reset its internal structures
    // when flushing it (so for a few modules it may be better to
    // continue after a flush as if nothing changed, for others a
    // cleanup of the internal structures may be required).
    NETDATA_DOUBLE (*flush)(struct rrdresult *r, RRDR_VALUE_FLAGS *rrdr_value_options_ptr);

    TIER_QUERY_FETCH tier_query_fetch;
} api_v1_data_groups[] = {
        {.name = "average",
                .hash  = 0,
                .value = RRDR_GROUPING_AVERAGE,
                .add_flush = RRDR_GROUPING_AVERAGE,
                .init  = NULL,
                .create= tg_average_create,
                .reset = tg_average_reset,
                .free  = tg_average_free,
                .add   = tg_average_add,
                .flush = tg_average_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "avg",                             // alias on 'average'
                .hash  = 0,
                .value = RRDR_GROUPING_AVERAGE,
                .add_flush = RRDR_GROUPING_AVERAGE,
                .init  = NULL,
                .create= tg_average_create,
                .reset = tg_average_reset,
                .free  = tg_average_free,
                .add   = tg_average_add,
                .flush = tg_average_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "mean",                            // alias on 'average'
                .hash  = 0,
                .value = RRDR_GROUPING_AVERAGE,
                .add_flush = RRDR_GROUPING_AVERAGE,
                .init  = NULL,
                .create= tg_average_create,
                .reset = tg_average_reset,
                .free  = tg_average_free,
                .add   = tg_average_add,
                .flush = tg_average_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean1",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN1,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_1,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean2",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN2,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_2,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean3",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN3,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_3,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean5",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_5,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean10",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN10,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_10,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean15",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN15,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_15,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean20",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN20,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_20,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean25",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN25,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_25,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-mean",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEAN,
                .add_flush = RRDR_GROUPING_TRIMMED_MEAN,
                .init  = NULL,
                .create= tg_trimmed_mean_create_5,
                .reset = tg_trimmed_mean_reset,
                .free  = tg_trimmed_mean_free,
                .add   = tg_trimmed_mean_add,
                .flush = tg_trimmed_mean_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name  = "incremental_sum",
                .hash  = 0,
                .value = RRDR_GROUPING_INCREMENTAL_SUM,
                .add_flush = RRDR_GROUPING_INCREMENTAL_SUM,
                .init  = NULL,
                .create= tg_incremental_sum_create,
                .reset = tg_incremental_sum_reset,
                .free  = tg_incremental_sum_free,
                .add   = tg_incremental_sum_add,
                .flush = tg_incremental_sum_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "incremental-sum",
                .hash  = 0,
                .value = RRDR_GROUPING_INCREMENTAL_SUM,
                .add_flush = RRDR_GROUPING_INCREMENTAL_SUM,
                .init  = NULL,
                .create= tg_incremental_sum_create,
                .reset = tg_incremental_sum_reset,
                .free  = tg_incremental_sum_free,
                .add   = tg_incremental_sum_add,
                .flush = tg_incremental_sum_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "median",
                .hash  = 0,
                .value = RRDR_GROUPING_MEDIAN,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median1",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN1,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_1,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median2",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN2,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_2,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median3",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN3,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_3,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median5",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN5,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_5,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median10",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN10,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_10,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median15",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN15,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_15,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median20",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN20,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_20,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median25",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN25,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_25,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "trimmed-median",
                .hash  = 0,
                .value = RRDR_GROUPING_TRIMMED_MEDIAN5,
                .add_flush = RRDR_GROUPING_MEDIAN,
                .init  = NULL,
                .create= tg_median_create_trimmed_5,
                .reset = tg_median_reset,
                .free  = tg_median_free,
                .add   = tg_median_add,
                .flush = tg_median_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile25",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE25,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_25,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile50",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE50,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_50,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile75",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE75,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_75,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile80",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE80,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_80,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile90",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE90,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_90,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile95",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_95,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile97",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE97,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_97,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile98",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE98,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_98,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile99",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE99,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_99,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "percentile",
                .hash  = 0,
                .value = RRDR_GROUPING_PERCENTILE,
                .add_flush = RRDR_GROUPING_PERCENTILE,
                .init  = NULL,
                .create= tg_percentile_create_95,
                .reset = tg_percentile_reset,
                .free  = tg_percentile_free,
                .add   = tg_percentile_add,
                .flush = tg_percentile_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "min",
                .hash  = 0,
                .value = RRDR_GROUPING_MIN,
                .add_flush = RRDR_GROUPING_MIN,
                .init  = NULL,
                .create= tg_min_create,
                .reset = tg_min_reset,
                .free  = tg_min_free,
                .add   = tg_min_add,
                .flush = tg_min_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_MIN
        },
        {.name = "max",
                .hash  = 0,
                .value = RRDR_GROUPING_MAX,
                .add_flush = RRDR_GROUPING_MAX,
                .init  = NULL,
                .create= tg_max_create,
                .reset = tg_max_reset,
                .free  = tg_max_free,
                .add   = tg_max_add,
                .flush = tg_max_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_MAX
        },
        {.name = "sum",
                .hash  = 0,
                .value = RRDR_GROUPING_SUM,
                .add_flush = RRDR_GROUPING_SUM,
                .init  = NULL,
                .create= tg_sum_create,
                .reset = tg_sum_reset,
                .free  = tg_sum_free,
                .add   = tg_sum_add,
                .flush = tg_sum_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_SUM
        },

        // standard deviation
        {.name = "stddev",
                .hash  = 0,
                .value = RRDR_GROUPING_STDDEV,
                .add_flush = RRDR_GROUPING_STDDEV,
                .init  = NULL,
                .create= tg_stddev_create,
                .reset = tg_stddev_reset,
                .free  = tg_stddev_free,
                .add   = tg_stddev_add,
                .flush = tg_stddev_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "cv",                           // coefficient variation is calculated by stddev
                .hash  = 0,
                .value = RRDR_GROUPING_CV,
                .add_flush = RRDR_GROUPING_CV,
                .init  = NULL,
                .create= tg_stddev_create, // not an error, stddev calculates this too
                .reset = tg_stddev_reset,  // not an error, stddev calculates this too
                .free  = tg_stddev_free,   // not an error, stddev calculates this too
                .add   = tg_stddev_add,    // not an error, stddev calculates this too
                .flush = tg_stddev_coefficient_of_variation_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "rsd",                          // alias of 'cv'
                .hash  = 0,
                .value = RRDR_GROUPING_CV,
                .add_flush = RRDR_GROUPING_CV,
                .init  = NULL,
                .create= tg_stddev_create, // not an error, stddev calculates this too
                .reset = tg_stddev_reset,  // not an error, stddev calculates this too
                .free  = tg_stddev_free,   // not an error, stddev calculates this too
                .add   = tg_stddev_add,    // not an error, stddev calculates this too
                .flush = tg_stddev_coefficient_of_variation_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },

        // single exponential smoothing
        {.name = "ses",
                .hash  = 0,
                .value = RRDR_GROUPING_SES,
                .add_flush = RRDR_GROUPING_SES,
                .init  = tg_ses_init,
                .create= tg_ses_create,
                .reset = tg_ses_reset,
                .free  = tg_ses_free,
                .add   = tg_ses_add,
                .flush = tg_ses_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "ema",                         // alias for 'ses'
                .hash  = 0,
                .value = RRDR_GROUPING_SES,
                .add_flush = RRDR_GROUPING_SES,
                .init  = NULL,
                .create= tg_ses_create,
                .reset = tg_ses_reset,
                .free  = tg_ses_free,
                .add   = tg_ses_add,
                .flush = tg_ses_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },
        {.name = "ewma",                        // alias for ses
                .hash  = 0,
                .value = RRDR_GROUPING_SES,
                .add_flush = RRDR_GROUPING_SES,
                .init  = NULL,
                .create= tg_ses_create,
                .reset = tg_ses_reset,
                .free  = tg_ses_free,
                .add   = tg_ses_add,
                .flush = tg_ses_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },

        // double exponential smoothing
        {.name = "des",
                .hash  = 0,
                .value = RRDR_GROUPING_DES,
                .add_flush = RRDR_GROUPING_DES,
                .init  = tg_des_init,
                .create= tg_des_create,
                .reset = tg_des_reset,
                .free  = tg_des_free,
                .add   = tg_des_add,
                .flush = tg_des_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },

        {.name = "countif",
                .hash  = 0,
                .value = RRDR_GROUPING_COUNTIF,
                .add_flush = RRDR_GROUPING_COUNTIF,
                .init = NULL,
                .create= tg_countif_create,
                .reset = tg_countif_reset,
                .free  = tg_countif_free,
                .add   = tg_countif_add,
                .flush = tg_countif_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        },

        // terminator
        {.name = NULL,
                .hash  = 0,
                .value = RRDR_GROUPING_UNDEFINED,
                .add_flush = RRDR_GROUPING_AVERAGE,
                .init = NULL,
                .create= tg_average_create,
                .reset = tg_average_reset,
                .free  = tg_average_free,
                .add   = tg_average_add,
                .flush = tg_average_flush,
                .tier_query_fetch = TIER_QUERY_FETCH_AVERAGE
        }
};

void time_grouping_init(void) {
    int i;

    for(i = 0; api_v1_data_groups[i].name ; i++) {
        api_v1_data_groups[i].hash = simple_hash(api_v1_data_groups[i].name);

        if(api_v1_data_groups[i].init)
            api_v1_data_groups[i].init();
    }
}

const char *time_grouping_method2string(RRDR_TIME_GROUPING group) {
    int i;

    for(i = 0; api_v1_data_groups[i].name ; i++) {
        if(api_v1_data_groups[i].value == group) {
            return api_v1_data_groups[i].name;
        }
    }

    return "unknown-group-method";
}

RRDR_TIME_GROUPING time_grouping_parse(const char *name, RRDR_TIME_GROUPING def) {
    int i;

    uint32_t hash = simple_hash(name);
    for(i = 0; api_v1_data_groups[i].name ; i++)
        if(unlikely(hash == api_v1_data_groups[i].hash && !strcmp(name, api_v1_data_groups[i].name)))
            return api_v1_data_groups[i].value;

    return def;
}

const char *time_grouping_tostring(RRDR_TIME_GROUPING group) {
    int i;

    for(i = 0; api_v1_data_groups[i].name ; i++)
        if(unlikely(group == api_v1_data_groups[i].value))
            return api_v1_data_groups[i].name;

    return "unknown";
}

static void rrdr_set_grouping_function(RRDR *r, RRDR_TIME_GROUPING group_method) {
    int i, found = 0;
    for(i = 0; !found && api_v1_data_groups[i].name ;i++) {
        if(api_v1_data_groups[i].value == group_method) {
            r->time_grouping.create  = api_v1_data_groups[i].create;
            r->time_grouping.reset   = api_v1_data_groups[i].reset;
            r->time_grouping.free    = api_v1_data_groups[i].free;
            r->time_grouping.add     = api_v1_data_groups[i].add;
            r->time_grouping.flush   = api_v1_data_groups[i].flush;
            r->time_grouping.tier_query_fetch = api_v1_data_groups[i].tier_query_fetch;
            r->time_grouping.add_flush = api_v1_data_groups[i].add_flush;
            found = 1;
        }
    }
    if(!found) {
        errno = 0;
        internal_error(true, "QUERY: grouping method %u not found. Using 'average'", (unsigned int)group_method);
        r->time_grouping.create  = tg_average_create;
        r->time_grouping.reset   = tg_average_reset;
        r->time_grouping.free    = tg_average_free;
        r->time_grouping.add     = tg_average_add;
        r->time_grouping.flush   = tg_average_flush;
        r->time_grouping.tier_query_fetch = TIER_QUERY_FETCH_AVERAGE;
        r->time_grouping.add_flush = RRDR_GROUPING_AVERAGE;
    }
}

static inline void time_grouping_add(RRDR *r, NETDATA_DOUBLE value, const RRDR_TIME_GROUPING add_flush) {
    switch(add_flush) {
        case RRDR_GROUPING_AVERAGE:
            tg_average_add(r, value);
            break;

        case RRDR_GROUPING_MAX:
            tg_max_add(r, value);
            break;

        case RRDR_GROUPING_MIN:
            tg_min_add(r, value);
            break;

        case RRDR_GROUPING_MEDIAN:
            tg_median_add(r, value);
            break;

        case RRDR_GROUPING_STDDEV:
        case RRDR_GROUPING_CV:
            tg_stddev_add(r, value);
            break;

        case RRDR_GROUPING_SUM:
            tg_sum_add(r, value);
            break;

        case RRDR_GROUPING_COUNTIF:
            tg_countif_add(r, value);
            break;

        case RRDR_GROUPING_TRIMMED_MEAN:
            tg_trimmed_mean_add(r, value);
            break;

        case RRDR_GROUPING_PERCENTILE:
            tg_percentile_add(r, value);
            break;

        case RRDR_GROUPING_SES:
            tg_ses_add(r, value);
            break;

        case RRDR_GROUPING_DES:
            tg_des_add(r, value);
            break;

        case RRDR_GROUPING_INCREMENTAL_SUM:
            tg_incremental_sum_add(r, value);
            break;

        default:
            r->time_grouping.add(r, value);
            break;
    }
}

static inline NETDATA_DOUBLE time_grouping_flush(RRDR *r, RRDR_VALUE_FLAGS *rrdr_value_options_ptr, const RRDR_TIME_GROUPING add_flush) {
    switch(add_flush) {
        case RRDR_GROUPING_AVERAGE:
            return tg_average_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_MAX:
            return tg_max_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_MIN:
            return tg_min_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_MEDIAN:
            return tg_median_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_STDDEV:
            return tg_stddev_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_CV:
            return tg_stddev_coefficient_of_variation_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_SUM:
            return tg_sum_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_COUNTIF:
            return tg_countif_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_TRIMMED_MEAN:
            return tg_trimmed_mean_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_PERCENTILE:
            return tg_percentile_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_SES:
            return tg_ses_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_DES:
            return tg_des_flush(r, rrdr_value_options_ptr);

        case RRDR_GROUPING_INCREMENTAL_SUM:
            return tg_incremental_sum_flush(r, rrdr_value_options_ptr);

        default:
            return r->time_grouping.flush(r, rrdr_value_options_ptr);
    }
}

RRDR_GROUP_BY group_by_parse(char *s) {
    RRDR_GROUP_BY group_by = RRDR_GROUP_BY_NONE;

    while(s) {
        char *key = strsep_skip_consecutive_separators(&s, ",| ");
        if (!key || !*key) continue;

        if (strcmp(key, "selected") == 0)
            group_by |= RRDR_GROUP_BY_SELECTED;

        if (strcmp(key, "dimension") == 0)
            group_by |= RRDR_GROUP_BY_DIMENSION;

        if (strcmp(key, "instance") == 0)
            group_by |= RRDR_GROUP_BY_INSTANCE;

        if (strcmp(key, "percentage-of-instance") == 0)
            group_by |= RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE;

        if (strcmp(key, "label") == 0)
            group_by |= RRDR_GROUP_BY_LABEL;

        if (strcmp(key, "node") == 0)
            group_by |= RRDR_GROUP_BY_NODE;

        if (strcmp(key, "context") == 0)
            group_by |= RRDR_GROUP_BY_CONTEXT;

        if (strcmp(key, "units") == 0)
            group_by |= RRDR_GROUP_BY_UNITS;
    }

    if((group_by & RRDR_GROUP_BY_SELECTED) && (group_by & ~RRDR_GROUP_BY_SELECTED)) {
        internal_error(true, "group-by given by query has 'selected' together with more groupings");
        group_by = RRDR_GROUP_BY_SELECTED; // remove all other groupings
    }

    if(group_by & RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE)
        group_by = RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE; // remove all other groupings

    return group_by;
}

void buffer_json_group_by_to_array(BUFFER *wb, RRDR_GROUP_BY group_by) {
    if(group_by == RRDR_GROUP_BY_NONE)
        buffer_json_add_array_item_string(wb, "none");
    else {
        if (group_by & RRDR_GROUP_BY_DIMENSION)
            buffer_json_add_array_item_string(wb, "dimension");

        if (group_by & RRDR_GROUP_BY_INSTANCE)
            buffer_json_add_array_item_string(wb, "instance");

        if (group_by & RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE)
            buffer_json_add_array_item_string(wb, "percentage-of-instance");

        if (group_by & RRDR_GROUP_BY_LABEL)
            buffer_json_add_array_item_string(wb, "label");

        if (group_by & RRDR_GROUP_BY_NODE)
            buffer_json_add_array_item_string(wb, "node");

        if (group_by & RRDR_GROUP_BY_CONTEXT)
            buffer_json_add_array_item_string(wb, "context");

        if (group_by & RRDR_GROUP_BY_UNITS)
            buffer_json_add_array_item_string(wb, "units");

        if (group_by & RRDR_GROUP_BY_SELECTED)
            buffer_json_add_array_item_string(wb, "selected");
    }
}

RRDR_GROUP_BY_FUNCTION group_by_aggregate_function_parse(const char *s) {
    if(strcmp(s, "average") == 0)
        return RRDR_GROUP_BY_FUNCTION_AVERAGE;

    if(strcmp(s, "avg") == 0)
        return RRDR_GROUP_BY_FUNCTION_AVERAGE;

    if(strcmp(s, "min") == 0)
        return RRDR_GROUP_BY_FUNCTION_MIN;

    if(strcmp(s, "max") == 0)
        return RRDR_GROUP_BY_FUNCTION_MAX;

    if(strcmp(s, "sum") == 0)
        return RRDR_GROUP_BY_FUNCTION_SUM;

    return RRDR_GROUP_BY_FUNCTION_AVERAGE;
}

const char *group_by_aggregate_function_to_string(RRDR_GROUP_BY_FUNCTION group_by_function) {
    switch(group_by_function) {
        default:
        case RRDR_GROUP_BY_FUNCTION_AVERAGE:
            return "average";

        case RRDR_GROUP_BY_FUNCTION_MIN:
            return "min";

        case RRDR_GROUP_BY_FUNCTION_MAX:
            return "max";

        case RRDR_GROUP_BY_FUNCTION_SUM:
            return "sum";
    }
}

// ----------------------------------------------------------------------------
// helpers to find our way in RRDR

static inline RRDR_VALUE_FLAGS *UNUSED_FUNCTION(rrdr_line_options)(RRDR *r, long rrdr_line) {
    return &r->o[ rrdr_line * r->d ];
}

static inline NETDATA_DOUBLE *UNUSED_FUNCTION(rrdr_line_values)(RRDR *r, long rrdr_line) {
    return &r->v[ rrdr_line * r->d ];
}

static inline long rrdr_line_init(RRDR *r __maybe_unused, time_t t __maybe_unused, long rrdr_line) {
    rrdr_line++;

    internal_fatal(rrdr_line >= (long)r->n,
                   "QUERY: requested to step above RRDR size for query '%s'",
                   r->internal.qt->id);

    internal_fatal(r->t[rrdr_line] != t,
                   "QUERY: wrong timestamp at RRDR line %ld, expected %ld, got %ld, of query '%s'",
                   rrdr_line, r->t[rrdr_line], t, r->internal.qt->id);

    return rrdr_line;
}

// ----------------------------------------------------------------------------
// tier management

static bool query_metric_is_valid_tier(QUERY_METRIC *qm, size_t tier) {
    if(!qm->tiers[tier].db_metric_handle || !qm->tiers[tier].db_first_time_s || !qm->tiers[tier].db_last_time_s || !qm->tiers[tier].db_update_every_s)
        return false;

    return true;
}

static size_t query_metric_first_working_tier(QUERY_METRIC *qm) {
    for(size_t tier = 0; tier < storage_tiers ; tier++) {

        // find the db time-range for this tier for all metrics
        STORAGE_METRIC_HANDLE *db_metric_handle = qm->tiers[tier].db_metric_handle;
        time_t first_time_s = qm->tiers[tier].db_first_time_s;
        time_t last_time_s  = qm->tiers[tier].db_last_time_s;
        time_t update_every_s = qm->tiers[tier].db_update_every_s;

        if(!db_metric_handle || !first_time_s || !last_time_s || !update_every_s)
            continue;

        return tier;
    }

    return 0;
}

static long query_plan_points_coverage_weight(time_t db_first_time_s, time_t db_last_time_s, time_t db_update_every_s, time_t after_wanted, time_t before_wanted, size_t points_wanted, size_t tier __maybe_unused) {
    if(db_first_time_s == 0 ||
        db_last_time_s == 0 ||
        db_update_every_s == 0 ||
        db_first_time_s > before_wanted ||
        db_last_time_s < after_wanted)
        return -LONG_MAX;

    long long common_first_t = MAX(db_first_time_s, after_wanted);
    long long common_last_t = MIN(db_last_time_s, before_wanted);

    long long time_coverage = (common_last_t - common_first_t) * 1000000LL / (before_wanted - after_wanted);
    long long points_wanted_in_coverage = (long long)points_wanted * time_coverage / 1000000LL;

    long long points_available = (common_last_t - common_first_t) / db_update_every_s;
    long long points_delta = (long)(points_available - points_wanted_in_coverage);
    long long points_coverage = (points_delta < 0) ? (long)(points_available * time_coverage / points_wanted_in_coverage) : time_coverage;

    // a way to benefit higher tiers
    // points_coverage += (long)tier * 10000;

    if(points_available <= 0)
        return -LONG_MAX;

    return (long)(points_coverage + (25000LL * tier)); // 2.5% benefit for each higher tier
}

static size_t query_metric_best_tier_for_timeframe(QUERY_METRIC *qm, time_t after_wanted, time_t before_wanted, size_t points_wanted) {
    if(unlikely(storage_tiers < 2))
        return 0;

    if(unlikely(after_wanted == before_wanted || points_wanted <= 0))
        return query_metric_first_working_tier(qm);

    time_t min_first_time_s = 0;
    time_t max_last_time_s = 0;

    for(size_t tier = 0; tier < storage_tiers ; tier++) {
        time_t first_time_s = qm->tiers[tier].db_first_time_s;
        time_t last_time_s  = qm->tiers[tier].db_last_time_s;

        if(!min_first_time_s || (first_time_s && first_time_s < min_first_time_s))
            min_first_time_s = first_time_s;

        if(!max_last_time_s || (last_time_s && last_time_s > max_last_time_s))
            max_last_time_s = last_time_s;
    }

    for(size_t tier = 0; tier < storage_tiers ; tier++) {

        // find the db time-range for this tier for all metrics
        STORAGE_METRIC_HANDLE *db_metric_handle = qm->tiers[tier].db_metric_handle;
        time_t first_time_s = qm->tiers[tier].db_first_time_s;
        time_t last_time_s  = qm->tiers[tier].db_last_time_s;
        time_t update_every_s = qm->tiers[tier].db_update_every_s;

        if( !db_metric_handle ||
            !first_time_s ||
            !last_time_s ||
            !update_every_s ||
            first_time_s > before_wanted ||
            last_time_s < after_wanted
            ) {
            qm->tiers[tier].weight = -LONG_MAX;
            continue;
        }

        internal_fatal(first_time_s > before_wanted || last_time_s < after_wanted, "QUERY: invalid db durations");

        qm->tiers[tier].weight = query_plan_points_coverage_weight(
                min_first_time_s, max_last_time_s, update_every_s,
                after_wanted, before_wanted, points_wanted, tier);
    }

    size_t best_tier = 0;
    for(size_t tier = 1; tier < storage_tiers ; tier++) {
        if(qm->tiers[tier].weight >= qm->tiers[best_tier].weight)
            best_tier = tier;
    }

    return best_tier;
}

static size_t rrddim_find_best_tier_for_timeframe(QUERY_TARGET *qt, time_t after_wanted, time_t before_wanted, size_t points_wanted) {
    if(unlikely(storage_tiers < 2))
        return 0;

    if(unlikely(after_wanted == before_wanted || points_wanted <= 0)) {
        internal_error(true, "QUERY: '%s' has invalid params to tier calculation", qt->id);
        return 0;
    }

    long weight[storage_tiers];

    for(size_t tier = 0; tier < storage_tiers ; tier++) {

        time_t common_first_time_s = 0;
        time_t common_last_time_s = 0;
        time_t common_update_every_s = 0;

        // find the db time-range for this tier for all metrics
        for(size_t i = 0, used = qt->query.used; i < used ; i++) {
            QUERY_METRIC *qm = query_metric(qt, i);

            time_t first_time_s = qm->tiers[tier].db_first_time_s;
            time_t last_time_s  = qm->tiers[tier].db_last_time_s;
            time_t update_every_s = qm->tiers[tier].db_update_every_s;

            if(!first_time_s || !last_time_s || !update_every_s)
                continue;

            if(!common_first_time_s)
                common_first_time_s = first_time_s;
            else
                common_first_time_s = MIN(first_time_s, common_first_time_s);

            if(!common_last_time_s)
                common_last_time_s = last_time_s;
            else
                common_last_time_s = MAX(last_time_s, common_last_time_s);

            if(!common_update_every_s)
                common_update_every_s = update_every_s;
            else
                common_update_every_s = MIN(update_every_s, common_update_every_s);
        }

        weight[tier] = query_plan_points_coverage_weight(common_first_time_s, common_last_time_s, common_update_every_s, after_wanted, before_wanted, points_wanted, tier);
    }

    size_t best_tier = 0;
    for(size_t tier = 1; tier < storage_tiers ; tier++) {
        if(weight[tier] >= weight[best_tier])
            best_tier = tier;
    }

    if(weight[best_tier] == -LONG_MAX)
        best_tier = 0;

    return best_tier;
}

static time_t rrdset_find_natural_update_every_for_timeframe(QUERY_TARGET *qt, time_t after_wanted, time_t before_wanted, size_t points_wanted, RRDR_OPTIONS options, size_t tier) {
    size_t best_tier;
    if((options & RRDR_OPTION_SELECTED_TIER) && tier < storage_tiers)
        best_tier = tier;
    else
        best_tier = rrddim_find_best_tier_for_timeframe(qt, after_wanted, before_wanted, points_wanted);

    // find the db minimum update every for this tier for all metrics
    time_t common_update_every_s = default_rrd_update_every;
    for(size_t i = 0, used = qt->query.used; i < used ; i++) {
        QUERY_METRIC *qm = query_metric(qt, i);

        time_t update_every_s = qm->tiers[best_tier].db_update_every_s;

        if(!i)
            common_update_every_s = update_every_s;
        else
            common_update_every_s = MIN(update_every_s, common_update_every_s);
    }

    return common_update_every_s;
}

// ----------------------------------------------------------------------------
// query ops

typedef struct query_point {
    STORAGE_POINT sp;
    NETDATA_DOUBLE value;
    bool added;
#ifdef NETDATA_INTERNAL_CHECKS
    size_t id;
#endif
} QUERY_POINT;

QUERY_POINT QUERY_POINT_EMPTY = {
        .sp = STORAGE_POINT_UNSET,
        .value = NAN,
        .added = false,
#ifdef NETDATA_INTERNAL_CHECKS
        .id = 0,
#endif
};

#ifdef NETDATA_INTERNAL_CHECKS
#define query_point_set_id(point, point_id) (point).id = point_id
#else
#define query_point_set_id(point, point_id) debug_dummy()
#endif

typedef struct query_engine_ops {
    // configuration
    RRDR *r;
    QUERY_METRIC *qm;
    time_t view_update_every;
    time_t query_granularity;
    TIER_QUERY_FETCH tier_query_fetch;

    // query planer
    size_t current_plan;
    time_t current_plan_expire_time;
    time_t plan_expanded_after;
    time_t plan_expanded_before;

    // storage queries
    size_t tier;
    struct query_metric_tier *tier_ptr;
    struct storage_engine_query_handle *handle;

    // aggregating points over time
    size_t group_points_non_zero;
    size_t group_points_added;
    STORAGE_POINT group_point;          // aggregates min, max, sum, count, anomaly count for each group point
    STORAGE_POINT query_point;          // aggregates min, max, sum, count, anomaly count across the whole query
    RRDR_VALUE_FLAGS group_value_flags;

    // statistics
    size_t db_total_points_read;
    size_t db_points_read_per_tier[RRD_STORAGE_TIERS];

    struct {
        time_t expanded_after;
        time_t expanded_before;
        struct storage_engine_query_handle handle;
        bool initialized;
        bool finalized;
    } plans[QUERY_PLANS_MAX];

    struct query_engine_ops *next;
} QUERY_ENGINE_OPS;


// ----------------------------------------------------------------------------
// query planer

#define query_plan_should_switch_plan(ops, now) ((now) >= (ops)->current_plan_expire_time)

static size_t query_planer_expand_duration_in_points(time_t this_update_every, time_t next_update_every) {

    time_t delta = this_update_every - next_update_every;
    if(delta < 0) delta = -delta;

    size_t points;
    if(delta < this_update_every * POINTS_TO_EXPAND_QUERY)
        points = POINTS_TO_EXPAND_QUERY;
    else
        points = (delta + this_update_every - 1) / this_update_every;

    return points;
}

static void query_planer_initialize_plans(QUERY_ENGINE_OPS *ops) {
    QUERY_METRIC *qm = ops->qm;

    for(size_t p = 0; p < qm->plan.used ; p++) {
        size_t tier = qm->plan.array[p].tier;
        time_t update_every = qm->tiers[tier].db_update_every_s;

        size_t points_to_add_to_after;
        if(p > 0) {
            // there is another plan before to this

            size_t tier0 = qm->plan.array[p - 1].tier;
            time_t update_every0 = qm->tiers[tier0].db_update_every_s;

            points_to_add_to_after = query_planer_expand_duration_in_points(update_every, update_every0);
        }
        else
            points_to_add_to_after = (tier == 0) ? 0 : POINTS_TO_EXPAND_QUERY;

        size_t points_to_add_to_before;
        if(p + 1 < qm->plan.used) {
            // there is another plan after to this

            size_t tier1 = qm->plan.array[p+1].tier;
            time_t update_every1 = qm->tiers[tier1].db_update_every_s;

            points_to_add_to_before = query_planer_expand_duration_in_points(update_every, update_every1);
        }
        else
            points_to_add_to_before = POINTS_TO_EXPAND_QUERY;

        time_t after = qm->plan.array[p].after - (time_t)(update_every * points_to_add_to_after);
        time_t before = qm->plan.array[p].before + (time_t)(update_every * points_to_add_to_before);

        ops->plans[p].expanded_after = after;
        ops->plans[p].expanded_before = before;

        ops->r->internal.qt->db.tiers[tier].queries++;

        struct query_metric_tier *tier_ptr = &qm->tiers[tier];
        STORAGE_ENGINE *eng = query_metric_storage_engine(ops->r->internal.qt, qm, tier);
        storage_engine_query_init(eng->backend, tier_ptr->db_metric_handle, &ops->plans[p].handle,
                after, before, ops->r->internal.qt->request.priority);

        ops->plans[p].initialized = true;
        ops->plans[p].finalized = false;
    }
}

static void query_planer_finalize_plan(QUERY_ENGINE_OPS *ops, size_t plan_id) {
    // QUERY_METRIC *qm = ops->qm;

    if(ops->plans[plan_id].initialized && !ops->plans[plan_id].finalized) {
        storage_engine_query_finalize(&ops->plans[plan_id].handle);
        ops->plans[plan_id].initialized = false;
        ops->plans[plan_id].finalized = true;
    }
}

static void query_planer_finalize_remaining_plans(QUERY_ENGINE_OPS *ops) {
    QUERY_METRIC *qm = ops->qm;

    for(size_t p = 0; p < qm->plan.used ; p++)
        query_planer_finalize_plan(ops, p);
}

static void query_planer_activate_plan(QUERY_ENGINE_OPS *ops, size_t plan_id, time_t overwrite_after __maybe_unused) {
    QUERY_METRIC *qm = ops->qm;

    internal_fatal(plan_id >= qm->plan.used, "QUERY: invalid plan_id given");
    internal_fatal(!ops->plans[plan_id].initialized, "QUERY: plan has not been initialized");
    internal_fatal(ops->plans[plan_id].finalized, "QUERY: plan has been finalized");

    internal_fatal(qm->plan.array[plan_id].after > qm->plan.array[plan_id].before, "QUERY: flipped after/before");

    ops->tier = qm->plan.array[plan_id].tier;
    ops->tier_ptr = &qm->tiers[ops->tier];
    ops->handle = &ops->plans[plan_id].handle;
    ops->current_plan = plan_id;

    if(plan_id + 1 < qm->plan.used && qm->plan.array[plan_id + 1].after < qm->plan.array[plan_id].before)
        ops->current_plan_expire_time = qm->plan.array[plan_id + 1].after;
    else
        ops->current_plan_expire_time = qm->plan.array[plan_id].before;

    ops->plan_expanded_after = ops->plans[plan_id].expanded_after;
    ops->plan_expanded_before = ops->plans[plan_id].expanded_before;
}

static bool query_planer_next_plan(QUERY_ENGINE_OPS *ops, time_t now, time_t last_point_end_time) {
    QUERY_METRIC *qm = ops->qm;

    size_t old_plan = ops->current_plan;

    time_t next_plan_before_time;
    do {
        ops->current_plan++;

        if (ops->current_plan >= qm->plan.used) {
            ops->current_plan = old_plan;
            ops->current_plan_expire_time = ops->r->internal.qt->window.before;
            // let the query run with current plan
            // we will not switch it
            return false;
        }

        next_plan_before_time = qm->plan.array[ops->current_plan].before;
    } while(now >= next_plan_before_time || last_point_end_time >= next_plan_before_time);

    if(!query_metric_is_valid_tier(qm, qm->plan.array[ops->current_plan].tier)) {
        ops->current_plan = old_plan;
        ops->current_plan_expire_time = ops->r->internal.qt->window.before;
        return false;
    }

    query_planer_finalize_plan(ops, old_plan);
    query_planer_activate_plan(ops, ops->current_plan, MIN(now, last_point_end_time));
    return true;
}

static int compare_query_plan_entries_on_start_time(const void *a, const void *b) {
    QUERY_PLAN_ENTRY *p1 = (QUERY_PLAN_ENTRY *)a;
    QUERY_PLAN_ENTRY *p2 = (QUERY_PLAN_ENTRY *)b;
    return (p1->after < p2->after)?-1:1;
}

static bool query_plan(QUERY_ENGINE_OPS *ops, time_t after_wanted, time_t before_wanted, size_t points_wanted) {
    QUERY_METRIC *qm = ops->qm;

    // put our selected tier as the first plan
    size_t selected_tier;
    bool switch_tiers = true;

    if((ops->r->internal.qt->window.options & RRDR_OPTION_SELECTED_TIER)
       && ops->r->internal.qt->window.tier < storage_tiers
       && query_metric_is_valid_tier(qm, ops->r->internal.qt->window.tier)) {
        selected_tier = ops->r->internal.qt->window.tier;
        switch_tiers = false;
    }
    else {
        selected_tier = query_metric_best_tier_for_timeframe(qm, after_wanted, before_wanted, points_wanted);

        if(!query_metric_is_valid_tier(qm, selected_tier))
            return false;

        if(qm->tiers[selected_tier].db_first_time_s > before_wanted ||
           qm->tiers[selected_tier].db_last_time_s < after_wanted)
            return false;
    }

    qm->plan.used = 1;
    qm->plan.array[0].tier = selected_tier;
    qm->plan.array[0].after = (qm->tiers[selected_tier].db_first_time_s < after_wanted) ? after_wanted : qm->tiers[selected_tier].db_first_time_s;
    qm->plan.array[0].before = (qm->tiers[selected_tier].db_last_time_s > before_wanted) ? before_wanted : qm->tiers[selected_tier].db_last_time_s;

    if(switch_tiers) {
        // the selected tier
        time_t selected_tier_first_time_s = qm->plan.array[0].after;
        time_t selected_tier_last_time_s = qm->plan.array[0].before;

        // check if our selected tier can start the query
        if (selected_tier_first_time_s > after_wanted) {
            // we need some help from other tiers
            for (size_t tr = (int)selected_tier + 1; tr < storage_tiers && qm->plan.used < QUERY_PLANS_MAX ; tr++) {
                if(!query_metric_is_valid_tier(qm, tr))
                    continue;

                // find the first time of this tier
                time_t tier_first_time_s = qm->tiers[tr].db_first_time_s;

                // can it help?
                if (tier_first_time_s < selected_tier_first_time_s) {
                    // it can help us add detail at the beginning of the query
                    QUERY_PLAN_ENTRY t = {
                        .tier = tr,
                        .after = (tier_first_time_s < after_wanted) ? after_wanted : tier_first_time_s,
                        .before = selected_tier_first_time_s,
                    };
                    ops->plans[qm->plan.used].initialized = false;
                    ops->plans[qm->plan.used].finalized = false;
                    qm->plan.array[qm->plan.used++] = t;

                    internal_fatal(!t.after || !t.before, "QUERY: invalid plan selected");

                    // prepare for the tier
                    selected_tier_first_time_s = t.after;

                    if (t.after <= after_wanted)
                        break;
                }
            }
        }

        // check if our selected tier can finish the query
        if (selected_tier_last_time_s < before_wanted) {
            // we need some help from other tiers
            for (int tr = (int)selected_tier - 1; tr >= 0 && qm->plan.used < QUERY_PLANS_MAX ; tr--) {
                if(!query_metric_is_valid_tier(qm, tr))
                    continue;

                // find the last time of this tier
                time_t tier_last_time_s = qm->tiers[tr].db_last_time_s;

                //buffer_sprintf(wb, ": EVAL BEFORE tier %d, %ld", tier, last_time_s);

                // can it help?
                if (tier_last_time_s > selected_tier_last_time_s) {
                    // it can help us add detail at the end of the query
                    QUERY_PLAN_ENTRY t = {
                        .tier = tr,
                        .after = selected_tier_last_time_s,
                        .before = (tier_last_time_s > before_wanted) ? before_wanted : tier_last_time_s,
                    };
                    ops->plans[qm->plan.used].initialized = false;
                    ops->plans[qm->plan.used].finalized = false;
                    qm->plan.array[qm->plan.used++] = t;

                    // prepare for the tier
                    selected_tier_last_time_s = t.before;

                    internal_fatal(!t.after || !t.before, "QUERY: invalid plan selected");

                    if (t.before >= before_wanted)
                        break;
                }
            }
        }
    }

    // sort the query plan
    if(qm->plan.used > 1)
        qsort(&qm->plan.array, qm->plan.used, sizeof(QUERY_PLAN_ENTRY), compare_query_plan_entries_on_start_time);

    if(!query_metric_is_valid_tier(qm, qm->plan.array[0].tier))
        return false;

#ifdef NETDATA_INTERNAL_CHECKS
    for(size_t p = 0; p < qm->plan.used ;p++) {
        internal_fatal(qm->plan.array[p].after > qm->plan.array[p].before, "QUERY: flipped after/before");
        internal_fatal(qm->plan.array[p].after < after_wanted, "QUERY: too small plan first time");
        internal_fatal(qm->plan.array[p].before > before_wanted, "QUERY: too big plan last time");
    }
#endif

    query_planer_initialize_plans(ops);
    query_planer_activate_plan(ops, 0, 0);

    return true;
}


// ----------------------------------------------------------------------------
// dimension level query engine

#define query_interpolate_point(this_point, last_point, now)      do {  \
    if(likely(                                                          \
            /* the point to interpolate is more than 1s wide */         \
            (this_point).sp.end_time_s - (this_point).sp.start_time_s > 1 \
                                                                        \
            /* the two points are exactly next to each other */         \
         && (last_point).sp.end_time_s == (this_point).sp.start_time_s  \
                                                                        \
            /* both points are valid numbers */                         \
         && netdata_double_isnumber((this_point).value)                 \
         && netdata_double_isnumber((last_point).value)                 \
                                                                        \
        )) {                                                            \
            (this_point).value = (last_point).value + ((this_point).value - (last_point).value) * (1.0 - (NETDATA_DOUBLE)((this_point).sp.end_time_s - (now)) / (NETDATA_DOUBLE)((this_point).sp.end_time_s - (this_point).sp.start_time_s)); \
            (this_point).sp.end_time_s = now;                           \
        }                                                               \
} while(0)

#define query_add_point_to_group(r, point, ops, add_flush)        do {  \
    if(likely(netdata_double_isnumber((point).value))) {                \
        if(likely(fpclassify((point).value) != FP_ZERO))                \
            (ops)->group_points_non_zero++;                             \
                                                                        \
        if(unlikely((point).sp.flags & SN_FLAG_RESET))                  \
            (ops)->group_value_flags |= RRDR_VALUE_RESET;               \
                                                                        \
        time_grouping_add(r, (point).value, add_flush);                 \
                                                                        \
        storage_point_merge_to((ops)->group_point, (point).sp);         \
        if(!(point).added)                                              \
            storage_point_merge_to((ops)->query_point, (point).sp);     \
    }                                                                   \
                                                                        \
    (ops)->group_points_added++;                                        \
} while(0)

static __thread QUERY_ENGINE_OPS *released_ops = NULL;

static void rrd2rrdr_query_ops_freeall(RRDR *r __maybe_unused) {
    while(released_ops) {
        QUERY_ENGINE_OPS *ops = released_ops;
        released_ops = ops->next;

        onewayalloc_freez(r->internal.owa, ops);
    }
}

static void rrd2rrdr_query_ops_release(QUERY_ENGINE_OPS *ops) {
    if(!ops) return;

    ops->next = released_ops;
    released_ops = ops;
}

static QUERY_ENGINE_OPS *rrd2rrdr_query_ops_get(RRDR *r) {
    QUERY_ENGINE_OPS *ops;
    if(released_ops) {
        ops = released_ops;
        released_ops = ops->next;
    }
    else {
        ops = onewayalloc_mallocz(r->internal.owa, sizeof(QUERY_ENGINE_OPS));
    }

    memset(ops, 0, sizeof(*ops));
    return ops;
}

static QUERY_ENGINE_OPS *rrd2rrdr_query_ops_prep(RRDR *r, size_t query_metric_id) {
    QUERY_TARGET *qt = r->internal.qt;

    QUERY_ENGINE_OPS *ops = rrd2rrdr_query_ops_get(r);
    *ops = (QUERY_ENGINE_OPS) {
            .r = r,
            .qm = query_metric(qt, query_metric_id),
            .tier_query_fetch = r->time_grouping.tier_query_fetch,
            .view_update_every = r->view.update_every,
            .query_granularity = (time_t)(r->view.update_every / r->view.group),
            .group_value_flags = RRDR_VALUE_NOTHING,
    };

    if(!query_plan(ops, qt->window.after, qt->window.before, qt->window.points)) {
        rrd2rrdr_query_ops_release(ops);
        return NULL;
    }

    return ops;
}

static void rrd2rrdr_query_execute(RRDR *r, size_t dim_id_in_rrdr, QUERY_ENGINE_OPS *ops) {
    QUERY_TARGET *qt = r->internal.qt;
    QUERY_METRIC *qm = ops->qm;

    const RRDR_TIME_GROUPING add_flush = r->time_grouping.add_flush;

    ops->group_point = STORAGE_POINT_UNSET;
    ops->query_point = STORAGE_POINT_UNSET;

    RRDR_OPTIONS options = qt->window.options;
    size_t points_wanted = qt->window.points;
    time_t after_wanted = qt->window.after;
    time_t before_wanted = qt->window.before; (void)before_wanted;

//    bool debug_this = false;
//    if(strcmp("user", string2str(rd->id)) == 0 && strcmp("system.cpu", string2str(rd->rrdset->id)) == 0)
//        debug_this = true;

    size_t points_added = 0;

    long rrdr_line = -1;
    bool use_anomaly_bit_as_value = (r->internal.qt->window.options & RRDR_OPTION_ANOMALY_BIT) ? true : false;

    NETDATA_DOUBLE min = r->view.min, max = r->view.max;

    QUERY_POINT last2_point = QUERY_POINT_EMPTY;
    QUERY_POINT last1_point = QUERY_POINT_EMPTY;
    QUERY_POINT new_point   = QUERY_POINT_EMPTY;

    // ONE POINT READ-AHEAD
    // when we switch plans, we read-ahead a point from the next plan
    // to join them smoothly at the exact time the next plan begins
    STORAGE_POINT next1_point = STORAGE_POINT_UNSET;

    time_t now_start_time = after_wanted - ops->query_granularity;
    time_t now_end_time   = after_wanted + ops->view_update_every - ops->query_granularity;

    size_t db_points_read_since_plan_switch = 0; (void)db_points_read_since_plan_switch;
    size_t query_is_finished_counter = 0;

    // The main loop, based on the query granularity we need
    for( ; points_added < points_wanted && query_is_finished_counter <= 10 ;
        now_start_time = now_end_time, now_end_time += ops->view_update_every) {

        if(unlikely(query_plan_should_switch_plan(ops, now_end_time))) {
            query_planer_next_plan(ops, now_end_time, new_point.sp.end_time_s);
            db_points_read_since_plan_switch = 0;
        }

        // read all the points of the db, prior to the time we need (now_end_time)

        size_t count_same_end_time = 0;
        while(count_same_end_time < 100) {
            if(likely(count_same_end_time == 0)) {
                last2_point = last1_point;
                last1_point = new_point;
            }

            if(unlikely(storage_engine_query_is_finished(ops->handle))) {
                query_is_finished_counter++;

                if(count_same_end_time != 0) {
                    last2_point = last1_point;
                    last1_point = new_point;
                }
                new_point = QUERY_POINT_EMPTY;
                new_point.sp.start_time_s = last1_point.sp.end_time_s;
                new_point.sp.end_time_s   = now_end_time;
//
//                if(debug_this) info("QUERY: is finished() returned true");
//
                break;
            }
            else
                query_is_finished_counter = 0;

            // fetch the new point
            {
                STORAGE_POINT sp;
                if(likely(storage_point_is_unset(next1_point))) {
                    db_points_read_since_plan_switch++;
                    sp = storage_engine_query_next_metric(ops->handle);
                    ops->db_points_read_per_tier[ops->tier]++;
                    ops->db_total_points_read++;

                    if(unlikely(options & RRDR_OPTION_ABSOLUTE))
                        storage_point_make_positive(sp);
                }
                else {
                    // ONE POINT READ-AHEAD
                    sp = next1_point;
                    storage_point_unset(next1_point);
                    db_points_read_since_plan_switch = 1;
                }

                // ONE POINT READ-AHEAD
                if(unlikely(query_plan_should_switch_plan(ops, sp.end_time_s) &&
                    query_planer_next_plan(ops, now_end_time, new_point.sp.end_time_s))) {

                    // The end time of the current point, crosses our plans (tiers)
                    // so, we switched plan (tier)
                    //
                    // There are 2 cases now:
                    //
                    // A. the entire point of the previous plan is to the future of point from the next plan
                    // B. part of the point of the previous plan overlaps with the point from the next plan

                    STORAGE_POINT sp2 = storage_engine_query_next_metric(ops->handle);
                    ops->db_points_read_per_tier[ops->tier]++;
                    ops->db_total_points_read++;

                    if(unlikely(options & RRDR_OPTION_ABSOLUTE))
                        storage_point_make_positive(sp);

                    if(sp.start_time_s > sp2.start_time_s)
                        // the point from the previous plan is useless
                        sp = sp2;
                    else
                        // let the query run from the previous plan
                        // but setting this will also cut off the interpolation
                        // of the point from the previous plan
                        next1_point = sp2;
                }

                new_point.sp = sp;
                new_point.added = false;
                query_point_set_id(new_point, ops->db_total_points_read);

//                if(debug_this)
//                    info("QUERY: got point %zu, from time %ld to %ld   //   now from %ld to %ld   //   query from %ld to %ld",
//                         new_point.id, new_point.start_time, new_point.end_time, now_start_time, now_end_time, after_wanted, before_wanted);
//
                // get the right value from the point we got
                if(likely(!storage_point_is_unset(sp) && !storage_point_is_gap(sp))) {

                    if(unlikely(use_anomaly_bit_as_value))
                        new_point.value = storage_point_anomaly_rate(new_point.sp);

                    else {
                        switch (ops->tier_query_fetch) {
                            default:
                            case TIER_QUERY_FETCH_AVERAGE:
                                new_point.value = sp.sum / (NETDATA_DOUBLE)sp.count;
                                break;

                            case TIER_QUERY_FETCH_MIN:
                                new_point.value = sp.min;
                                break;

                            case TIER_QUERY_FETCH_MAX:
                                new_point.value = sp.max;
                                break;

                            case TIER_QUERY_FETCH_SUM:
                                new_point.value = sp.sum;
                                break;
                        };
                    }
                }
                else
                    new_point.value      = NAN;
            }

            // check if the db is giving us zero duration points
            if(unlikely(db_points_read_since_plan_switch > 1 &&
                        new_point.sp.start_time_s == new_point.sp.end_time_s)) {

                internal_error(true, "QUERY: '%s', dimension '%s' next_metric() returned "
                                     "point %zu from %ld to %ld, that are both equal",
                               qt->id, query_metric_id(qt, qm),
                               new_point.id, new_point.sp.start_time_s, new_point.sp.end_time_s);

                new_point.sp.start_time_s = new_point.sp.end_time_s - ops->tier_ptr->db_update_every_s;
            }

            // check if the db is advancing the query
            if(unlikely(db_points_read_since_plan_switch > 1 &&
                        new_point.sp.end_time_s <= last1_point.sp.end_time_s)) {

                internal_error(true,
                               "QUERY: '%s', dimension '%s' next_metric() returned "
                               "point %zu from %ld to %ld, before the "
                               "last point %zu from %ld to %ld, "
                               "now is %ld to %ld",
                               qt->id, query_metric_id(qt, qm),
                               new_point.id, new_point.sp.start_time_s, new_point.sp.end_time_s,
                               last1_point.id, last1_point.sp.start_time_s, last1_point.sp.end_time_s,
                               now_start_time, now_end_time);

                count_same_end_time++;
                continue;
            }
            count_same_end_time = 0;

            // decide how to use this point
            if(likely(new_point.sp.end_time_s < now_end_time)) { // likely to favor tier0
                // this db point ends before our now_end_time

                if(likely(new_point.sp.end_time_s >= now_start_time)) { // likely to favor tier0
                    // this db point ends after our now_start time

                    query_add_point_to_group(r, new_point, ops, add_flush);
                    new_point.added = true;
                }
                else {
                    // we don't need this db point
                    // it is totally outside our current time-frame

                    // this is desirable for the first point of the query
                    // because it allows us to interpolate the next point
                    // at exactly the time we will want

                    // we only log if this is not point 1
                    internal_error(new_point.sp.end_time_s < ops->plan_expanded_after &&
                                   db_points_read_since_plan_switch > 1,
                                   "QUERY: '%s', dimension '%s' next_metric() "
                                   "returned point %zu from %ld time %ld, "
                                   "which is entirely before our current timeframe %ld to %ld "
                                   "(and before the entire query, after %ld, before %ld)",
                                   qt->id, query_metric_id(qt, qm),
                                   new_point.id, new_point.sp.start_time_s, new_point.sp.end_time_s,
                                   now_start_time, now_end_time,
                                   ops->plan_expanded_after, ops->plan_expanded_before);
                }

            }
            else {
                // the point ends in the future
                // so, we will interpolate it below, at the inner loop
                break;
            }
        }

        if(unlikely(count_same_end_time)) {
            internal_error(true,
                           "QUERY: '%s', dimension '%s', the database does not advance the query,"
                           " it returned an end time less or equal to the end time of the last "
                           "point we got %ld, %zu times",
                           qt->id, query_metric_id(qt, qm),
                           last1_point.sp.end_time_s, count_same_end_time);

            if(unlikely(new_point.sp.end_time_s <= last1_point.sp.end_time_s))
                new_point.sp.end_time_s = now_end_time;
        }

        time_t stop_time = new_point.sp.end_time_s;
        if(unlikely(!storage_point_is_unset(next1_point) && next1_point.start_time_s >= now_end_time)) {
            // ONE POINT READ-AHEAD
            // the point crosses the start time of the
            // read ahead storage point we have read
            stop_time = next1_point.start_time_s;
        }

        // the inner loop
        // we have 3 points in memory: last2, last1, new
        // we select the one to use based on their timestamps

        internal_fatal(now_end_time > stop_time || points_added >= points_wanted,
            "QUERY: first part of query provides invalid point to interpolate (now_end_time %ld, stop_time %ld",
            now_end_time, stop_time);

        do {
            // now_start_time is wrong in this loop
            // but, we don't need it

            QUERY_POINT current_point;

            if(likely(now_end_time > new_point.sp.start_time_s)) {
                // it is time for our NEW point to be used
                current_point = new_point;
                new_point.added = true; // first copy, then set it, so that new_point will not be added again
                query_interpolate_point(current_point, last1_point, now_end_time);

//                internal_error(current_point.id > 0
//                                && last1_point.id == 0
//                                && current_point.end_time > after_wanted
//                                && current_point.end_time > now_end_time,
//                               "QUERY: '%s', dimension '%s', after %ld, before %ld, view update every %ld,"
//                               " query granularity %ld, interpolating point %zu (from %ld to %ld) at %ld,"
//                               " but we could really favor by having last_point1 in this query.",
//                               qt->id, string2str(qm->dimension.id),
//                               after_wanted, before_wanted,
//                               ops.view_update_every, ops.query_granularity,
//                               current_point.id, current_point.start_time, current_point.end_time,
//                               now_end_time);
            }
            else if(likely(now_end_time <= last1_point.sp.end_time_s)) {
                // our LAST point is still valid
                current_point = last1_point;
                last1_point.added = true; // first copy, then set it, so that last1_point will not be added again
                query_interpolate_point(current_point, last2_point, now_end_time);

//                internal_error(current_point.id > 0
//                                && last2_point.id == 0
//                                && current_point.end_time > after_wanted
//                                && current_point.end_time > now_end_time,
//                               "QUERY: '%s', dimension '%s', after %ld, before %ld, view update every %ld,"
//                               " query granularity %ld, interpolating point %zu (from %ld to %ld) at %ld,"
//                               " but we could really favor by having last_point2 in this query.",
//                               qt->id, string2str(qm->dimension.id),
//                               after_wanted, before_wanted, ops.view_update_every, ops.query_granularity,
//                               current_point.id, current_point.start_time, current_point.end_time,
//                               now_end_time);
            }
            else {
                // a GAP, we don't have a value this time
                current_point = QUERY_POINT_EMPTY;
            }

            query_add_point_to_group(r, current_point, ops, add_flush);

            rrdr_line = rrdr_line_init(r, now_end_time, rrdr_line);
            size_t rrdr_o_v_index = rrdr_line * r->d + dim_id_in_rrdr;

            // find the place to store our values
            RRDR_VALUE_FLAGS *rrdr_value_options_ptr = &r->o[rrdr_o_v_index];

            // update the dimension options
            if(likely(ops->group_points_non_zero))
                r->od[dim_id_in_rrdr] |= RRDR_DIMENSION_NONZERO;

            // store the specific point options
            *rrdr_value_options_ptr = ops->group_value_flags;

            // store the group value
            NETDATA_DOUBLE group_value = time_grouping_flush(r, rrdr_value_options_ptr, add_flush);
            r->v[rrdr_o_v_index] = group_value;

            r->ar[rrdr_o_v_index] = storage_point_anomaly_rate(ops->group_point);

            if(likely(points_added || r->internal.queries_count)) {
                // find the min/max across all dimensions

                if(unlikely(group_value < min)) min = group_value;
                if(unlikely(group_value > max)) max = group_value;

            }
            else {
                // runs only when r->internal.queries_count == 0 && points_added == 0
                // so, on the first point added for the query.
                min = max = group_value;
            }

            points_added++;
            ops->group_points_added = 0;
            ops->group_value_flags = RRDR_VALUE_NOTHING;
            ops->group_points_non_zero = 0;
            ops->group_point = STORAGE_POINT_UNSET;

            now_end_time += ops->view_update_every;
        } while(now_end_time <= stop_time && points_added < points_wanted);

        // the loop above increased "now" by ops->view_update_every,
        // but the main loop will increase it too,
        // so, let's undo the last iteration of this loop
        now_end_time -= ops->view_update_every;
    }
    query_planer_finalize_remaining_plans(ops);

    qm->query_points = ops->query_point;

    // fill the rest of the points with empty values
    while (points_added < points_wanted) {
        rrdr_line++;
        size_t rrdr_o_v_index = rrdr_line * r->d + dim_id_in_rrdr;
        r->o[rrdr_o_v_index] = RRDR_VALUE_EMPTY;
        r->v[rrdr_o_v_index] = 0.0;
        r->ar[rrdr_o_v_index] = 0.0;
        points_added++;
    }

    r->internal.queries_count++;
    r->view.min = min;
    r->view.max = max;

    r->stats.result_points_generated += points_added;
    r->stats.db_points_read += ops->db_total_points_read;
    for(size_t tr = 0; tr < storage_tiers ; tr++)
        qt->db.tiers[tr].points += ops->db_points_read_per_tier[tr];
}

// ----------------------------------------------------------------------------
// fill the gap of a tier

void store_metric_at_tier(RRDDIM *rd, size_t tier, struct rrddim_tier *t, STORAGE_POINT sp, usec_t now_ut);

void rrdr_fill_tier_gap_from_smaller_tiers(RRDDIM *rd, size_t tier, time_t now_s) {
    if(unlikely(tier >= storage_tiers)) return;
    if(storage_tiers_backfill[tier] == RRD_BACKFILL_NONE) return;

    struct rrddim_tier *t = &rd->tiers[tier];
    if(unlikely(!t)) return;

    time_t latest_time_s = storage_engine_latest_time_s(t->backend, t->db_metric_handle);
    time_t granularity = (time_t)t->tier_grouping * (time_t)rd->update_every;
    time_t time_diff   = now_s - latest_time_s;

    // if the user wants only NEW backfilling, and we don't have any data
    if(storage_tiers_backfill[tier] == RRD_BACKFILL_NEW && latest_time_s <= 0) return;

    // there is really nothing we can do
    if(now_s <= latest_time_s || time_diff < granularity) return;

    struct storage_engine_query_handle handle;

    // for each lower tier
    for(int read_tier = (int)tier - 1; read_tier >= 0 ; read_tier--){
        time_t smaller_tier_first_time = storage_engine_oldest_time_s(rd->tiers[read_tier].backend, rd->tiers[read_tier].db_metric_handle);
        time_t smaller_tier_last_time = storage_engine_latest_time_s(rd->tiers[read_tier].backend, rd->tiers[read_tier].db_metric_handle);
        if(smaller_tier_last_time <= latest_time_s) continue;  // it is as bad as we are

        long after_wanted = (latest_time_s < smaller_tier_first_time) ? smaller_tier_first_time : latest_time_s;
        long before_wanted = smaller_tier_last_time;

        struct rrddim_tier *tmp = &rd->tiers[read_tier];
        storage_engine_query_init(tmp->backend, tmp->db_metric_handle, &handle, after_wanted, before_wanted, STORAGE_PRIORITY_HIGH);

        size_t points_read = 0;

        while(!storage_engine_query_is_finished(&handle)) {

            STORAGE_POINT sp = storage_engine_query_next_metric(&handle);
            points_read++;

            if(sp.end_time_s > latest_time_s) {
                latest_time_s = sp.end_time_s;
                store_metric_at_tier(rd, tier, t, sp, sp.end_time_s * USEC_PER_SEC);
            }
        }

        storage_engine_query_finalize(&handle);
        store_metric_collection_completed();
        global_statistics_backfill_query_completed(points_read);

        //internal_error(true, "DBENGINE: backfilled chart '%s', dimension '%s', tier %d, from %ld to %ld, with %zu points from tier %d",
        //               rd->rrdset->name, rd->name, tier, after_wanted, before_wanted, points, tr);
    }
}

// ----------------------------------------------------------------------------
// fill RRDR for the whole chart

#ifdef NETDATA_INTERNAL_CHECKS
static void rrd2rrdr_log_request_response_metadata(RRDR *r
        , RRDR_OPTIONS options __maybe_unused
        , RRDR_TIME_GROUPING group_method
        , bool aligned
        , size_t group
        , time_t resampling_time
        , size_t resampling_group
        , time_t after_wanted
        , time_t after_requested
        , time_t before_wanted
        , time_t before_requested
        , size_t points_requested
        , size_t points_wanted
        //, size_t after_slot
        //, size_t before_slot
        , const char *msg
        ) {

    QUERY_TARGET *qt = r->internal.qt;
    time_t first_entry_s = qt->db.first_time_s;
    time_t last_entry_s = qt->db.last_time_s;

    internal_error(
    true,
    "rrd2rrdr() on %s update every %ld with %s grouping %s (group: %zu, resampling_time: %ld, resampling_group: %zu), "
         "after (got: %ld, want: %ld, req: %ld, db: %ld), "
         "before (got: %ld, want: %ld, req: %ld, db: %ld), "
         "duration (got: %ld, want: %ld, req: %ld, db: %ld), "
         "points (got: %zu, want: %zu, req: %zu), "
         "%s"
         , qt->id
         , qt->window.query_granularity

         // grouping
         , (aligned) ? "aligned" : "unaligned"
         , time_grouping_method2string(group_method)
         , group
         , resampling_time
         , resampling_group

         // after
         , r->view.after
         , after_wanted
         , after_requested
         , first_entry_s

         // before
         , r->view.before
         , before_wanted
         , before_requested
         , last_entry_s

         // duration
         , (long)(r->view.before - r->view.after + qt->window.query_granularity)
         , (long)(before_wanted - after_wanted + qt->window.query_granularity)
         , (long)before_requested - after_requested
         , (long)((last_entry_s - first_entry_s) + qt->window.query_granularity)

         // points
         , r->rows
         , points_wanted
         , points_requested

         // message
         , msg
    );
}
#endif // NETDATA_INTERNAL_CHECKS

// Returns 1 if an absolute period was requested or 0 if it was a relative period
bool rrdr_relative_window_to_absolute(time_t *after, time_t *before, time_t *now_ptr) {
    time_t now = now_realtime_sec() - 1;

    if(now_ptr)
        *now_ptr = now;

    int absolute_period_requested = -1;
    long long after_requested, before_requested;

    before_requested = *before;
    after_requested = *after;

    // allow relative for before (smaller than API_RELATIVE_TIME_MAX)
    if(ABS(before_requested) <= API_RELATIVE_TIME_MAX) {
        // if the user asked for a positive relative time,
        // flip it to a negative
        if(before_requested > 0)
            before_requested = -before_requested;

        before_requested = now + before_requested;
        absolute_period_requested = 0;
    }

    // allow relative for after (smaller than API_RELATIVE_TIME_MAX)
    if(ABS(after_requested) <= API_RELATIVE_TIME_MAX) {
        if(after_requested > 0)
            after_requested = -after_requested;

        // if the user didn't give an after, use the number of points
        // to give a sane default
        if(after_requested == 0)
            after_requested = -600;

        // since the query engine now returns inclusive timestamps
        // it is awkward to return 6 points when after=-5 is given
        // so for relative queries we add 1 second, to give
        // more predictable results to users.
        after_requested = before_requested + after_requested + 1;
        absolute_period_requested = 0;
    }

    if(absolute_period_requested == -1)
        absolute_period_requested = 1;

    // check if the parameters are flipped
    if(after_requested > before_requested) {
        long long t = before_requested;
        before_requested = after_requested;
        after_requested = t;
    }

    // if the query requests future data
    // shift the query back to be in the present time
    // (this may also happen because of the rules above)
    if(before_requested > now) {
        long long delta = before_requested - now;
        before_requested -= delta;
        after_requested  -= delta;
    }

    time_t absolute_minimum_time = now - (10 * 365 * 86400);
    time_t absolute_maximum_time = now + (1 * 365 * 86400);

    if (after_requested < absolute_minimum_time && !unittest_running)
        after_requested = absolute_minimum_time;

    if (after_requested > absolute_maximum_time && !unittest_running)
        after_requested = absolute_maximum_time;

    if (before_requested < absolute_minimum_time && !unittest_running)
        before_requested = absolute_minimum_time;

    if (before_requested > absolute_maximum_time && !unittest_running)
        before_requested = absolute_maximum_time;

    *before = before_requested;
    *after = after_requested;

    return (absolute_period_requested != 1);
}

// #define DEBUG_QUERY_LOGIC 1

#ifdef DEBUG_QUERY_LOGIC
#define query_debug_log_init() BUFFER *debug_log = buffer_create(1000)
#define query_debug_log(args...) buffer_sprintf(debug_log, ##args)
#define query_debug_log_fin() { \
        info("QUERY: '%s', after:%ld, before:%ld, duration:%ld, points:%zu, res:%ld - wanted => after:%ld, before:%ld, points:%zu, group:%zu, granularity:%ld, resgroup:%ld, resdiv:" NETDATA_DOUBLE_FORMAT_AUTO " %s", qt->id, after_requested, before_requested, before_requested - after_requested, points_requested, resampling_time_requested, after_wanted, before_wanted, points_wanted, group, query_granularity, resampling_group, resampling_divisor, buffer_tostring(debug_log)); \
        buffer_free(debug_log); \
        debug_log = NULL; \
    }
#define query_debug_log_free() do { buffer_free(debug_log); } while(0)
#else
#define query_debug_log_init() debug_dummy()
#define query_debug_log(args...) debug_dummy()
#define query_debug_log_fin() debug_dummy()
#define query_debug_log_free() debug_dummy()
#endif

bool query_target_calculate_window(QUERY_TARGET *qt) {
    if (unlikely(!qt)) return false;

    size_t points_requested = (long)qt->request.points;
    time_t after_requested = qt->request.after;
    time_t before_requested = qt->request.before;
    RRDR_TIME_GROUPING group_method = qt->request.time_group_method;
    time_t resampling_time_requested = qt->request.resampling_time;
    RRDR_OPTIONS options = qt->window.options;
    size_t tier = qt->request.tier;
    time_t update_every = qt->db.minimum_latest_update_every_s ? qt->db.minimum_latest_update_every_s : 1;

    // RULES
    // points_requested = 0
    // the user wants all the natural points the database has
    //
    // after_requested = 0
    // the user wants to start the query from the oldest point in our database
    //
    // before_requested = 0
    // the user wants the query to end to the latest point in our database
    //
    // when natural points are wanted, the query has to be aligned to the update_every
    // of the database

    size_t points_wanted = points_requested;
    time_t after_wanted = after_requested;
    time_t before_wanted = before_requested;

    bool aligned = !(options & RRDR_OPTION_NOT_ALIGNED);
    bool automatic_natural_points = (points_wanted == 0);
    bool relative_period_requested = false;
    bool natural_points = (options & RRDR_OPTION_NATURAL_POINTS) || automatic_natural_points;
    bool before_is_aligned_to_db_end = false;

    query_debug_log_init();

    if (ABS(before_requested) <= API_RELATIVE_TIME_MAX || ABS(after_requested) <= API_RELATIVE_TIME_MAX) {
        relative_period_requested = true;
        natural_points = true;
        options |= RRDR_OPTION_NATURAL_POINTS;
        query_debug_log(":relative+natural");
    }

    // if the user wants virtual points, make sure we do it
    if (options & RRDR_OPTION_VIRTUAL_POINTS)
        natural_points = false;

    // set the right flag about natural and virtual points
    if (natural_points) {
        options |= RRDR_OPTION_NATURAL_POINTS;

        if (options & RRDR_OPTION_VIRTUAL_POINTS)
            options &= ~RRDR_OPTION_VIRTUAL_POINTS;
    }
    else {
        options |= RRDR_OPTION_VIRTUAL_POINTS;

        if (options & RRDR_OPTION_NATURAL_POINTS)
            options &= ~RRDR_OPTION_NATURAL_POINTS;
    }

    if (after_wanted == 0 || before_wanted == 0) {
        relative_period_requested = true;

        time_t first_entry_s = qt->db.first_time_s;
        time_t last_entry_s = qt->db.last_time_s;

        if (first_entry_s == 0 || last_entry_s == 0) {
            internal_error(true, "QUERY: no data detected on query '%s' (db first_entry_t = %ld, last_entry_t = %ld)", qt->id, first_entry_s, last_entry_s);
            after_wanted = qt->window.after;
            before_wanted = qt->window.before;

            if(after_wanted == before_wanted)
                after_wanted = before_wanted - update_every;

            if (points_wanted == 0) {
                points_wanted = (before_wanted - after_wanted) / update_every;
                query_debug_log(":zero points_wanted %zu", points_wanted);
            }
        }
        else {
            query_debug_log(":first_entry_t %ld, last_entry_t %ld", first_entry_s, last_entry_s);

            if (after_wanted == 0) {
                after_wanted = first_entry_s;
                query_debug_log(":zero after_wanted %ld", after_wanted);
            }

            if (before_wanted == 0) {
                before_wanted = last_entry_s;
                before_is_aligned_to_db_end = true;
                query_debug_log(":zero before_wanted %ld", before_wanted);
            }

            if (points_wanted == 0) {
                points_wanted = (last_entry_s - first_entry_s) / update_every;
                query_debug_log(":zero points_wanted %zu", points_wanted);
            }
        }
    }

    if (points_wanted == 0) {
        points_wanted = 600;
        query_debug_log(":zero600 points_wanted %zu", points_wanted);
    }

    // convert our before_wanted and after_wanted to absolute
    rrdr_relative_window_to_absolute(&after_wanted, &before_wanted, NULL);
    query_debug_log(":relative2absolute after %ld, before %ld", after_wanted, before_wanted);

    if (natural_points && (options & RRDR_OPTION_SELECTED_TIER) && tier > 0 && storage_tiers > 1) {
        update_every = rrdset_find_natural_update_every_for_timeframe(
                qt, after_wanted, before_wanted, points_wanted, options, tier);

        if (update_every <= 0) update_every = qt->db.minimum_latest_update_every_s;
        query_debug_log(":natural update every %ld", update_every);
    }

    // this is the update_every of the query
    // it may be different to the update_every of the database
    time_t query_granularity = (natural_points) ? update_every : 1;
    if (query_granularity <= 0) query_granularity = 1;
    query_debug_log(":query_granularity %ld", query_granularity);

    // align before_wanted and after_wanted to query_granularity
    if (before_wanted % query_granularity) {
        before_wanted -= before_wanted % query_granularity;
        query_debug_log(":granularity align before_wanted %ld", before_wanted);
    }

    if (after_wanted % query_granularity) {
        after_wanted -= after_wanted % query_granularity;
        query_debug_log(":granularity align after_wanted %ld", after_wanted);
    }

    // automatic_natural_points is set when the user wants all the points available in the database
    if (automatic_natural_points) {
        points_wanted = (before_wanted - after_wanted + 1) / query_granularity;
        if (unlikely(points_wanted <= 0)) points_wanted = 1;
        query_debug_log(":auto natural points_wanted %zu", points_wanted);
    }

    time_t duration = before_wanted - after_wanted;

    // if the resampling time is too big, extend the duration to the past
    if (unlikely(resampling_time_requested > duration)) {
        after_wanted = before_wanted - resampling_time_requested;
        duration = before_wanted - after_wanted;
        query_debug_log(":resampling after_wanted %ld", after_wanted);
    }

    // if the duration is not aligned to resampling time
    // extend the duration to the past, to avoid a gap at the chart
    // only when the missing duration is above 1/10th of a point
    if (resampling_time_requested > query_granularity && duration % resampling_time_requested) {
        time_t delta = duration % resampling_time_requested;
        if (delta > resampling_time_requested / 10) {
            after_wanted -= resampling_time_requested - delta;
            duration = before_wanted - after_wanted;
            query_debug_log(":resampling2 after_wanted %ld", after_wanted);
        }
    }

    // the available points of the query
    size_t points_available = (duration + 1) / query_granularity;
    if (unlikely(points_available <= 0)) points_available = 1;
    query_debug_log(":points_available %zu", points_available);

    if (points_wanted > points_available) {
        points_wanted = points_available;
        query_debug_log(":max points_wanted %zu", points_wanted);
    }

    if(points_wanted > 86400 && !unittest_running) {
        points_wanted = 86400;
        query_debug_log(":absolute max points_wanted %zu", points_wanted);
    }

    // calculate the desired grouping of source data points
    size_t group = points_available / points_wanted;
    if (group == 0) group = 1;

    // round "group" to the closest integer
    if (points_available % points_wanted > points_wanted / 2)
        group++;

    query_debug_log(":group %zu", group);

    if (points_wanted * group * query_granularity < (size_t)duration) {
        // the grouping we are going to do, is not enough
        // to cover the entire duration requested, so
        // we have to change the number of points, to make sure we will
        // respect the timeframe as closely as possibly

        // let's see how many points are the optimal
        points_wanted = points_available / group;

        if (points_wanted * group < points_available)
            points_wanted++;

        if (unlikely(points_wanted == 0))
            points_wanted = 1;

        query_debug_log(":optimal points %zu", points_wanted);
    }

    // resampling_time_requested enforces a certain grouping multiple
    NETDATA_DOUBLE resampling_divisor = 1.0;
    size_t resampling_group = 1;
    if (unlikely(resampling_time_requested > query_granularity)) {
        // the points we should group to satisfy gtime
        resampling_group = resampling_time_requested / query_granularity;
        if (unlikely(resampling_time_requested % query_granularity))
            resampling_group++;

        query_debug_log(":resampling group %zu", resampling_group);

        // adapt group according to resampling_group
        if (unlikely(group < resampling_group)) {
            group = resampling_group; // do not allow grouping below the desired one
            query_debug_log(":group less res %zu", group);
        }
        if (unlikely(group % resampling_group)) {
            group += resampling_group - (group % resampling_group); // make sure group is multiple of resampling_group
            query_debug_log(":group mod res %zu", group);
        }

        // resampling_divisor = group / resampling_group;
        resampling_divisor = (NETDATA_DOUBLE) (group * query_granularity) / (NETDATA_DOUBLE) resampling_time_requested;
        query_debug_log(":resampling divisor " NETDATA_DOUBLE_FORMAT, resampling_divisor);
    }

    // now that we have group, align the requested timeframe to fit it.
    if (aligned && before_wanted % (group * query_granularity)) {
        if (before_is_aligned_to_db_end)
            before_wanted -= before_wanted % (time_t)(group * query_granularity);
        else
            before_wanted += (time_t)(group * query_granularity) - before_wanted % (time_t)(group * query_granularity);
        query_debug_log(":align before_wanted %ld", before_wanted);
    }

    after_wanted = before_wanted - (time_t)(points_wanted * group * query_granularity) + query_granularity;
    query_debug_log(":final after_wanted %ld", after_wanted);

    duration = before_wanted - after_wanted;
    query_debug_log(":final duration %ld", duration + 1);

    query_debug_log_fin();

    internal_error(points_wanted != duration / (query_granularity * group) + 1,
                   "QUERY: points_wanted %zu is not points %zu",
                   points_wanted, (size_t)(duration / (query_granularity * group) + 1));

    internal_error(group < resampling_group,
                   "QUERY: group %zu is less than the desired group points %zu",
                   group, resampling_group);

    internal_error(group > resampling_group && group % resampling_group,
                   "QUERY: group %zu is not a multiple of the desired group points %zu",
                   group, resampling_group);

    // -------------------------------------------------------------------------
    // update QUERY_TARGET with our calculations

    qt->window.after = after_wanted;
    qt->window.before = before_wanted;
    qt->window.relative = relative_period_requested;
    qt->window.points = points_wanted;
    qt->window.group = group;
    qt->window.time_group_method = group_method;
    qt->window.time_group_options = qt->request.time_group_options;
    qt->window.query_granularity = query_granularity;
    qt->window.resampling_group = resampling_group;
    qt->window.resampling_divisor = resampling_divisor;
    qt->window.options = options;
    qt->window.tier = tier;
    qt->window.aligned = aligned;

    return true;
}

// ----------------------------------------------------------------------------
// group by

struct group_by_label_key {
    DICTIONARY *values;
};

static void group_by_label_key_insert_cb(const DICTIONARY_ITEM *item __maybe_unused, void *value, void *data) {
    // add the key to our r->label_keys global keys dictionary
    DICTIONARY *label_keys = data;
    dictionary_set(label_keys, dictionary_acquired_item_name(item), NULL, 0);

    // create a dictionary for the values of this key
    struct group_by_label_key *k = value;
    k->values = dictionary_create_advanced(DICT_OPTION_SINGLE_THREADED | DICT_OPTION_DONT_OVERWRITE_VALUE, NULL, 0);
}

static void group_by_label_key_delete_cb(const DICTIONARY_ITEM *item __maybe_unused, void *value, void *data __maybe_unused) {
    struct group_by_label_key *k = value;
    dictionary_destroy(k->values);
}

static int rrdlabels_traversal_cb_to_group_by_label_key(const char *name, const char *value, RRDLABEL_SRC ls __maybe_unused, void *data) {
    DICTIONARY *dl = data;
    struct group_by_label_key *k = dictionary_set(dl, name, NULL, sizeof(struct group_by_label_key));
    dictionary_set(k->values, value, NULL, 0);
    return 1;
}

void rrdr_json_group_by_labels(BUFFER *wb, const char *key, RRDR *r, RRDR_OPTIONS options) {
    if(!r->label_keys || !r->dl)
        return;

    buffer_json_member_add_object(wb, key);

    void *t;
    dfe_start_read(r->label_keys, t) {
                buffer_json_member_add_array(wb, t_dfe.name);

                for(size_t d = 0; d < r->d ;d++) {
                    if(!rrdr_dimension_should_be_exposed(r->od[d], options))
                        continue;

                    struct group_by_label_key *k = dictionary_get(r->dl[d], t_dfe.name);
                    if(k) {
                        buffer_json_add_array_item_array(wb);
                        void *tt;
                        dfe_start_read(k->values, tt) {
                                    buffer_json_add_array_item_string(wb, tt_dfe.name);
                                }
                        dfe_done(tt);
                        buffer_json_array_close(wb);
                    }
                    else
                        buffer_json_add_array_item_string(wb, NULL);
                }

                buffer_json_array_close(wb);
            }
    dfe_done(t);

    buffer_json_object_close(wb); // key
}

static int group_by_label_is_space(char c) {
    if(c == ',' || c == '|')
        return 1;

    return 0;
}

static void rrd2rrdr_set_timestamps(RRDR *r) {
    QUERY_TARGET *qt = r->internal.qt;

    internal_fatal(qt->window.points != r->n, "QUERY: mismatch to the number of points in qt and r");

    r->view.group = qt->window.group;
    r->view.update_every = (int) query_view_update_every(qt);
    r->view.before = qt->window.before;
    r->view.after = qt->window.after;

    r->time_grouping.points_wanted = qt->window.points;
    r->time_grouping.resampling_group = qt->window.resampling_group;
    r->time_grouping.resampling_divisor = qt->window.resampling_divisor;

    r->rows = qt->window.points;

    size_t points_wanted = qt->window.points;
    time_t after_wanted = qt->window.after;
    time_t before_wanted = qt->window.before; (void)before_wanted;

    time_t view_update_every = r->view.update_every;
    time_t query_granularity = (time_t)(r->view.update_every / r->view.group);

    size_t rrdr_line = 0;
    time_t first_point_end_time = after_wanted + view_update_every - query_granularity;
    time_t now_end_time = first_point_end_time;

    while (rrdr_line < points_wanted) {
        r->t[rrdr_line++] = now_end_time;
        now_end_time += view_update_every;
    }

    internal_fatal(r->t[0] != first_point_end_time, "QUERY: wrong first timestamp in the query");
    internal_error(r->t[points_wanted - 1] != before_wanted,
                   "QUERY: wrong last timestamp in the query, expected %ld, found %ld",
                   before_wanted, r->t[points_wanted - 1]);
}

static void query_group_by_make_dimension_key(BUFFER *key, RRDR_GROUP_BY group_by, size_t group_by_id, QUERY_TARGET *qt, QUERY_NODE *qn, QUERY_CONTEXT *qc, QUERY_INSTANCE *qi, QUERY_DIMENSION *qd __maybe_unused, QUERY_METRIC *qm, bool query_has_percentage_of_instance) {
    buffer_flush(key);
    if(unlikely(!query_has_percentage_of_instance && qm->status & RRDR_DIMENSION_HIDDEN)) {
        buffer_strcat(key, "__hidden_dimensions__");
    }
    else if(unlikely(group_by & RRDR_GROUP_BY_SELECTED)) {
        buffer_strcat(key, "selected");
    }
    else {
        if (group_by & RRDR_GROUP_BY_DIMENSION) {
            buffer_fast_strcat(key, "|", 1);
            buffer_strcat(key, query_metric_name(qt, qm));
        }

        if (group_by & (RRDR_GROUP_BY_INSTANCE|RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE)) {
            buffer_fast_strcat(key, "|", 1);
            buffer_strcat(key, string2str(query_instance_id_fqdn(qi, qt->request.version)));
        }

        if (group_by & RRDR_GROUP_BY_LABEL) {
            DICTIONARY *labels = rrdinstance_acquired_labels(qi->ria);
            for (size_t l = 0; l < qt->group_by[group_by_id].used; l++) {
                buffer_fast_strcat(key, "|", 1);
                rrdlabels_get_value_to_buffer_or_unset(labels, key, qt->group_by[group_by_id].label_keys[l], "[unset]");
            }
        }

        if (group_by & RRDR_GROUP_BY_NODE) {
            buffer_fast_strcat(key, "|", 1);
            buffer_strcat(key, qn->rrdhost->machine_guid);
        }

        if (group_by & RRDR_GROUP_BY_CONTEXT) {
            buffer_fast_strcat(key, "|", 1);
            buffer_strcat(key, rrdcontext_acquired_id(qc->rca));
        }

        if (group_by & RRDR_GROUP_BY_UNITS) {
            buffer_fast_strcat(key, "|", 1);
            buffer_strcat(key, query_target_has_percentage_units(qt) ? "%" : rrdinstance_acquired_units(qi->ria));
        }
    }
}

static void query_group_by_make_dimension_id(BUFFER *key, RRDR_GROUP_BY group_by, size_t group_by_id, QUERY_TARGET *qt, QUERY_NODE *qn, QUERY_CONTEXT *qc, QUERY_INSTANCE *qi, QUERY_DIMENSION *qd __maybe_unused, QUERY_METRIC *qm, bool query_has_percentage_of_instance) {
    buffer_flush(key);
    if(unlikely(!query_has_percentage_of_instance && qm->status & RRDR_DIMENSION_HIDDEN)) {
        buffer_strcat(key, "__hidden_dimensions__");
    }
    else if(unlikely(group_by & RRDR_GROUP_BY_SELECTED)) {
        buffer_strcat(key, "selected");
    }
    else {
        if (group_by & RRDR_GROUP_BY_DIMENSION) {
            buffer_strcat(key, query_metric_name(qt, qm));
        }

        if (group_by & (RRDR_GROUP_BY_INSTANCE|RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE)) {
            if (buffer_strlen(key) != 0)
                buffer_fast_strcat(key, ",", 1);

            if (group_by & RRDR_GROUP_BY_NODE)
                buffer_strcat(key, rrdinstance_acquired_id(qi->ria));
            else
                buffer_strcat(key, string2str(query_instance_id_fqdn(qi, qt->request.version)));
        }

        if (group_by & RRDR_GROUP_BY_LABEL) {
            DICTIONARY *labels = rrdinstance_acquired_labels(qi->ria);
            for (size_t l = 0; l < qt->group_by[group_by_id].used; l++) {
                if (buffer_strlen(key) != 0)
                    buffer_fast_strcat(key, ",", 1);
                rrdlabels_get_value_to_buffer_or_unset(labels, key, qt->group_by[group_by_id].label_keys[l], "[unset]");
            }
        }

        if (group_by & RRDR_GROUP_BY_NODE) {
            if (buffer_strlen(key) != 0)
                buffer_fast_strcat(key, ",", 1);

            buffer_strcat(key, qn->rrdhost->machine_guid);
        }

        if (group_by & RRDR_GROUP_BY_CONTEXT) {
            if (buffer_strlen(key) != 0)
                buffer_fast_strcat(key, ",", 1);

            buffer_strcat(key, rrdcontext_acquired_id(qc->rca));
        }

        if (group_by & RRDR_GROUP_BY_UNITS) {
            if (buffer_strlen(key) != 0)
                buffer_fast_strcat(key, ",", 1);

            buffer_strcat(key, query_target_has_percentage_units(qt) ? "%" : rrdinstance_acquired_units(qi->ria));
        }
    }
}

static void query_group_by_make_dimension_name(BUFFER *key, RRDR_GROUP_BY group_by, size_t group_by_id, QUERY_TARGET *qt, QUERY_NODE *qn, QUERY_CONTEXT *qc, QUERY_INSTANCE *qi, QUERY_DIMENSION *qd __maybe_unused, QUERY_METRIC *qm, bool query_has_percentage_of_instance) {
    buffer_flush(key);
    if(unlikely(!query_has_percentage_of_instance && qm->status & RRDR_DIMENSION_HIDDEN)) {
        buffer_strcat(key, "__hidden_dimensions__");
    }
    else if(unlikely(group_by & RRDR_GROUP_BY_SELECTED)) {
        buffer_strcat(key, "selected");
    }
    else {
        if (group_by & RRDR_GROUP_BY_DIMENSION) {
            buffer_strcat(key, query_metric_name(qt, qm));
        }

        if (group_by & (RRDR_GROUP_BY_INSTANCE|RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE)) {
            if (buffer_strlen(key) != 0)
                buffer_fast_strcat(key, ",", 1);

            if (group_by & RRDR_GROUP_BY_NODE)
                buffer_strcat(key, rrdinstance_acquired_name(qi->ria));
            else
                buffer_strcat(key, string2str(query_instance_name_fqdn(qi, qt->request.version)));
        }

        if (group_by & RRDR_GROUP_BY_LABEL) {
            DICTIONARY *labels = rrdinstance_acquired_labels(qi->ria);
            for (size_t l = 0; l < qt->group_by[group_by_id].used; l++) {
                if (buffer_strlen(key) != 0)
                    buffer_fast_strcat(key, ",", 1);
                rrdlabels_get_value_to_buffer_or_unset(labels, key, qt->group_by[group_by_id].label_keys[l], "[unset]");
            }
        }

        if (group_by & RRDR_GROUP_BY_NODE) {
            if (buffer_strlen(key) != 0)
                buffer_fast_strcat(key, ",", 1);

            buffer_strcat(key, rrdhost_hostname(qn->rrdhost));
        }

        if (group_by & RRDR_GROUP_BY_CONTEXT) {
            if (buffer_strlen(key) != 0)
                buffer_fast_strcat(key, ",", 1);

            buffer_strcat(key, rrdcontext_acquired_id(qc->rca));
        }

        if (group_by & RRDR_GROUP_BY_UNITS) {
            if (buffer_strlen(key) != 0)
                buffer_fast_strcat(key, ",", 1);

            buffer_strcat(key, query_target_has_percentage_units(qt) ? "%" : rrdinstance_acquired_units(qi->ria));
        }
    }
}

struct rrdr_group_by_entry {
    size_t priority;
    size_t count;
    STRING *id;
    STRING *name;
    STRING *units;
    RRDR_DIMENSION_FLAGS od;
    DICTIONARY *dl;
};

static RRDR *rrd2rrdr_group_by_initialize(ONEWAYALLOC *owa, QUERY_TARGET *qt) {
    RRDR *r_tmp = NULL;
    RRDR_OPTIONS options = qt->window.options;

    if(qt->request.version < 2) {
        // v1 query
        RRDR *r = rrdr_create(owa, qt, qt->query.used, qt->window.points);
         if(unlikely(!r)) {
             internal_error(true, "QUERY: cannot create RRDR for %s, after=%ld, before=%ld, dimensions=%u, points=%zu",
                            qt->id, qt->window.after, qt->window.before, qt->query.used, qt->window.points);
             return NULL;
         }
         r->group_by.r = NULL;

         for(size_t d = 0; d < qt->query.used ; d++) {
             QUERY_METRIC *qm = query_metric(qt, d);
             QUERY_DIMENSION *qd = query_dimension(qt, qm->link.query_dimension_id);
             r->di[d] = rrdmetric_acquired_id_dup(qd->rma);
             r->dn[d] = rrdmetric_acquired_name_dup(qd->rma);
         }

         rrd2rrdr_set_timestamps(r);
         return r;
    }
    // v2 query

    // parse all the group-by label keys
    for(size_t g = 0; g < MAX_QUERY_GROUP_BY_PASSES ;g++) {
        if (qt->request.group_by[g].group_by & RRDR_GROUP_BY_LABEL &&
            qt->request.group_by[g].group_by_label && *qt->request.group_by[g].group_by_label)
            qt->group_by[g].used = quoted_strings_splitter(
                    qt->request.group_by[g].group_by_label, qt->group_by[g].label_keys,
                    GROUP_BY_MAX_LABEL_KEYS, group_by_label_is_space);

        if (!qt->group_by[g].used)
            qt->request.group_by[g].group_by &= ~RRDR_GROUP_BY_LABEL;
    }

    // make sure there are valid group-by methods
    bool query_has_percentage_of_instance = false;
    for(size_t g = 0; g < MAX_QUERY_GROUP_BY_PASSES - 1 ;g++) {
        if(!(qt->request.group_by[g].group_by & SUPPORTED_GROUP_BY_METHODS))
            qt->request.group_by[g].group_by = (g == 0) ? RRDR_GROUP_BY_DIMENSION : RRDR_GROUP_BY_NONE;

        if(qt->request.group_by[g].group_by & RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE)
            query_has_percentage_of_instance = true;
    }

    // merge all group-by options to upper levels
    for(size_t g = 0; g < MAX_QUERY_GROUP_BY_PASSES - 1 ;g++) {
        if(qt->request.group_by[g].group_by == RRDR_GROUP_BY_NONE)
            continue;

        if(qt->request.group_by[g].group_by == RRDR_GROUP_BY_SELECTED) {
            for (size_t r = g + 1; r < MAX_QUERY_GROUP_BY_PASSES; r++)
                qt->request.group_by[r].group_by = RRDR_GROUP_BY_NONE;
        }
        else {
            for (size_t r = g + 1; r < MAX_QUERY_GROUP_BY_PASSES; r++) {
                if (qt->request.group_by[r].group_by == RRDR_GROUP_BY_NONE)
                    continue;

                if (qt->request.group_by[r].group_by != RRDR_GROUP_BY_SELECTED) {
                    if(qt->request.group_by[r].group_by & RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE)
                        qt->request.group_by[g].group_by |= RRDR_GROUP_BY_INSTANCE;
                    else
                        qt->request.group_by[g].group_by |= qt->request.group_by[r].group_by;

                    if(qt->request.group_by[r].group_by & RRDR_GROUP_BY_LABEL) {
                        for (size_t lr = 0; lr < qt->group_by[r].used; lr++) {
                            bool found = false;
                            for (size_t lg = 0; lg < qt->group_by[g].used; lg++) {
                                if (strcmp(qt->group_by[g].label_keys[lg], qt->group_by[r].label_keys[lr]) == 0) {
                                    found = true;
                                    break;
                                }
                            }

                            if (!found && qt->group_by[g].used < GROUP_BY_MAX_LABEL_KEYS * MAX_QUERY_GROUP_BY_PASSES)
                                qt->group_by[g].label_keys[qt->group_by[g].used++] = qt->group_by[r].label_keys[lr];
                        }
                    }
                }
            }
        }
    }

    int added = 0;
    RRDR *first_r = NULL, *last_r = NULL;
    BUFFER *key = buffer_create(0, NULL);
    struct rrdr_group_by_entry *entries = onewayalloc_mallocz(owa, qt->query.used * sizeof(struct rrdr_group_by_entry));
    DICTIONARY *groups = dictionary_create(DICT_OPTION_SINGLE_THREADED | DICT_OPTION_DONT_OVERWRITE_VALUE);
    DICTIONARY *label_keys = NULL;

    for(size_t g = 0; g < MAX_QUERY_GROUP_BY_PASSES ;g++) {
        RRDR_GROUP_BY group_by = qt->request.group_by[g].group_by;

        if(group_by == RRDR_GROUP_BY_NONE)
            break;

        memset(entries, 0, qt->query.used * sizeof(struct rrdr_group_by_entry));
        dictionary_flush(groups);
        added = 0;

        size_t hidden_dimensions = 0;
        bool final_grouping = (g == MAX_QUERY_GROUP_BY_PASSES - 1 || qt->request.group_by[g + 1].group_by == RRDR_GROUP_BY_NONE) ? true : false;

        if (final_grouping && (options & RRDR_OPTION_GROUP_BY_LABELS))
            label_keys = dictionary_create_advanced(DICT_OPTION_SINGLE_THREADED | DICT_OPTION_DONT_OVERWRITE_VALUE, NULL, 0);

        QUERY_INSTANCE *last_qi = NULL;
        size_t priority = 0;
        time_t update_every_max = 0;
        for (size_t d = 0; d < qt->query.used; d++) {
            QUERY_METRIC *qm = query_metric(qt, d);
            QUERY_DIMENSION *qd = query_dimension(qt, qm->link.query_dimension_id);
            QUERY_INSTANCE *qi = query_instance(qt, qm->link.query_instance_id);
            QUERY_CONTEXT *qc = query_context(qt, qm->link.query_context_id);
            QUERY_NODE *qn = query_node(qt, qm->link.query_node_id);

            if (qi != last_qi) {
                last_qi = qi;

                time_t update_every = rrdinstance_acquired_update_every(qi->ria);
                if (update_every > update_every_max)
                    update_every_max = update_every;
            }

            priority = qd->priority;

            if(qm->status & RRDR_DIMENSION_HIDDEN)
                hidden_dimensions++;

            // --------------------------------------------------------------------
            // generate the group by key

            query_group_by_make_dimension_key(key, group_by, g, qt, qn, qc, qi, qd, qm, query_has_percentage_of_instance);

            // lookup the key in the dictionary

            int pos = -1;
            int *set = dictionary_set(groups, buffer_tostring(key), &pos, sizeof(pos));
            if (*set == -1) {
                // the key just added to the dictionary

                *set = pos = added++;

                // ----------------------------------------------------------------
                // generate the dimension id

                query_group_by_make_dimension_id(key, group_by, g, qt, qn, qc, qi, qd, qm, query_has_percentage_of_instance);
                entries[pos].id = string_strdupz(buffer_tostring(key));

                // ----------------------------------------------------------------
                // generate the dimension name

                query_group_by_make_dimension_name(key, group_by, g, qt, qn, qc, qi, qd, qm, query_has_percentage_of_instance);
                entries[pos].name = string_strdupz(buffer_tostring(key));

                // add the rest of the info
                entries[pos].units = rrdinstance_acquired_units_dup(qi->ria);
                entries[pos].priority = priority;

                if (label_keys) {
                    entries[pos].dl = dictionary_create_advanced(
                            DICT_OPTION_SINGLE_THREADED | DICT_OPTION_FIXED_SIZE | DICT_OPTION_DONT_OVERWRITE_VALUE,
                            NULL, sizeof(struct group_by_label_key));
                    dictionary_register_insert_callback(entries[pos].dl, group_by_label_key_insert_cb, label_keys);
                    dictionary_register_delete_callback(entries[pos].dl, group_by_label_key_delete_cb, label_keys);
                }
            } else {
                // the key found in the dictionary
                pos = *set;
            }

            entries[pos].count++;

            if (unlikely(priority < entries[pos].priority))
                entries[pos].priority = priority;

            if(g > 0)
                last_r->dgbs[qm->grouped_as.slot] = pos;
            else
                qm->grouped_as.first_slot = pos;

            qm->grouped_as.slot = pos;
            qm->grouped_as.id = entries[pos].id;
            qm->grouped_as.name = entries[pos].name;
            qm->grouped_as.units = entries[pos].units;

            // copy the dimension flags decided by the query target
            // we need this, because if a dimension is explicitly selected
            // the query target adds to it the non-zero flag
            qm->status |= RRDR_DIMENSION_GROUPED;

            if(query_has_percentage_of_instance)
                // when the query has percentage of instance
                // there will be no hidden dimensions in the final query
                // so we have to remove the hidden flag from all dimensions
                entries[pos].od |= qm->status & ~RRDR_DIMENSION_HIDDEN;
            else
                entries[pos].od |= qm->status;

            if (entries[pos].dl)
                rrdlabels_walkthrough_read(rrdinstance_acquired_labels(qi->ria),
                                           rrdlabels_traversal_cb_to_group_by_label_key, entries[pos].dl);
        }

        RRDR *r = rrdr_create(owa, qt, added, qt->window.points);
        if (!r) {
            internal_error(true,
                           "QUERY: cannot create group by RRDR for %s, after=%ld, before=%ld, dimensions=%d, points=%zu",
                           qt->id, qt->window.after, qt->window.before, added, qt->window.points);
            goto cleanup;
        }

        bool hidden_dimension_on_percentage_of_instance = hidden_dimensions && (group_by & RRDR_GROUP_BY_PERCENTAGE_OF_INSTANCE);

        // prevent double cleanup in case of error
        added = 0;

        if(!last_r)
            first_r = last_r = r;
        else
            last_r->group_by.r = r;

        last_r = r;

        rrd2rrdr_set_timestamps(r);
        r->dp = onewayalloc_callocz(owa, r->d, sizeof(*r->dp));
        r->dview = onewayalloc_callocz(owa, r->d, sizeof(*r->dview));
        r->dgbc = onewayalloc_callocz(owa, r->d, sizeof(*r->dgbc));
        r->gbc = onewayalloc_callocz(owa, r->n * r->d, sizeof(*r->gbc));
        r->dqp = onewayalloc_callocz(owa, r->d, sizeof(STORAGE_POINT));

        if(hidden_dimension_on_percentage_of_instance)
            // this is where we are going to group the hidden dimensions
            r->vh = onewayalloc_mallocz(owa, r->n * r->d * sizeof(*r->vh));

        if(!final_grouping)
            // this is where we are going to store the slot in the next RRDR
            // that we are going to group by the dimension of this RRDR
            r->dgbs = onewayalloc_callocz(owa, r->d, sizeof(*r->dgbs));

        if (label_keys) {
            r->dl = onewayalloc_callocz(owa, r->d, sizeof(DICTIONARY *));
            r->label_keys = label_keys;
            label_keys = NULL;
        }

        // zero r (dimension options, names, and ids)
        // this is required, because group-by may lead to empty dimensions
        for (size_t d = 0; d < r->d; d++) {
            r->di[d] = entries[d].id;
            r->dn[d] = entries[d].name;

            r->od[d] = entries[d].od;
            r->du[d] = entries[d].units;
            r->dp[d] = entries[d].priority;
            r->dgbc[d] = entries[d].count;

            if (r->dl)
                r->dl[d] = entries[d].dl;
        }

        // initialize partial trimming
        r->partial_data_trimming.max_update_every = update_every_max;
        r->partial_data_trimming.expected_after =
                (!(qt->window.options & RRDR_OPTION_RETURN_RAW) &&
                 qt->window.before >= qt->window.now - update_every_max) ?
                qt->window.before - update_every_max :
                qt->window.before;
        r->partial_data_trimming.trimmed_after = qt->window.before;

        // make all values empty
        for (size_t i = 0; i != r->n; i++) {
            NETDATA_DOUBLE *cn = &r->v[i * r->d];
            RRDR_VALUE_FLAGS *co = &r->o[i * r->d];
            NETDATA_DOUBLE *ar = &r->ar[i * r->d];
            NETDATA_DOUBLE *vh = r->vh ? &r->vh[i * r->d] : NULL;

            for (size_t d = 0; d < r->d; d++) {
                cn[d] = NAN;
                ar[d] = 0.0;
                co[d] = RRDR_VALUE_EMPTY;

                if(vh)
                    *vh = NAN;
            }
        }
    }

    if(!first_r || !last_r)
        goto cleanup;

    r_tmp = rrdr_create(owa, qt, 1, qt->window.points);
    if (!r_tmp) {
        internal_error(true,
                       "QUERY: cannot create group by temporary RRDR for %s, after=%ld, before=%ld, dimensions=%d, points=%zu",
                       qt->id, qt->window.after, qt->window.before, 1, qt->window.points);
        goto cleanup;
    }
    rrd2rrdr_set_timestamps(r_tmp);
    r_tmp->group_by.r = first_r;

cleanup:
    if(!first_r || !last_r || !r_tmp) {
        if(r_tmp) {
            r_tmp->group_by.r = NULL;
            rrdr_free(owa, r_tmp);
        }

        if(first_r) {
            RRDR *r = first_r;
            while (r) {
                r_tmp = r->group_by.r;
                r->group_by.r = NULL;
                rrdr_free(owa, r);
                r = r_tmp;
            }
        }

        if(entries && added) {
            for (int d = 0; d < added; d++) {
                string_freez(entries[d].id);
                string_freez(entries[d].name);
                string_freez(entries[d].units);
                dictionary_destroy(entries[d].dl);
            }
        }
        dictionary_destroy(label_keys);

        first_r = last_r = r_tmp = NULL;
    }

    buffer_free(key);
    onewayalloc_freez(owa, entries);
    dictionary_destroy(groups);

    return r_tmp;
}

static void rrd2rrdr_group_by_add_metric(RRDR *r_dst, size_t d_dst, RRDR *r_tmp, size_t d_tmp,
                                         RRDR_GROUP_BY_FUNCTION group_by_aggregate_function,
                                         STORAGE_POINT *query_points, size_t pass __maybe_unused) {
    if(!r_tmp || r_dst == r_tmp || !(r_tmp->od[d_tmp] & RRDR_DIMENSION_QUERIED))
        return;

    internal_fatal(r_dst->n != r_tmp->n, "QUERY: group-by source and destination do not have the same number of rows");
    internal_fatal(d_dst >= r_dst->d, "QUERY: group-by destination dimension number exceeds destination RRDR size");
    internal_fatal(d_tmp >= r_tmp->d, "QUERY: group-by source dimension number exceeds source RRDR size");
    internal_fatal(!r_dst->dqp, "QUERY: group-by destination is not properly prepared (missing dqp array)");
    internal_fatal(!r_dst->gbc, "QUERY: group-by destination is not properly prepared (missing gbc array)");

    bool hidden_dimension_on_percentage_of_instance = (r_tmp->od[d_tmp] & RRDR_DIMENSION_HIDDEN) && r_dst->vh;

    if(!hidden_dimension_on_percentage_of_instance) {
        r_dst->od[d_dst] |= r_tmp->od[d_tmp];
        storage_point_merge_to(r_dst->dqp[d_dst], *query_points);
    }

    // do the group_by
    for(size_t i = 0; i != rrdr_rows(r_tmp) ; i++) {

        size_t idx_tmp = i * r_tmp->d + d_tmp;
        NETDATA_DOUBLE n_tmp = r_tmp->v[ idx_tmp ];
        RRDR_VALUE_FLAGS o_tmp = r_tmp->o[ idx_tmp ];
        NETDATA_DOUBLE ar_tmp = r_tmp->ar[ idx_tmp ];

        if(o_tmp & RRDR_VALUE_EMPTY)
            continue;

        size_t idx_dst = i * r_dst->d + d_dst;
        NETDATA_DOUBLE *cn = (hidden_dimension_on_percentage_of_instance) ? &r_dst->vh[ idx_dst ] : &r_dst->v[ idx_dst ];
        RRDR_VALUE_FLAGS *co = &r_dst->o[ idx_dst ];
        NETDATA_DOUBLE *ar = &r_dst->ar[ idx_dst ];
        uint32_t *gbc = &r_dst->gbc[ idx_dst ];

        switch(group_by_aggregate_function) {
            default:
            case RRDR_GROUP_BY_FUNCTION_AVERAGE:
            case RRDR_GROUP_BY_FUNCTION_SUM:
                if(isnan(*cn))
                    *cn = n_tmp;
                else
                    *cn += n_tmp;
                break;

            case RRDR_GROUP_BY_FUNCTION_MIN:
                if(isnan(*cn) || n_tmp < *cn)
                    *cn = n_tmp;
                break;

            case RRDR_GROUP_BY_FUNCTION_MAX:
                if(isnan(*cn) || n_tmp > *cn)
                    *cn = n_tmp;
                break;
        }

        if(!hidden_dimension_on_percentage_of_instance) {
            *co &= ~RRDR_VALUE_EMPTY;
            *co |= (o_tmp & (RRDR_VALUE_RESET | RRDR_VALUE_PARTIAL));
            *ar += ar_tmp;
            (*gbc)++;
        }
    }
}

static void rrdr2rrdr_group_by_partial_trimming(RRDR *r) {
    time_t trimmable_after = r->partial_data_trimming.expected_after;

    // find the point just before the trimmable ones
    ssize_t i = (ssize_t)r->n - 1;
    for( ; i >= 0 ;i--) {
        if (r->t[i] < trimmable_after)
            break;
    }

    if(unlikely(i < 0))
        return;

    size_t last_row_gbc = 0;
    for (; i < (ssize_t)r->n; i++) {
        size_t row_gbc = 0;
        for (size_t d = 0; d < r->d; d++) {
            if (unlikely(!(r->od[d] & RRDR_DIMENSION_QUERIED)))
                continue;

            row_gbc += r->gbc[ i * r->d + d ];
        }

        if (unlikely(r->t[i] >= trimmable_after && row_gbc < last_row_gbc)) {
            // discard the rest of the points
            r->partial_data_trimming.trimmed_after = r->t[i];
            r->rows = i;
            break;
        }
        else
            last_row_gbc = row_gbc;
    }
}

static void rrdr2rrdr_group_by_calculate_percentage_of_instance(RRDR *r) {
    if(!r->vh)
        return;

    for(size_t i = 0; i < r->n ;i++) {
        NETDATA_DOUBLE *cn = &r->v[ i * r->d ];
        NETDATA_DOUBLE *ch = &r->vh[ i * r->d ];

        for(size_t d = 0; d < r->d ;d++) {
            NETDATA_DOUBLE n = cn[d];
            NETDATA_DOUBLE h = ch[d];

            if(isnan(n))
                cn[d] = 0.0;

            else if(isnan(h))
                cn[d] = 100.0;

            else
                cn[d] = n * 100.0 / (n + h);
        }
    }
}

static void rrd2rrdr_convert_to_percentage(RRDR *r) {
    size_t global_min_max_values = 0;
    NETDATA_DOUBLE global_min = NAN, global_max = NAN;

    for(size_t i = 0; i != r->n ;i++) {
        NETDATA_DOUBLE *cn = &r->v[ i * r->d ];
        RRDR_VALUE_FLAGS *co = &r->o[ i * r->d ];

        NETDATA_DOUBLE total = 0;
        for (size_t d = 0; d < r->d; d++) {
            if (unlikely(!(r->od[d] & RRDR_DIMENSION_QUERIED)))
                continue;

            if(co[d] & RRDR_VALUE_EMPTY)
                continue;

            total += cn[d];
        }

        if(total == 0.0)
            total = 1.0;

        for (size_t d = 0; d < r->d; d++) {
            if (unlikely(!(r->od[d] & RRDR_DIMENSION_QUERIED)))
                continue;

            if(co[d] & RRDR_VALUE_EMPTY)
                continue;

            NETDATA_DOUBLE n = cn[d];
            n = cn[d] = n * 100.0 / total;

            if(unlikely(!global_min_max_values++))
                global_min = global_max = n;
            else {
                if(n < global_min)
                    global_min = n;
                if(n > global_max)
                    global_max = n;
            }
        }
    }

    r->view.min = global_min;
    r->view.max = global_max;

    if(!r->dview)
        // v1 query
        return;

    // v2 query

    for (size_t d = 0; d < r->d; d++) {
        if (unlikely(!(r->od[d] & RRDR_DIMENSION_QUERIED)))
            continue;

        size_t count = 0;
        NETDATA_DOUBLE min = 0.0, max = 0.0, sum = 0.0, ars = 0.0;
        for(size_t i = 0; i != r->rows ;i++) { // we use r->rows to respect trimming
            size_t idx = i * r->d + d;

            RRDR_VALUE_FLAGS o = r->o[ idx ];

            if (o & RRDR_VALUE_EMPTY)
                continue;

            NETDATA_DOUBLE ar = r->ar[ idx ];
            ars += ar;

            NETDATA_DOUBLE n = r->v[ idx ];
            sum += n;

            if(!count++)
                min = max = n;
            else {
                if(n < min)
                    min = n;
                if(n > max)
                    max = n;
            }
        }

        r->dview[d] = (STORAGE_POINT) {
            .sum = sum,
            .count = count,
            .min = min,
            .max = max,
            .anomaly_count = (size_t)(ars * (NETDATA_DOUBLE)count),
        };
    }
}

static RRDR *rrd2rrdr_group_by_finalize(RRDR *r_tmp) {
    QUERY_TARGET *qt = r_tmp->internal.qt;
    RRDR_OPTIONS options = qt->window.options;

    if(!r_tmp->group_by.r) {
        // v1 query
        if(options & RRDR_OPTION_PERCENTAGE)
            rrd2rrdr_convert_to_percentage(r_tmp);
        return r_tmp;
    }
    // v2 query

    // do the additional passes on RRDRs
    RRDR *last_r = r_tmp->group_by.r;
    rrdr2rrdr_group_by_calculate_percentage_of_instance(last_r);

    RRDR *r = last_r->group_by.r;
    size_t pass = 0;
    while(r) {
        pass++;
        for(size_t d = 0; d < last_r->d ;d++) {
            rrd2rrdr_group_by_add_metric(r, last_r->dgbs[d], last_r, d,
                                         qt->request.group_by[pass].aggregation,
                                         &last_r->dqp[d], pass);
        }
        rrdr2rrdr_group_by_calculate_percentage_of_instance(r);

        last_r = r;
        r = last_r->group_by.r;
    }

    // free all RRDRs except the last one
    r = r_tmp;
    while(r != last_r) {
        r_tmp = r->group_by.r;
        r->group_by.r = NULL;
        rrdr_free(r->internal.owa, r);
        r = r_tmp;
    }
    r = last_r;

    // find the final aggregation
    RRDR_GROUP_BY_FUNCTION aggregation = qt->request.group_by[0].aggregation;
    for(size_t g = 0; g < MAX_QUERY_GROUP_BY_PASSES ;g++)
        if(qt->request.group_by[g].group_by != RRDR_GROUP_BY_NONE)
            aggregation = qt->request.group_by[g].aggregation;

    if(!(options & RRDR_OPTION_RETURN_RAW) && r->partial_data_trimming.expected_after < qt->window.before)
        rrdr2rrdr_group_by_partial_trimming(r);

    // apply averaging, remove RRDR_VALUE_EMPTY, find the non-zero dimensions, min and max
    size_t global_min_max_values = 0;
    size_t dimensions_nonzero = 0;
    NETDATA_DOUBLE global_min = NAN, global_max = NAN;
    for (size_t d = 0; d < r->d; d++) {
        if (unlikely(!(r->od[d] & RRDR_DIMENSION_QUERIED)))
            continue;

        size_t points_nonzero = 0;
        NETDATA_DOUBLE min = 0, max = 0, sum = 0, ars = 0;
        size_t count = 0;

        for(size_t i = 0; i != r->n ;i++) {
            size_t idx = i * r->d + d;

            NETDATA_DOUBLE *cn = &r->v[ idx ];
            RRDR_VALUE_FLAGS *co = &r->o[ idx ];
            NETDATA_DOUBLE *ar = &r->ar[ idx ];
            uint32_t gbc = r->gbc[ idx ];

            if(likely(gbc)) {
                *co &= ~RRDR_VALUE_EMPTY;

                if(gbc != r->dgbc[d])
                    *co |= RRDR_VALUE_PARTIAL;

                NETDATA_DOUBLE n;

                sum += *cn;
                ars += *ar;

                if(aggregation == RRDR_GROUP_BY_FUNCTION_AVERAGE && !query_target_aggregatable(qt))
                    n = (*cn /= gbc);
                else
                    n = *cn;

                if(!query_target_aggregatable(qt))
                    *ar /= gbc;

                if(islessgreater(n, 0.0))
                    points_nonzero++;

                if(unlikely(!count))
                    min = max = n;
                else {
                    if(n < min)
                        min = n;

                    if(n > max)
                        max = n;
                }

                if(unlikely(!global_min_max_values++))
                    global_min = global_max = n;
                else {
                    if(n < global_min)
                        global_min = n;

                    if(n > global_max)
                        global_max = n;
                }

                count += gbc;
            }
        }

        if(points_nonzero) {
            r->od[d] |= RRDR_DIMENSION_NONZERO;
            dimensions_nonzero++;
        }

        r->dview[d] = (STORAGE_POINT) {
                .sum = sum,
                .count = count,
                .min = min,
                .max = max,
                .anomaly_count = (size_t)(ars * RRDR_DVIEW_ANOMALY_COUNT_MULTIPLIER / 100.0),
        };
    }

    r->view.min = global_min;
    r->view.max = global_max;

    if(!dimensions_nonzero && (qt->window.options & RRDR_OPTION_NONZERO)) {
        // all dimensions are zero
        // remove the nonzero option
        qt->window.options &= ~RRDR_OPTION_NONZERO;
    }

    if(options & RRDR_OPTION_PERCENTAGE && !(options & RRDR_OPTION_RETURN_RAW))
        rrd2rrdr_convert_to_percentage(r);

    // update query instance counts in query host and query context
    {
        size_t h = 0, c = 0, i = 0;
        for(; h < qt->nodes.used ; h++) {
            QUERY_NODE *qn = &qt->nodes.array[h];

            for(; c < qt->contexts.used ;c++) {
                QUERY_CONTEXT *qc = &qt->contexts.array[c];

                if(!rrdcontext_acquired_belongs_to_host(qc->rca, qn->rrdhost))
                    break;

                for(; i < qt->instances.used ;i++) {
                    QUERY_INSTANCE *qi = &qt->instances.array[i];

                    if(!rrdinstance_acquired_belongs_to_context(qi->ria, qc->rca))
                        break;

                    if(qi->metrics.queried) {
                        qc->instances.queried++;
                        qn->instances.queried++;
                    }
                    else if(qi->metrics.failed) {
                        qc->instances.failed++;
                        qn->instances.failed++;
                    }
                }
            }
        }
    }

    return r;
}

// ----------------------------------------------------------------------------
// query entry point

RRDR *rrd2rrdr_legacy(
        ONEWAYALLOC *owa,
        RRDSET *st, size_t points, time_t after, time_t before,
        RRDR_TIME_GROUPING group_method, time_t resampling_time, RRDR_OPTIONS options, const char *dimensions,
        const char *group_options, time_t timeout_ms, size_t tier, QUERY_SOURCE query_source,
        STORAGE_PRIORITY priority) {

    QUERY_TARGET_REQUEST qtr = {
            .version = 1,
            .st = st,
            .points = points,
            .after = after,
            .before = before,
            .time_group_method = group_method,
            .resampling_time = resampling_time,
            .options = options,
            .dimensions = dimensions,
            .time_group_options = group_options,
            .timeout_ms = timeout_ms,
            .tier = tier,
            .query_source = query_source,
            .priority = priority,
    };

    QUERY_TARGET *qt = query_target_create(&qtr);
    RRDR *r = rrd2rrdr(owa, qt);
    if(!r) {
        query_target_release(qt);
        return NULL;
    }

    r->internal.release_with_rrdr_qt = qt;
    return r;
}

RRDR *rrd2rrdr(ONEWAYALLOC *owa, QUERY_TARGET *qt) {
    if(!qt || !owa)
        return NULL;

    // qt.window members are the WANTED ones.
    // qt.request members are the REQUESTED ones.

    RRDR *r_tmp = rrd2rrdr_group_by_initialize(owa, qt);
    if(!r_tmp)
        return NULL;

    // the RRDR we group-by at
    RRDR *r = (r_tmp->group_by.r) ? r_tmp->group_by.r : r_tmp;

    // the final RRDR to return to callers
    RRDR *last_r = r_tmp;
    while(last_r->group_by.r)
        last_r = last_r->group_by.r;

    if(qt->window.relative)
        last_r->view.flags |= RRDR_RESULT_FLAG_RELATIVE;
    else
        last_r->view.flags |= RRDR_RESULT_FLAG_ABSOLUTE;

    // -------------------------------------------------------------------------
    // assign the processor functions
    rrdr_set_grouping_function(r_tmp, qt->window.time_group_method);

    // allocate any memory required by the grouping method
    r_tmp->time_grouping.create(r_tmp, qt->window.time_group_options);

    // -------------------------------------------------------------------------
    // do the work for each dimension

    time_t max_after = 0, min_before = 0;
    size_t max_rows = 0;

    long dimensions_used = 0, dimensions_nonzero = 0;
    size_t last_db_points_read = 0;
    size_t last_result_points_generated = 0;

    internal_fatal(released_ops, "QUERY: released_ops should be NULL when the query starts");

    QUERY_ENGINE_OPS **ops = NULL;
    if(qt->query.used)
        ops = onewayalloc_callocz(owa, qt->query.used, sizeof(QUERY_ENGINE_OPS *));

    size_t capacity = libuv_worker_threads * 10;
    size_t max_queries_to_prepare = (qt->query.used > (capacity - 1)) ? (capacity - 1) : qt->query.used;
    size_t queries_prepared = 0;
    while(queries_prepared < max_queries_to_prepare) {
        // preload another query
        ops[queries_prepared] = rrd2rrdr_query_ops_prep(r_tmp, queries_prepared);
        queries_prepared++;
    }

    QUERY_NODE *last_qn = NULL;
    usec_t last_ut = now_monotonic_usec();
    usec_t last_qn_ut = last_ut;

    for(size_t d = 0; d < qt->query.used ; d++) {
        QUERY_METRIC *qm = query_metric(qt, d);
        QUERY_DIMENSION *qd = query_dimension(qt, qm->link.query_dimension_id);
        QUERY_INSTANCE *qi = query_instance(qt, qm->link.query_instance_id);
        QUERY_CONTEXT *qc = query_context(qt, qm->link.query_context_id);
        QUERY_NODE *qn = query_node(qt, qm->link.query_node_id);

        usec_t now_ut = last_ut;
        if(qn != last_qn) {
            if(last_qn)
                last_qn->duration_ut = now_ut - last_qn_ut;

            last_qn = qn;
            last_qn_ut = now_ut;
        }

        if(queries_prepared < qt->query.used) {
            // preload another query
            ops[queries_prepared] = rrd2rrdr_query_ops_prep(r_tmp, queries_prepared);
            queries_prepared++;
        }

        size_t dim_in_rrdr_tmp = (r_tmp != r) ? 0 : d;

        // set the query target dimension options to rrdr
        r_tmp->od[dim_in_rrdr_tmp] = qm->status;

        // reset the grouping for the new dimension
        r_tmp->time_grouping.reset(r_tmp);

        if(ops[d]) {
            rrd2rrdr_query_execute(r_tmp, dim_in_rrdr_tmp, ops[d]);
            r_tmp->od[dim_in_rrdr_tmp] |= RRDR_DIMENSION_QUERIED;

            now_ut = now_monotonic_usec();
            qm->duration_ut = now_ut - last_ut;
            last_ut = now_ut;

            if(r_tmp != r) {
                // copy back whatever got updated from the temporary r

                // the query updates RRDR_DIMENSION_NONZERO
                qm->status = r_tmp->od[dim_in_rrdr_tmp];

                // the query updates these
                r->view.min = r_tmp->view.min;
                r->view.max = r_tmp->view.max;
                r->view.after = r_tmp->view.after;
                r->view.before = r_tmp->view.before;
                r->rows = r_tmp->rows;

                rrd2rrdr_group_by_add_metric(r, qm->grouped_as.first_slot, r_tmp, dim_in_rrdr_tmp,
                                             qt->request.group_by[0].aggregation, &qm->query_points, 0);
            }

            rrd2rrdr_query_ops_release(ops[d]); // reuse this ops allocation
            ops[d] = NULL;

            qi->metrics.queried++;
            qc->metrics.queried++;
            qn->metrics.queried++;

            qd->status |= QUERY_STATUS_QUERIED;
            qm->status |= RRDR_DIMENSION_QUERIED;

            if(qt->request.version >= 2) {
                // we need to make the query points positive now
                // since we will aggregate it across multiple dimensions
                storage_point_make_positive(qm->query_points);
                storage_point_merge_to(qi->query_points, qm->query_points);
                storage_point_merge_to(qc->query_points, qm->query_points);
                storage_point_merge_to(qn->query_points, qm->query_points);
                storage_point_merge_to(qt->query_points, qm->query_points);
            }
        }
        else {
            qi->metrics.failed++;
            qc->metrics.failed++;
            qn->metrics.failed++;

            qd->status |= QUERY_STATUS_FAILED;
            qm->status |= RRDR_DIMENSION_FAILED;

            continue;
        }

        global_statistics_rrdr_query_completed(
                1,
                r_tmp->stats.db_points_read - last_db_points_read,
                r_tmp->stats.result_points_generated - last_result_points_generated,
                qt->request.query_source);

        last_db_points_read = r_tmp->stats.db_points_read;
        last_result_points_generated = r_tmp->stats.result_points_generated;

        if(qm->status & RRDR_DIMENSION_NONZERO)
            dimensions_nonzero++;

        // verify all dimensions are aligned
        if(unlikely(!dimensions_used)) {
            min_before = r->view.before;
            max_after = r->view.after;
            max_rows = r->rows;
        }
        else {
            if(r->view.after != max_after) {
                internal_error(true, "QUERY: 'after' mismatch between dimensions for chart '%s': max is %zu, dimension '%s' has %zu",
                               rrdinstance_acquired_id(qi->ria), (size_t)max_after, rrdmetric_acquired_id(qd->rma), (size_t)r->view.after);

                r->view.after = (r->view.after > max_after) ? r->view.after : max_after;
            }

            if(r->view.before != min_before) {
                internal_error(true, "QUERY: 'before' mismatch between dimensions for chart '%s': max is %zu, dimension '%s' has %zu",
                               rrdinstance_acquired_id(qi->ria), (size_t)min_before, rrdmetric_acquired_id(qd->rma), (size_t)r->view.before);

                r->view.before = (r->view.before < min_before) ? r->view.before : min_before;
            }

            if(r->rows != max_rows) {
                internal_error(true, "QUERY: 'rows' mismatch between dimensions for chart '%s': max is %zu, dimension '%s' has %zu",
                               rrdinstance_acquired_id(qi->ria), (size_t)max_rows, rrdmetric_acquired_id(qd->rma), (size_t)r->rows);

                r->rows = (r->rows > max_rows) ? r->rows : max_rows;
            }
        }

        dimensions_used++;

        bool cancel = false;
        if (qt->request.interrupt_callback && qt->request.interrupt_callback(qt->request.interrupt_callback_data)) {
            cancel = true;
            log_access("QUERY INTERRUPTED");
        }

        if (qt->request.timeout_ms && ((NETDATA_DOUBLE)(now_ut - qt->timings.received_ut) / 1000.0) > (NETDATA_DOUBLE)qt->request.timeout_ms) {
            cancel = true;
            log_access("QUERY CANCELED RUNTIME EXCEEDED %0.2f ms (LIMIT %lld ms)",
                       (NETDATA_DOUBLE)(now_ut - qt->timings.received_ut) / 1000.0, (long long)qt->request.timeout_ms);
        }

        if(cancel) {
            r->view.flags |= RRDR_RESULT_FLAG_CANCEL;

            for(size_t i = d + 1; i < queries_prepared ; i++) {
                if(ops[i]) {
                    query_planer_finalize_remaining_plans(ops[i]);
                    rrd2rrdr_query_ops_release(ops[i]);
                    ops[i] = NULL;
                }
            }

            break;
        }
    }

    // free all resources used by the grouping method
    r_tmp->time_grouping.free(r_tmp);

    // get the final RRDR to send to the caller
    r = rrd2rrdr_group_by_finalize(r_tmp);

#ifdef NETDATA_INTERNAL_CHECKS
    if (dimensions_used && !(r->view.flags & RRDR_RESULT_FLAG_CANCEL)) {
        if(r->internal.log)
            rrd2rrdr_log_request_response_metadata(r, qt->window.options, qt->window.time_group_method, qt->window.aligned, qt->window.group, qt->request.resampling_time, qt->window.resampling_group,
                                                   qt->window.after, qt->request.after, qt->window.before, qt->request.before,
                                                   qt->request.points, qt->window.points, /*after_slot, before_slot,*/
                                                   r->internal.log);

        if(r->rows != qt->window.points)
            rrd2rrdr_log_request_response_metadata(r, qt->window.options, qt->window.time_group_method, qt->window.aligned, qt->window.group, qt->request.resampling_time, qt->window.resampling_group,
                                                   qt->window.after, qt->request.after, qt->window.before, qt->request.before,
                                                   qt->request.points, qt->window.points, /*after_slot, before_slot,*/
                                                   "got 'points' is not wanted 'points'");

        if(qt->window.aligned && (r->view.before % query_view_update_every(qt)) != 0)
            rrd2rrdr_log_request_response_metadata(r, qt->window.options, qt->window.time_group_method, qt->window.aligned, qt->window.group, qt->request.resampling_time, qt->window.resampling_group,
                                                   qt->window.after, qt->request.after, qt->window.before, qt->request.before,
                                                   qt->request.points, qt->window.points, /*after_slot, before_slot,*/
                                                   "'before' is not aligned but alignment is required");

        // 'after' should not be aligned, since we start inside the first group
        //if(qt->window.aligned && (r->after % group) != 0)
        //    rrd2rrdr_log_request_response_metadata(r, qt->window.options, qt->window.group_method, qt->window.aligned, qt->window.group, qt->request.resampling_time, qt->window.resampling_group, qt->window.after, after_requested, before_wanted, before_requested, points_requested, points_wanted, after_slot, before_slot, "'after' is not aligned but alignment is required");

        if(r->view.before != qt->window.before)
            rrd2rrdr_log_request_response_metadata(r, qt->window.options, qt->window.time_group_method, qt->window.aligned, qt->window.group, qt->request.resampling_time, qt->window.resampling_group,
                                                   qt->window.after, qt->request.after, qt->window.before, qt->request.before,
                                                   qt->request.points, qt->window.points, /*after_slot, before_slot,*/
                                                   "chart is not aligned to requested 'before'");

        if(r->view.before != qt->window.before)
            rrd2rrdr_log_request_response_metadata(r, qt->window.options, qt->window.time_group_method, qt->window.aligned, qt->window.group, qt->request.resampling_time, qt->window.resampling_group,
                                                   qt->window.after, qt->request.after, qt->window.before, qt->request.before,
                                                   qt->request.points, qt->window.points, /*after_slot, before_slot,*/
                                                   "got 'before' is not wanted 'before'");

        // reported 'after' varies, depending on group
        if(r->view.after != qt->window.after)
            rrd2rrdr_log_request_response_metadata(r, qt->window.options, qt->window.time_group_method, qt->window.aligned, qt->window.group, qt->request.resampling_time, qt->window.resampling_group,
                                                   qt->window.after, qt->request.after, qt->window.before, qt->request.before,
                                                   qt->request.points, qt->window.points, /*after_slot, before_slot,*/
                                                   "got 'after' is not wanted 'after'");

    }
#endif

    // free the query pipelining ops
    for(size_t d = 0; d < qt->query.used ; d++) {
        rrd2rrdr_query_ops_release(ops[d]);
        ops[d] = NULL;
    }
    rrd2rrdr_query_ops_freeall(r);
    internal_fatal(released_ops, "QUERY: released_ops should be NULL when the query ends");

    onewayalloc_freez(owa, ops);

    if(likely(dimensions_used && (qt->window.options & RRDR_OPTION_NONZERO) && !dimensions_nonzero))
        // when all the dimensions are zero, we should return all of them
        qt->window.options &= ~RRDR_OPTION_NONZERO;

    qt->timings.executed_ut = now_monotonic_usec();

    return r;
}