netdata/database/sqlite/sqlite_metadata.c

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

#include "sqlite_metadata.h"

// SQL statements

#define SQL_STORE_CLAIM_ID  "INSERT INTO node_instance " \
    "(host_id, claim_id, date_created) VALUES (@host_id, @claim_id, unixepoch()) " \
    "ON CONFLICT(host_id) DO UPDATE SET claim_id = excluded.claim_id;"

#define SQL_DELETE_HOST_LABELS  "DELETE FROM host_label WHERE host_id = @uuid;"

#define STORE_HOST_LABEL                                                                                               \
    "INSERT OR REPLACE INTO host_label (host_id, source_type, label_key, label_value, date_created) VALUES "

#define STORE_CHART_LABEL                                                                                              \
    "INSERT OR REPLACE INTO chart_label (chart_id, source_type, label_key, label_value, date_created) VALUES "

#define STORE_HOST_OR_CHART_LABEL_VALUE "(u2h('%s'), %d,'%s','%s', unixepoch())"

#define DELETE_DIMENSION_UUID   "DELETE FROM dimension WHERE dim_id = @uuid;"

#define SQL_STORE_HOST_INFO "INSERT OR REPLACE INTO host " \
        "(host_id, hostname, registry_hostname, update_every, os, timezone," \
        "tags, hops, memory_mode, abbrev_timezone, utc_offset, program_name, program_version," \
        "entries, health_enabled) " \
        "values (@host_id, @hostname, @registry_hostname, @update_every, @os, @timezone, @tags, @hops, @memory_mode, " \
        "@abbrev_timezone, @utc_offset, @program_name, @program_version, " \
        "@entries, @health_enabled);"

#define SQL_STORE_CHART "insert or replace into chart (chart_id, host_id, type, id, " \
    "name, family, context, title, unit, plugin, module, priority, update_every , chart_type , memory_mode , " \
    "history_entries) values (?1,?2,?3,?4,?5,?6,?7,?8,?9,?10,?11,?12,?13,?14,?15,?16);"

#define SQL_STORE_DIMENSION "INSERT OR REPLACE INTO dimension (dim_id, chart_id, id, name, multiplier, divisor , algorithm, options) " \
        "VALUES (@dim_id, @chart_id, @id, @name, @multiplier, @divisor, @algorithm, @options);"

#define SELECT_DIMENSION_LIST "SELECT dim_id, rowid FROM dimension WHERE rowid > @row_id"

#define STORE_HOST_INFO "INSERT OR REPLACE INTO host_info (host_id, system_key, system_value, date_created) VALUES "
#define STORE_HOST_INFO_VALUES "(u2h('%s'), '%s','%s', unixepoch())"

#define MIGRATE_LOCALHOST_TO_NEW_MACHINE_GUID                                                                          \
    "UPDATE chart SET host_id = @host_id WHERE host_id in (SELECT host_id FROM host where host_id <> @host_id and hops = 0);"
#define DELETE_NON_EXISTING_LOCALHOST "DELETE FROM host WHERE hops = 0 AND host_id <> @host_id;"
#define DELETE_MISSING_NODE_INSTANCES "DELETE FROM node_instance WHERE host_id NOT IN (SELECT host_id FROM host);"

#define METADATA_CMD_Q_MAX_SIZE (1024)              // Max queue size; callers will block until there is room
#define METADATA_MAINTENANCE_FIRST_CHECK (1800)     // Maintenance first run after agent startup in seconds
#define METADATA_MAINTENANCE_RETRY (60)             // Retry run if already running or last run did actual work
#define METADATA_MAINTENANCE_INTERVAL (3600)        // Repeat maintenance after latest successful

#define METADATA_HOST_CHECK_FIRST_CHECK (5)         // First check for pending metadata
#define METADATA_HOST_CHECK_INTERVAL (30)           // Repeat check for pending metadata
#define METADATA_HOST_CHECK_IMMEDIATE (5)           // Repeat immediate run because we have more metadata to write

#define MAX_METADATA_CLEANUP (500)                  // Maximum metadata write operations (e.g  deletes before retrying)
#define METADATA_MAX_BATCH_SIZE (512)               // Maximum commands to execute before running the event loop

enum metadata_opcode {
    METADATA_DATABASE_NOOP = 0,
    METADATA_DATABASE_TIMER,
    METADATA_DEL_DIMENSION,
    METADATA_STORE_CLAIM_ID,
    METADATA_ADD_HOST_INFO,
    METADATA_SCAN_HOSTS,
    METADATA_LOAD_HOST_CONTEXT,
    METADATA_MAINTENANCE,
    METADATA_SYNC_SHUTDOWN,
    METADATA_UNITTEST,
    METADATA_ML_LOAD_MODELS,
    // leave this last
    // we need it to check for worker utilization
    METADATA_MAX_ENUMERATIONS_DEFINED
};

#define MAX_PARAM_LIST  (2)
struct metadata_cmd {
    enum metadata_opcode opcode;
    struct completion *completion;
    const void *param[MAX_PARAM_LIST];
};

struct metadata_database_cmdqueue {
    unsigned head, tail;
    struct metadata_cmd cmd_array[METADATA_CMD_Q_MAX_SIZE];
};

typedef enum {
    METADATA_FLAG_CLEANUP           = (1 << 0), // Cleanup is running
    METADATA_FLAG_SCANNING_HOSTS    = (1 << 1), // Scanning of hosts in worker thread
    METADATA_FLAG_SHUTDOWN          = (1 << 2), // Shutting down
} METADATA_FLAG;

#define METADATA_WORKER_BUSY    (METADATA_FLAG_CLEANUP | METADATA_FLAG_SCANNING_HOSTS)

struct metadata_wc {
    uv_thread_t thread;
    uv_loop_t *loop;
    uv_async_t async;
    uv_timer_t timer_req;
    time_t check_metadata_after;
    time_t check_hosts_after;
    volatile unsigned queue_size;
    METADATA_FLAG flags;
    uint64_t row_id;
    struct completion init_complete;
    /* FIFO command queue */
    uv_mutex_t cmd_mutex;
    uv_cond_t cmd_cond;
    struct metadata_database_cmdqueue cmd_queue;
};

#define metadata_flag_check(target_flags, flag) (__atomic_load_n(&((target_flags)->flags), __ATOMIC_SEQ_CST) & (flag))
#define metadata_flag_set(target_flags, flag)   __atomic_or_fetch(&((target_flags)->flags), (flag), __ATOMIC_SEQ_CST)
#define metadata_flag_clear(target_flags, flag) __atomic_and_fetch(&((target_flags)->flags), ~(flag), __ATOMIC_SEQ_CST)

//
// For unittest
//
struct thread_unittest {
    int join;
    unsigned added;
    unsigned processed;
    unsigned *done;
};


// Metadata functions

struct query_build {
    BUFFER *sql;
    int count;
    char uuid_str[UUID_STR_LEN];
};

static int host_label_store_to_sql_callback(const char *name, const char *value, RRDLABEL_SRC ls, void *data) {
    struct query_build *lb = data;
    if (unlikely(!lb->count))
        buffer_sprintf(lb->sql, STORE_HOST_LABEL);
    else
        buffer_strcat(lb->sql, ", ");
    buffer_sprintf(lb->sql, STORE_HOST_OR_CHART_LABEL_VALUE, lb->uuid_str, (int)ls & ~(RRDLABEL_FLAG_INTERNAL), name, value);
    lb->count++;
    return 1;
}

static int chart_label_store_to_sql_callback(const char *name, const char *value, RRDLABEL_SRC ls, void *data) {
    struct query_build *lb = data;
    if (unlikely(!lb->count))
        buffer_sprintf(lb->sql, STORE_CHART_LABEL);
    else
        buffer_strcat(lb->sql, ", ");
    buffer_sprintf(lb->sql, STORE_HOST_OR_CHART_LABEL_VALUE, lb->uuid_str, ls, name, value);
    lb->count++;
    return 1;
}

#define SQL_DELETE_CHART_LABEL "DELETE FROM chart_label WHERE chart_id = @chart_id;"
#define SQL_DELETE_CHART_LABEL_HISTORY "DELETE FROM chart_label WHERE date_created < %ld AND chart_id = @chart_id;"

static void clean_old_chart_labels(RRDSET *st)
{
    char sql[512];
    time_t first_time_s = rrdset_first_entry_s(st);

    if (unlikely(!first_time_s))
        snprintfz(sql, 511,SQL_DELETE_CHART_LABEL);
    else
        snprintfz(sql, 511,SQL_DELETE_CHART_LABEL_HISTORY, first_time_s);

    int rc = exec_statement_with_uuid(sql, &st->chart_uuid);
    if (unlikely(rc))
        error_report("METADATA: 'host:%s' Failed to clean old labels for chart %s", rrdhost_hostname(st->rrdhost), rrdset_name(st));
}

static int check_and_update_chart_labels(RRDSET *st, BUFFER *work_buffer, size_t *query_counter)
{
    size_t old_version = st->rrdlabels_last_saved_version;
    size_t new_version = dictionary_version(st->rrdlabels);

    if (new_version == old_version)
        return 0;

    struct query_build tmp = {.sql = work_buffer, .count = 0};
    uuid_unparse_lower(st->chart_uuid, tmp.uuid_str);
    rrdlabels_walkthrough_read(st->rrdlabels, chart_label_store_to_sql_callback, &tmp);
    int rc = db_execute(db_meta, buffer_tostring(work_buffer));
    if (likely(!rc)) {
        st->rrdlabels_last_saved_version = new_version;
        (*query_counter)++;
    }

    clean_old_chart_labels(st);
    return rc;
}

// Migrate all hosts with hops zero to this host_uuid
void migrate_localhost(uuid_t *host_uuid)
{
    int rc;

    rc = exec_statement_with_uuid(MIGRATE_LOCALHOST_TO_NEW_MACHINE_GUID, host_uuid);
    if (!rc)
        rc = exec_statement_with_uuid(DELETE_NON_EXISTING_LOCALHOST, host_uuid);
    if (!rc) {
        if (unlikely(db_execute(db_meta, DELETE_MISSING_NODE_INSTANCES)))
            error_report("Failed to remove deleted hosts from node instances");
    }
}

static int store_claim_id(uuid_t *host_id, uuid_t *claim_id)
{
    sqlite3_stmt *res = NULL;
    int rc;

    if (unlikely(!db_meta)) {
        if (default_rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE)
            error_report("Database has not been initialized");
        return 1;
    }

    rc = sqlite3_prepare_v2(db_meta, SQL_STORE_CLAIM_ID, -1, &res, 0);
    if (unlikely(rc != SQLITE_OK)) {
        error_report("Failed to prepare statement to store host claim id");
        return 1;
    }

    rc = sqlite3_bind_blob(res, 1, host_id, sizeof(*host_id), SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK)) {
        error_report("Failed to bind host_id parameter to store claim id");
        goto failed;
    }

    if (claim_id)
        rc = sqlite3_bind_blob(res, 2, claim_id, sizeof(*claim_id), SQLITE_STATIC);
    else
        rc = sqlite3_bind_null(res, 2);
    if (unlikely(rc != SQLITE_OK)) {
        error_report("Failed to bind claim_id parameter to host claim id");
        goto failed;
    }

    rc = execute_insert(res);
    if (unlikely(rc != SQLITE_DONE))
        error_report("Failed to store host claim id rc = %d", rc);

failed:
    if (unlikely(sqlite3_finalize(res) != SQLITE_OK))
        error_report("Failed to finalize the prepared statement when storing a host claim id");

    return rc != SQLITE_DONE;
}

static void delete_dimension_uuid(uuid_t *dimension_uuid)
{
    static __thread sqlite3_stmt *res = NULL;
    int rc;

    if (unlikely(!res)) {
        rc = prepare_statement(db_meta, DELETE_DIMENSION_UUID, &res);
        if (rc != SQLITE_OK) {
            error_report("Failed to prepare statement to delete a dimension uuid");
            return;
        }
    }

    rc = sqlite3_bind_blob(res, 1, dimension_uuid,  sizeof(*dimension_uuid), SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto skip_execution;

    rc = sqlite3_step_monitored(res);
    if (unlikely(rc != SQLITE_DONE))
        error_report("Failed to delete dimension uuid, rc = %d", rc);

skip_execution:
    rc = sqlite3_reset(res);
    if (unlikely(rc != SQLITE_OK))
        error_report("Failed to reset statement when deleting dimension UUID, rc = %d", rc);
}

//
// Store host and host system info information in the database
static int store_host_metadata(RRDHOST *host)
{
    static __thread sqlite3_stmt *res = NULL;
    int rc, param = 0;

    if (unlikely(!db_meta)) {
        if (default_rrd_memory_mode != RRD_MEMORY_MODE_DBENGINE)
            return 0;
        error_report("Database has not been initialized");
        return 1;
    }

    if (unlikely((!res))) {
        rc = prepare_statement(db_meta, SQL_STORE_HOST_INFO, &res);
        if (unlikely(rc != SQLITE_OK)) {
            error_report("Failed to prepare statement to store host, rc = %d", rc);
            return 1;
        }
    }

    rc = sqlite3_bind_blob(res, ++param, &host->host_uuid, sizeof(host->host_uuid), SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = bind_text_null(res, ++param, rrdhost_hostname(host), 0);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = bind_text_null(res, ++param, rrdhost_registry_hostname(host), 1);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, host->rrd_update_every);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = bind_text_null(res, ++param, rrdhost_os(host), 1);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = bind_text_null(res, ++param, rrdhost_timezone(host), 1);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = bind_text_null(res, ++param, rrdhost_tags(host), 1);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, host->system_info ? host->system_info->hops : 0);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, host->rrd_memory_mode);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = bind_text_null(res, ++param, rrdhost_abbrev_timezone(host), 1);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, host->utc_offset);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = bind_text_null(res, ++param, rrdhost_program_name(host), 1);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = bind_text_null(res, ++param, rrdhost_program_version(host), 1);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int64(res, ++param, host->rrd_history_entries);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, (int ) host->health.health_enabled);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    int store_rc = sqlite3_step_monitored(res);
    if (unlikely(store_rc != SQLITE_DONE))
        error_report("Failed to store host %s, rc = %d", rrdhost_hostname(host), rc);

    rc = sqlite3_reset(res);
    if (unlikely(rc != SQLITE_OK))
        error_report("Failed to reset statement to store host %s, rc = %d", rrdhost_hostname(host), rc);

    return store_rc != SQLITE_DONE;
bind_fail:
    error_report("Failed to bind %d parameter to store host %s, rc = %d", param, rrdhost_hostname(host), rc);
    rc = sqlite3_reset(res);
    if (unlikely(rc != SQLITE_OK))
        error_report("Failed to reset statement to store host %s, rc = %d", rrdhost_hostname(host), rc);
    return 1;
}

static void add_host_sysinfo_key_value(const char *name, const char *value, void *data)
{
    struct query_build *lb = data;

    if (unlikely(!value))
        return;

    if (unlikely(!lb->count))
        buffer_sprintf(
            lb->sql, STORE_HOST_INFO);
    else
        buffer_strcat(lb->sql, ", ");
    buffer_sprintf(lb->sql, STORE_HOST_INFO_VALUES, lb->uuid_str, name, value);
    lb->count++;
}

static bool build_host_system_info_statements(RRDHOST *host, BUFFER *work_buffer)
{
    struct rrdhost_system_info *system_info = host->system_info;

    if (unlikely(!system_info))
        return false;

    buffer_flush(work_buffer);
    struct query_build key_data = {.sql = work_buffer, .count = 0};
    uuid_unparse_lower(host->host_uuid, key_data.uuid_str);

    add_host_sysinfo_key_value("NETDATA_CONTAINER_OS_NAME", system_info->container_os_name, &key_data);
    add_host_sysinfo_key_value("NETDATA_CONTAINER_OS_ID", system_info->container_os_id, &key_data);
    add_host_sysinfo_key_value("NETDATA_CONTAINER_OS_ID_LIKE", system_info->container_os_id_like, &key_data);
    add_host_sysinfo_key_value("NETDATA_CONTAINER_OS_VERSION", system_info->container_os_version, &key_data);
    add_host_sysinfo_key_value("NETDATA_CONTAINER_OS_VERSION_ID", system_info->container_os_version_id, &key_data);
    add_host_sysinfo_key_value("NETDATA_CONTAINER_OS_DETECTION", system_info->host_os_detection, &key_data);
    add_host_sysinfo_key_value("NETDATA_HOST_OS_NAME", system_info->host_os_name, &key_data);
    add_host_sysinfo_key_value("NETDATA_HOST_OS_ID", system_info->host_os_id, &key_data);
    add_host_sysinfo_key_value("NETDATA_HOST_OS_ID_LIKE", system_info->host_os_id_like, &key_data);
    add_host_sysinfo_key_value("NETDATA_HOST_OS_VERSION", system_info->host_os_version, &key_data);
    add_host_sysinfo_key_value("NETDATA_HOST_OS_VERSION_ID", system_info->host_os_version_id, &key_data);
    add_host_sysinfo_key_value("NETDATA_HOST_OS_DETECTION", system_info->host_os_detection, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_KERNEL_NAME", system_info->kernel_name, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_CPU_LOGICAL_CPU_COUNT", system_info->host_cores, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_CPU_FREQ", system_info->host_cpu_freq, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_TOTAL_RAM", system_info->host_ram_total, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_TOTAL_DISK_SIZE", system_info->host_disk_space, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_KERNEL_VERSION", system_info->kernel_version, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_ARCHITECTURE", system_info->architecture, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_VIRTUALIZATION", system_info->virtualization, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_VIRT_DETECTION", system_info->virt_detection, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_CONTAINER", system_info->container, &key_data);
    add_host_sysinfo_key_value("NETDATA_SYSTEM_CONTAINER_DETECTION", system_info->container_detection, &key_data);
    add_host_sysinfo_key_value("NETDATA_HOST_IS_K8S_NODE", system_info->is_k8s_node, &key_data);

    return true;
}


/*
 * Store a chart in the database
 */

static int store_chart_metadata(RRDSET *st)
{
    static __thread sqlite3_stmt *res = NULL;
    int rc, param = 0, store_rc = 0;

    if (unlikely(!db_meta)) {
        if (default_rrd_memory_mode != RRD_MEMORY_MODE_DBENGINE)
            return 0;
        error_report("Database has not been initialized");
        return 1;
    }

    if (unlikely(!res)) {
        rc = prepare_statement(db_meta, SQL_STORE_CHART, &res);
        if (unlikely(rc != SQLITE_OK)) {
            error_report("Failed to prepare statement to store chart, rc = %d", rc);
            return 1;
        }
    }

    rc = sqlite3_bind_blob(res, ++param, &st->chart_uuid, sizeof(st->chart_uuid), SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_blob(res, ++param, &st->rrdhost->host_uuid, sizeof(st->rrdhost->host_uuid), SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, string2str(st->parts.type), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, string2str(st->parts.id), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    const char *name = string2str(st->parts.name);
    if (name && *name)
        rc = sqlite3_bind_text(res, ++param, name, -1, SQLITE_STATIC);
    else
        rc = sqlite3_bind_null(res, ++param);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, rrdset_family(st), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, rrdset_context(st), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, rrdset_title(st), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, rrdset_units(st), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, rrdset_plugin_name(st), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, rrdset_module_name(st), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, (int) st->priority);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, st->update_every);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, st->chart_type);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, st->rrd_memory_mode);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, (int) st->entries);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    store_rc = execute_insert(res);
    if (unlikely(store_rc != SQLITE_DONE))
        error_report("Failed to store chart, rc = %d", store_rc);

    rc = sqlite3_reset(res);
    if (unlikely(rc != SQLITE_OK))
        error_report("Failed to reset statement in chart store function, rc = %d", rc);

    return store_rc != SQLITE_DONE;

bind_fail:
    error_report("Failed to bind parameter %d to store chart, rc = %d", param, rc);
    rc = sqlite3_reset(res);
    if (unlikely(rc != SQLITE_OK))
        error_report("Failed to reset statement in chart store function, rc = %d", rc);
    return 1;
}

/*
 * Store a dimension
 */
static int store_dimension_metadata(RRDDIM *rd)
{
    static __thread sqlite3_stmt *res = NULL;
    int rc, param = 0;

    if (unlikely(!db_meta)) {
        if (default_rrd_memory_mode != RRD_MEMORY_MODE_DBENGINE)
            return 0;
        error_report("Database has not been initialized");
        return 1;
    }

    if (unlikely(!res)) {
        rc = prepare_statement(db_meta, SQL_STORE_DIMENSION, &res);
        if (unlikely(rc != SQLITE_OK)) {
            error_report("Failed to prepare statement to store dimension, rc = %d", rc);
            return 1;
        }
    }

    rc = sqlite3_bind_blob(res, ++param, &rd->metric_uuid, sizeof(rd->metric_uuid), SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_blob(res, ++param, &rd->rrdset->chart_uuid, sizeof(rd->rrdset->chart_uuid), SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, string2str(rd->id), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_text(res, ++param, string2str(rd->name), -1, SQLITE_STATIC);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, (int) rd->multiplier);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, (int ) rd->divisor);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = sqlite3_bind_int(res, ++param, rd->algorithm);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    if (rrddim_option_check(rd, RRDDIM_OPTION_HIDDEN))
        rc = sqlite3_bind_text(res, ++param, "hidden", -1, SQLITE_STATIC);
    else
        rc = sqlite3_bind_null(res, ++param);
    if (unlikely(rc != SQLITE_OK))
        goto bind_fail;

    rc = execute_insert(res);
    if (unlikely(rc != SQLITE_DONE))
        error_report("Failed to store dimension, rc = %d", rc);

    rc = sqlite3_reset(res);
    if (unlikely(rc != SQLITE_OK))
        error_report("Failed to reset statement in store dimension, rc = %d", rc);
    return 0;

bind_fail:
    error_report("Failed to bind parameter %d to store dimension, rc = %d", param, rc);
    rc = sqlite3_reset(res);
    if (unlikely(rc != SQLITE_OK))
        error_report("Failed to reset statement in store dimension, rc = %d", rc);
    return 1;
}

static bool dimension_can_be_deleted(uuid_t *dim_uuid __maybe_unused)
{
#ifdef ENABLE_DBENGINE
    if(dbengine_enabled) {
        bool no_retention = true;
        for (size_t tier = 0; tier < storage_tiers; tier++) {
            if (!multidb_ctx[tier])
                continue;
            time_t first_time_t = 0, last_time_t = 0;
            if (rrdeng_metric_retention_by_uuid((void *) multidb_ctx[tier], dim_uuid, &first_time_t, &last_time_t)) {
                if (first_time_t > 0) {
                    no_retention = false;
                    break;
                }
            }
        }
        return no_retention;
    }
    else
        return false;
#else
    return false;
#endif
}

static void check_dimension_metadata(struct metadata_wc *wc)
{
    int rc;
    sqlite3_stmt *res = NULL;

    rc = sqlite3_prepare_v2(db_meta, SELECT_DIMENSION_LIST, -1, &res, 0);
    if (unlikely(rc != SQLITE_OK)) {
        error_report("Failed to prepare statement to fetch host dimensions");
        return;
    }

    rc = sqlite3_bind_int64(res, 1,  (sqlite3_int64) wc->row_id);
    if (unlikely(rc != SQLITE_OK)) {
        error_report("Failed to row parameter");
        goto skip_run;
    }

    uint32_t total_checked = 0;
    uint32_t total_deleted= 0;
    uint64_t last_row_id = wc->row_id;

    info("METADATA: Checking dimensions starting after row %"PRIu64, wc->row_id);

    while (sqlite3_step_monitored(res) == SQLITE_ROW && total_deleted < MAX_METADATA_CLEANUP) {
        if (unlikely(metadata_flag_check(wc, METADATA_FLAG_SHUTDOWN)))
            break;

        last_row_id = sqlite3_column_int64(res, 1);
        rc = dimension_can_be_deleted((uuid_t *)sqlite3_column_blob(res, 0));
        if (rc == true) {
            delete_dimension_uuid((uuid_t *)sqlite3_column_blob(res, 0));
            total_deleted++;
        }
        total_checked++;
    }
    wc->row_id = last_row_id;
    time_t now = now_realtime_sec();
    if (total_deleted > 0) {
        wc->check_metadata_after = now + METADATA_MAINTENANCE_RETRY;
    } else
        wc->row_id = 0;
    info("METADATA: Checked %u, deleted %u -- will resume after row %"PRIu64" in %lld seconds", total_checked, total_deleted, wc->row_id,
         (long long)(wc->check_metadata_after - now));

skip_run:
    rc = sqlite3_finalize(res);
    if (unlikely(rc != SQLITE_OK))
        error_report("Failed to finalize the prepared statement when reading dimensions");
}

static void cleanup_health_log(void)
{
    RRDHOST *host;
    dfe_start_reentrant(rrdhost_root_index, host) {
        if (rrdhost_flag_check(host, RRDHOST_FLAG_ARCHIVED))
            continue;
        sql_health_alarm_log_cleanup(host);
    }
    dfe_done(host);
}

//
// EVENT LOOP STARTS HERE
//

static void metadata_init_cmd_queue(struct metadata_wc *wc)
{
    wc->cmd_queue.head = wc->cmd_queue.tail = 0;
    wc->queue_size = 0;
    fatal_assert(0 == uv_cond_init(&wc->cmd_cond));
    fatal_assert(0 == uv_mutex_init(&wc->cmd_mutex));
}

int metadata_enq_cmd_noblock(struct metadata_wc *wc, struct metadata_cmd *cmd)
{
    unsigned queue_size;

    /* wait for free space in queue */
    uv_mutex_lock(&wc->cmd_mutex);

    if (cmd->opcode == METADATA_SYNC_SHUTDOWN) {
        metadata_flag_set(wc, METADATA_FLAG_SHUTDOWN);
        uv_mutex_unlock(&wc->cmd_mutex);
        return 0;
    }

    if (unlikely((queue_size = wc->queue_size) == METADATA_CMD_Q_MAX_SIZE ||
        metadata_flag_check(wc, METADATA_FLAG_SHUTDOWN))) {
        uv_mutex_unlock(&wc->cmd_mutex);
        return 1;
    }

    fatal_assert(queue_size < METADATA_CMD_Q_MAX_SIZE);
    /* enqueue command */
    wc->cmd_queue.cmd_array[wc->cmd_queue.tail] = *cmd;
    wc->cmd_queue.tail = wc->cmd_queue.tail != METADATA_CMD_Q_MAX_SIZE - 1 ?
                             wc->cmd_queue.tail + 1 : 0;
    wc->queue_size = queue_size + 1;
    uv_mutex_unlock(&wc->cmd_mutex);
    return 0;
}

static void metadata_enq_cmd(struct metadata_wc *wc, struct metadata_cmd *cmd)
{
    unsigned queue_size;

    /* wait for free space in queue */
    uv_mutex_lock(&wc->cmd_mutex);
    if (unlikely(metadata_flag_check(wc, METADATA_FLAG_SHUTDOWN))) {
        uv_mutex_unlock(&wc->cmd_mutex);
        (void) uv_async_send(&wc->async);
        return;
    }

    if (cmd->opcode == METADATA_SYNC_SHUTDOWN) {
        metadata_flag_set(wc, METADATA_FLAG_SHUTDOWN);
        uv_mutex_unlock(&wc->cmd_mutex);
        (void) uv_async_send(&wc->async);
        return;
    }

    while ((queue_size = wc->queue_size) == METADATA_CMD_Q_MAX_SIZE) {
        if (unlikely(metadata_flag_check(wc, METADATA_FLAG_SHUTDOWN))) {
            uv_mutex_unlock(&wc->cmd_mutex);
            return;
        }
        uv_cond_wait(&wc->cmd_cond, &wc->cmd_mutex);
    }
    fatal_assert(queue_size < METADATA_CMD_Q_MAX_SIZE);
    /* enqueue command */
    wc->cmd_queue.cmd_array[wc->cmd_queue.tail] = *cmd;
    wc->cmd_queue.tail = wc->cmd_queue.tail != METADATA_CMD_Q_MAX_SIZE - 1 ?
                             wc->cmd_queue.tail + 1 : 0;
    wc->queue_size = queue_size + 1;
    uv_mutex_unlock(&wc->cmd_mutex);

    /* wake up event loop */
    (void) uv_async_send(&wc->async);
}

static struct metadata_cmd metadata_deq_cmd(struct metadata_wc *wc)
{
    struct metadata_cmd ret;
    unsigned queue_size;

    uv_mutex_lock(&wc->cmd_mutex);
    queue_size = wc->queue_size;
    if (queue_size == 0) {
        memset(&ret, 0, sizeof(ret));
        ret.opcode = METADATA_DATABASE_NOOP;
        ret.completion = NULL;
    } else {
        /* dequeue command */
        ret = wc->cmd_queue.cmd_array[wc->cmd_queue.head];

        if (queue_size == 1) {
            wc->cmd_queue.head = wc->cmd_queue.tail = 0;
        } else {
            wc->cmd_queue.head = wc->cmd_queue.head != METADATA_CMD_Q_MAX_SIZE - 1 ?
                                     wc->cmd_queue.head + 1 : 0;
        }
        wc->queue_size = queue_size - 1;
        /* wake up producers */
        uv_cond_signal(&wc->cmd_cond);
    }
    uv_mutex_unlock(&wc->cmd_mutex);

    return ret;
}

static void async_cb(uv_async_t *handle)
{
    uv_stop(handle->loop);
    uv_update_time(handle->loop);
}

#define TIMER_INITIAL_PERIOD_MS (1000)
#define TIMER_REPEAT_PERIOD_MS (1000)

static void timer_cb(uv_timer_t* handle)
{
    uv_stop(handle->loop);
    uv_update_time(handle->loop);

   struct metadata_wc *wc = handle->data;
   struct metadata_cmd cmd;
   memset(&cmd, 0, sizeof(cmd));

   time_t now = now_realtime_sec();

   if (wc->check_metadata_after && wc->check_metadata_after < now) {
       cmd.opcode = METADATA_MAINTENANCE;
       if (!metadata_enq_cmd_noblock(wc, &cmd))
           wc->check_metadata_after = now + METADATA_MAINTENANCE_INTERVAL;
   }

   if (wc->check_hosts_after && wc->check_hosts_after < now) {
       cmd.opcode = METADATA_SCAN_HOSTS;
       if (!metadata_enq_cmd_noblock(wc, &cmd))
           wc->check_hosts_after = now + METADATA_HOST_CHECK_INTERVAL;
   }
}

static void after_metadata_cleanup(uv_work_t *req, int status)
{
    UNUSED(status);

    struct metadata_wc *wc = req->data;
    metadata_flag_clear(wc, METADATA_FLAG_CLEANUP);
}

static void start_metadata_cleanup(uv_work_t *req)
{
    register_libuv_worker_jobs();

    worker_is_busy(UV_EVENT_METADATA_CLEANUP);
    struct metadata_wc *wc = req->data;
    check_dimension_metadata(wc);
    cleanup_health_log();
    (void) sqlite3_wal_checkpoint(db_meta, NULL);
    worker_is_idle();
}

struct scan_metadata_payload {
    uv_work_t request;
    struct metadata_wc *wc;
    struct completion *completion;
    BUFFER *work_buffer;
    uint32_t max_count;
};

struct host_context_load_thread {
    uv_thread_t thread;
    RRDHOST *host;
    bool busy;
    bool finished;
};

static void restore_host_context(void *arg)
{
    struct host_context_load_thread *hclt = arg;
    RRDHOST *host = hclt->host;

    usec_t started_ut = now_monotonic_usec(); (void)started_ut;
    rrdhost_load_rrdcontext_data(host);
    usec_t ended_ut = now_monotonic_usec(); (void)ended_ut;

    rrdhost_flag_clear(host, RRDHOST_FLAG_PENDING_CONTEXT_LOAD | RRDHOST_FLAG_CONTEXT_LOAD_IN_PROGRESS);

#ifdef ENABLE_ACLK
    aclk_queue_node_info(host, false);
#endif

    internal_error(true, "METADATA: 'host:%s' context load in %0.2f ms", rrdhost_hostname(host),
        (double)(ended_ut - started_ut) / USEC_PER_MS);

    __atomic_store_n(&hclt->finished, true, __ATOMIC_RELEASE);
}

// Callback after scan of hosts is done
static void after_start_host_load_context(uv_work_t *req, int status __maybe_unused)
{
    struct scan_metadata_payload *data = req->data;
    freez(data);
}

#define MAX_FIND_THREAD_RETRIES (10)

static void cleanup_finished_threads(struct host_context_load_thread *hclt, size_t max_thread_slots, bool wait)
{
    for (size_t index = 0; index < max_thread_slots; index++) {
       if (__atomic_load_n(&(hclt[index].finished), __ATOMIC_RELAXED)
           || (wait && __atomic_load_n(&(hclt[index].busy), __ATOMIC_ACQUIRE))) {
           int rc = uv_thread_join(&(hclt[index].thread));
           if (rc)
                error("Failed to join thread, rc = %d",rc);
           __atomic_store_n(&(hclt[index].busy), false, __ATOMIC_RELEASE);
           __atomic_store_n(&(hclt[index].finished), false, __ATOMIC_RELEASE);
       }
    }
}

static size_t find_available_thread_slot(struct host_context_load_thread *hclt, size_t max_thread_slots, size_t *found_index)
{
    size_t retries = MAX_FIND_THREAD_RETRIES;
    while (retries--) {
       size_t index = 0;
       while (index < max_thread_slots) {
           if (false == __atomic_load_n(&(hclt[index].busy), __ATOMIC_ACQUIRE)) {
                *found_index = index;
                return true;
           }
           index++;
       }
       sleep_usec(10 * USEC_PER_MS);
    }
    return false;
}

static void start_all_host_load_context(uv_work_t *req __maybe_unused)
{
    register_libuv_worker_jobs();

    struct scan_metadata_payload *data = req->data;
    UNUSED(data);

    worker_is_busy(UV_EVENT_HOST_CONTEXT_LOAD);
    usec_t started_ut = now_monotonic_usec(); (void)started_ut;

    RRDHOST *host;

    size_t max_threads = MIN(get_netdata_cpus() / 2, 6);
    struct host_context_load_thread *hclt = callocz(max_threads, sizeof(*hclt));

    size_t thread_index;
    dfe_start_reentrant(rrdhost_root_index, host) {
       if (rrdhost_flag_check(host, RRDHOST_FLAG_CONTEXT_LOAD_IN_PROGRESS) ||
           !rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_CONTEXT_LOAD))
           continue;

       rrdhost_flag_set(host, RRDHOST_FLAG_CONTEXT_LOAD_IN_PROGRESS);
       internal_error(true, "METADATA: 'host:%s' loading context", rrdhost_hostname(host));

       cleanup_finished_threads(hclt, max_threads, false);
       bool found_slot = find_available_thread_slot(hclt, max_threads, &thread_index);

       if (unlikely(!found_slot)) {
           struct host_context_load_thread hclt_sync = {.host = host};
           restore_host_context(&hclt_sync);
       }
       else {
           __atomic_store_n(&hclt[thread_index].busy, true, __ATOMIC_RELAXED);
           hclt[thread_index].host = host;
           assert(0 == uv_thread_create(&hclt[thread_index].thread, restore_host_context, &hclt[thread_index]));
       }
    }
    dfe_done(host);

    cleanup_finished_threads(hclt, max_threads, true);
    freez(hclt);
    usec_t ended_ut = now_monotonic_usec(); (void)ended_ut;
    internal_error(true, "METADATA: 'host:ALL' contexts loaded in %0.2f ms", (double)(ended_ut - started_ut) / USEC_PER_MS);

    worker_is_idle();
}

// Callback after scan of hosts is done
static void after_metadata_hosts(uv_work_t *req, int status __maybe_unused)
{
    struct scan_metadata_payload *data = req->data;
    struct metadata_wc *wc = data->wc;

    metadata_flag_clear(wc, METADATA_FLAG_SCANNING_HOSTS);
    internal_error(true, "METADATA: scanning hosts complete");
    if (unlikely(data->completion)) {
        completion_mark_complete(data->completion);
        internal_error(true, "METADATA: Sending completion done");
    }
    freez(data);
}

static bool metadata_scan_host(RRDHOST *host, uint32_t max_count, bool use_transaction, BUFFER *work_buffer, size_t *query_counter) {
    RRDSET *st;
    int rc;

    bool more_to_do = false;
    uint32_t scan_count = 1;

    if (use_transaction)
        (void)db_execute(db_meta, "BEGIN TRANSACTION;");

    rrdset_foreach_reentrant(st, host) {
        if (scan_count == max_count) {
            more_to_do = true;
            break;
        }
        if(rrdset_flag_check(st, RRDSET_FLAG_METADATA_UPDATE)) {
            (*query_counter)++;

            rrdset_flag_clear(st, RRDSET_FLAG_METADATA_UPDATE);
            scan_count++;

            buffer_flush(work_buffer);
            rc = check_and_update_chart_labels(st, work_buffer, query_counter);
            if (unlikely(rc))
                error_report("METADATA: 'host:%s': Failed to update labels for chart %s", rrdhost_hostname(host), rrdset_name(st));
            else
                (*query_counter)++;

            rc = store_chart_metadata(st);
            if (unlikely(rc))
               error_report("METADATA: 'host:%s': Failed to store metadata for chart %s", rrdhost_hostname(host), rrdset_name(st));
        }

        RRDDIM *rd;
        rrddim_foreach_read(rd, st) {
            if(rrddim_flag_check(rd, RRDDIM_FLAG_METADATA_UPDATE)) {
                (*query_counter)++;

                rrddim_flag_clear(rd, RRDDIM_FLAG_METADATA_UPDATE);

                if (rrddim_option_check(rd, RRDDIM_OPTION_HIDDEN))
                    rrddim_flag_set(rd, RRDDIM_FLAG_META_HIDDEN);
                else
                    rrddim_flag_clear(rd, RRDDIM_FLAG_META_HIDDEN);

                rc = store_dimension_metadata(rd);
                if (unlikely(rc))
                    error_report("METADATA: 'host:%s': Failed to dimension metadata for chart %s. dimension %s",
                                 rrdhost_hostname(host), rrdset_name(st),
                                 rrddim_name(rd));
            }
        }
        rrddim_foreach_done(rd);
    }
    rrdset_foreach_done(st);

    if (use_transaction)
        (void)db_execute(db_meta, "COMMIT TRANSACTION;");

    return more_to_do;
}

static void store_host_and_system_info(RRDHOST *host, BUFFER *work_buffer, size_t *query_counter)
{
    bool free_work_buffer = (NULL == work_buffer);

    if (unlikely(free_work_buffer))
        work_buffer = buffer_create(1024, &netdata_buffers_statistics.buffers_sqlite);

    if (build_host_system_info_statements(host, work_buffer)) {
        int rc = db_execute(db_meta, buffer_tostring(work_buffer));
        if (unlikely(rc)) {
            error_report("METADATA: 'host:%s': Failed to store host updated information in the database", rrdhost_hostname(host));
            rrdhost_flag_set(host, RRDHOST_FLAG_METADATA_INFO | RRDHOST_FLAG_METADATA_UPDATE);
        }
        else {
            if (likely(query_counter))
                (*query_counter)++;
        }
    }

    if (unlikely(store_host_metadata(host))) {
        error_report("METADATA: 'host:%s': Failed to store host info in the database", rrdhost_hostname(host));
        rrdhost_flag_set(host, RRDHOST_FLAG_METADATA_INFO | RRDHOST_FLAG_METADATA_UPDATE);
    }
    else {
        if (likely(query_counter))
            (*query_counter)++;
    }

    if (unlikely(free_work_buffer))
        buffer_free(work_buffer);
}

// Worker thread to scan hosts for pending metadata to store
static void start_metadata_hosts(uv_work_t *req __maybe_unused)
{
    register_libuv_worker_jobs();

    RRDHOST *host;
    int transaction_started = 0;

    struct scan_metadata_payload *data = req->data;
    struct metadata_wc *wc = data->wc;

    BUFFER *work_buffer = data->work_buffer;
    usec_t all_started_ut = now_monotonic_usec(); (void)all_started_ut;
    internal_error(true, "METADATA: checking all hosts...");
    usec_t started_ut = now_monotonic_usec(); (void)started_ut;

    bool run_again = false;
    worker_is_busy(UV_EVENT_METADATA_STORE);

    if (!data->max_count)
        transaction_started = !db_execute(db_meta, "BEGIN TRANSACTION;");

    dfe_start_reentrant(rrdhost_root_index, host) {
        if (rrdhost_flag_check(host, RRDHOST_FLAG_ARCHIVED) || !rrdhost_flag_check(host, RRDHOST_FLAG_METADATA_UPDATE))
            continue;

        size_t query_counter = 0; (void)query_counter;

        rrdhost_flag_clear(host,RRDHOST_FLAG_METADATA_UPDATE);

        if (unlikely(rrdhost_flag_check(host, RRDHOST_FLAG_METADATA_LABELS))) {
            rrdhost_flag_clear(host, RRDHOST_FLAG_METADATA_LABELS);

            int rc = exec_statement_with_uuid(SQL_DELETE_HOST_LABELS, &host->host_uuid);
            if (likely(!rc)) {
                query_counter++;

                buffer_flush(work_buffer);
                struct query_build tmp = {.sql = work_buffer, .count = 0};
                uuid_unparse_lower(host->host_uuid, tmp.uuid_str);
                rrdlabels_walkthrough_read(host->rrdlabels, host_label_store_to_sql_callback, &tmp);
                rc = db_execute(db_meta, buffer_tostring(work_buffer));

                if (unlikely(rc)) {
                    error_report("METADATA: 'host:%s': failed to update metadata host labels", rrdhost_hostname(host));
                    rrdhost_flag_set(host, RRDHOST_FLAG_METADATA_LABELS | RRDHOST_FLAG_METADATA_UPDATE);
                }
                else
                    query_counter++;
            } else {
                error_report("METADATA: 'host:%s': failed to delete old host labels", rrdhost_hostname(host));
                rrdhost_flag_set(host, RRDHOST_FLAG_METADATA_LABELS | RRDHOST_FLAG_METADATA_UPDATE);
            }
        }

        if (unlikely(rrdhost_flag_check(host, RRDHOST_FLAG_METADATA_CLAIMID))) {
            rrdhost_flag_clear(host, RRDHOST_FLAG_METADATA_CLAIMID);
            uuid_t uuid;
            int rc;
            if (likely(host->aclk_state.claimed_id && !uuid_parse(host->aclk_state.claimed_id, uuid)))
                rc = store_claim_id(&host->host_uuid, &uuid);
            else
                rc = store_claim_id(&host->host_uuid, NULL);

            if (unlikely(rc))
                rrdhost_flag_set(host, RRDHOST_FLAG_METADATA_CLAIMID | RRDHOST_FLAG_METADATA_UPDATE);
            else
                query_counter++;
        }
        if (unlikely(rrdhost_flag_check(host, RRDHOST_FLAG_METADATA_INFO))) {
            rrdhost_flag_clear(host, RRDHOST_FLAG_METADATA_INFO);
            store_host_and_system_info(host, work_buffer, &query_counter);
        }

        // For clarity
        bool use_transaction = data->max_count;
        if (unlikely(metadata_scan_host(host, data->max_count, use_transaction, work_buffer, &query_counter))) {
            run_again = true;
            rrdhost_flag_set(host,RRDHOST_FLAG_METADATA_UPDATE);
            internal_error(true,"METADATA: 'host:%s': scheduling another run, more charts to store", rrdhost_hostname(host));
        }
        usec_t ended_ut = now_monotonic_usec(); (void)ended_ut;
        internal_error(true, "METADATA: 'host:%s': saved metadata with %zu SQL statements, in %0.2f ms",
                       rrdhost_hostname(host), query_counter,
                       (double)(ended_ut - started_ut) / USEC_PER_MS);
    }
    dfe_done(host);

    if (!data->max_count && transaction_started)
        transaction_started = db_execute(db_meta, "COMMIT TRANSACTION;");

    usec_t all_ended_ut = now_monotonic_usec(); (void)all_ended_ut;
    internal_error(true, "METADATA: checking all hosts completed in %0.2f ms",
                   (double)(all_ended_ut - all_started_ut) / USEC_PER_MS);

    if (unlikely(run_again))
        wc->check_hosts_after = now_realtime_sec() + METADATA_HOST_CHECK_IMMEDIATE;
    else
        wc->check_hosts_after = now_realtime_sec() + METADATA_HOST_CHECK_INTERVAL;
    worker_is_idle();
}

static void metadata_event_loop(void *arg)
{
    worker_register("METASYNC");
    worker_register_job_name(METADATA_DATABASE_NOOP,        "noop");
    worker_register_job_name(METADATA_DATABASE_TIMER,       "timer");
    worker_register_job_name(METADATA_DEL_DIMENSION,        "delete dimension");
    worker_register_job_name(METADATA_STORE_CLAIM_ID,       "add claim id");
    worker_register_job_name(METADATA_ADD_HOST_INFO,        "add host info");
    worker_register_job_name(METADATA_MAINTENANCE,          "maintenance");
    worker_register_job_name(METADATA_ML_LOAD_MODELS,       "ml load models");

    int ret;
    uv_loop_t *loop;
    unsigned cmd_batch_size;
    struct metadata_wc *wc = arg;
    enum metadata_opcode opcode;
    uv_work_t metadata_cleanup_worker;

    uv_thread_set_name_np(wc->thread, "METASYNC");
//    service_register(SERVICE_THREAD_TYPE_EVENT_LOOP, NULL, NULL, NULL, true);
    loop = wc->loop = mallocz(sizeof(uv_loop_t));
    ret = uv_loop_init(loop);
    if (ret) {
        error("uv_loop_init(): %s", uv_strerror(ret));
        goto error_after_loop_init;
    }
    loop->data = wc;

    ret = uv_async_init(wc->loop, &wc->async, async_cb);
    if (ret) {
        error("uv_async_init(): %s", uv_strerror(ret));
        goto error_after_async_init;
    }
    wc->async.data = wc;

    ret = uv_timer_init(loop, &wc->timer_req);
    if (ret) {
        error("uv_timer_init(): %s", uv_strerror(ret));
        goto error_after_timer_init;
    }
    wc->timer_req.data = wc;
    fatal_assert(0 == uv_timer_start(&wc->timer_req, timer_cb, TIMER_INITIAL_PERIOD_MS, TIMER_REPEAT_PERIOD_MS));

    info("Starting metadata sync thread with %d entries command queue", METADATA_CMD_Q_MAX_SIZE);

    struct metadata_cmd cmd;
    memset(&cmd, 0, sizeof(cmd));
    metadata_flag_clear(wc, METADATA_FLAG_CLEANUP);
    metadata_flag_clear(wc, METADATA_FLAG_SCANNING_HOSTS);

    wc->check_metadata_after = now_realtime_sec() + METADATA_MAINTENANCE_FIRST_CHECK;
    wc->check_hosts_after    = now_realtime_sec() + METADATA_HOST_CHECK_FIRST_CHECK;

    int shutdown = 0;
    wc->row_id = 0;
    completion_mark_complete(&wc->init_complete);
    BUFFER *work_buffer = buffer_create(1024, &netdata_buffers_statistics.buffers_sqlite);
    struct scan_metadata_payload *data;

    while (shutdown == 0 || (wc->flags & METADATA_WORKER_BUSY)) {
        uuid_t  *uuid;
        RRDHOST *host = NULL;

        worker_is_idle();
        uv_run(loop, UV_RUN_DEFAULT);

        /* wait for commands */
        cmd_batch_size = 0;
        do {
            if (unlikely(cmd_batch_size >= METADATA_MAX_BATCH_SIZE))
                break;

            cmd = metadata_deq_cmd(wc);
            opcode = cmd.opcode;

            if (unlikely(opcode == METADATA_DATABASE_NOOP && metadata_flag_check(wc, METADATA_FLAG_SHUTDOWN))) {
                shutdown = 1;
                continue;
            }

            ++cmd_batch_size;

            if (likely(opcode != METADATA_DATABASE_NOOP))
                worker_is_busy(opcode);

            switch (opcode) {
                case METADATA_DATABASE_NOOP:
                case METADATA_DATABASE_TIMER:
                    break;

                case METADATA_ML_LOAD_MODELS: {
                    RRDDIM *rd = (RRDDIM *) cmd.param[0];
                    ml_dimension_load_models(rd);
                    break;
                }
                case METADATA_DEL_DIMENSION:
                    uuid = (uuid_t *) cmd.param[0];
                    if (likely(dimension_can_be_deleted(uuid)))
                        delete_dimension_uuid(uuid);
                    freez(uuid);
                    break;
                case METADATA_STORE_CLAIM_ID:
                    store_claim_id((uuid_t *) cmd.param[0], (uuid_t *) cmd.param[1]);
                    freez((void *) cmd.param[0]);
                    freez((void *) cmd.param[1]);
                    break;
                case METADATA_ADD_HOST_INFO:
                    host = (RRDHOST *) cmd.param[0];
                    store_host_and_system_info(host, NULL, NULL);
                    break;
                case METADATA_SCAN_HOSTS:
                    if (unlikely(metadata_flag_check(wc, METADATA_FLAG_SCANNING_HOSTS)))
                        break;

                    if (unittest_running)
                        break;

                    data = mallocz(sizeof(*data));
                    data->request.data = data;
                    data->wc = wc;
                    data->completion = cmd.completion;  // Completion by the worker
                    data->work_buffer = work_buffer;

                    if (unlikely(cmd.completion)) {
                        data->max_count = 0;            // 0 will process all pending updates
                        cmd.completion = NULL;          // Do not complete after launching worker (worker will do)
                    }
                    else
                        data->max_count = 5000;

                    metadata_flag_set(wc, METADATA_FLAG_SCANNING_HOSTS);
                    if (unlikely(
                            uv_queue_work(loop,&data->request,
                                          start_metadata_hosts,
                                          after_metadata_hosts))) {
                        // Failed to launch worker -- let the event loop handle completion
                        cmd.completion = data->completion;
                        freez(data);
                        metadata_flag_clear(wc, METADATA_FLAG_SCANNING_HOSTS);
                    }
                    break;
                case METADATA_LOAD_HOST_CONTEXT:;
                    if (unittest_running)
                        break;

                    data = callocz(1,sizeof(*data));
                    data->request.data = data;
                    data->wc = wc;
                    if (unlikely(
                            uv_queue_work(loop,&data->request, start_all_host_load_context,
                                          after_start_host_load_context))) {
                        freez(data);
                    }
                    break;
                case METADATA_MAINTENANCE:
                    if (unlikely(metadata_flag_check(wc, METADATA_FLAG_CLEANUP)))
                        break;

                    metadata_cleanup_worker.data = wc;
                    metadata_flag_set(wc, METADATA_FLAG_CLEANUP);
                    if (unlikely(
                            uv_queue_work(loop, &metadata_cleanup_worker, start_metadata_cleanup, after_metadata_cleanup))) {
                        metadata_flag_clear(wc, METADATA_FLAG_CLEANUP);
                    }
                    break;
                case METADATA_UNITTEST:;
                    struct thread_unittest *tu = (struct thread_unittest *) cmd.param[0];
                    sleep_usec(1000); // processing takes 1ms
                    __atomic_fetch_add(&tu->processed, 1, __ATOMIC_SEQ_CST);
                    break;
                default:
                    break;
            }

            if (cmd.completion)
                completion_mark_complete(cmd.completion);
        } while (opcode != METADATA_DATABASE_NOOP);
    }

    if (!uv_timer_stop(&wc->timer_req))
        uv_close((uv_handle_t *)&wc->timer_req, NULL);

    uv_close((uv_handle_t *)&wc->async, NULL);
    uv_cond_destroy(&wc->cmd_cond);
    int rc;
    do {
        rc = uv_loop_close(loop);
    } while (rc != UV_EBUSY);

    buffer_free(work_buffer);
    freez(loop);
    worker_unregister();

    info("METADATA: Shutting down event loop");
    completion_mark_complete(&wc->init_complete);
    return;

error_after_timer_init:
    uv_close((uv_handle_t *)&wc->async, NULL);
error_after_async_init:
    fatal_assert(0 == uv_loop_close(loop));
error_after_loop_init:
    freez(loop);
    worker_unregister();
}

struct metadata_wc metasync_worker = {.loop = NULL};

void metadata_sync_shutdown(void)
{
    completion_init(&metasync_worker.init_complete);

    struct metadata_cmd cmd;
    memset(&cmd, 0, sizeof(cmd));
    info("METADATA: Sending a shutdown command");
    cmd.opcode = METADATA_SYNC_SHUTDOWN;
    metadata_enq_cmd(&metasync_worker, &cmd);

    /* wait for metadata thread to shut down */
    info("METADATA: Waiting for shutdown ACK");
    completion_wait_for(&metasync_worker.init_complete);
    completion_destroy(&metasync_worker.init_complete);
    info("METADATA: Shutdown complete");
}

void metadata_sync_shutdown_prepare(void)
{
    if (unlikely(!metasync_worker.loop))
        return;

    struct metadata_cmd cmd;
    memset(&cmd, 0, sizeof(cmd));

    struct completion compl;
    completion_init(&compl);

    info("METADATA: Sending a scan host command");
    uint32_t max_wait_iterations = 2000;
    while (unlikely(metadata_flag_check(&metasync_worker, METADATA_FLAG_SCANNING_HOSTS)) && max_wait_iterations--) {
        if (max_wait_iterations == 1999)
            info("METADATA: Current worker is running; waiting to finish");
        sleep_usec(1000);
    }

    cmd.opcode = METADATA_SCAN_HOSTS;
    cmd.completion = &compl;
    metadata_enq_cmd(&metasync_worker, &cmd);

    info("METADATA: Waiting for host scan completion");
    completion_wait_for(&compl);
    completion_destroy(&compl);
    info("METADATA: Host scan complete; can continue with shutdown");
}

// -------------------------------------------------------------
// Init function called on agent startup

void metadata_sync_init(void)
{
    struct metadata_wc *wc = &metasync_worker;

    memset(wc, 0, sizeof(*wc));
    metadata_init_cmd_queue(wc);
    completion_init(&wc->init_complete);

    fatal_assert(0 == uv_thread_create(&(wc->thread), metadata_event_loop, wc));

    completion_wait_for(&wc->init_complete);
    completion_destroy(&wc->init_complete);

    info("SQLite metadata sync initialization complete");
}


//  Helpers

static inline void queue_metadata_cmd(enum metadata_opcode opcode, const void *param0, const void *param1)
{
    struct metadata_cmd cmd;
    cmd.opcode = opcode;
    cmd.param[0] = param0;
    cmd.param[1] = param1;
    cmd.completion = NULL;
    metadata_enq_cmd(&metasync_worker, &cmd);

}

// Public
void metaqueue_delete_dimension_uuid(uuid_t *uuid)
{
    if (unlikely(!metasync_worker.loop))
        return;
    uuid_t *use_uuid = mallocz(sizeof(*uuid));
    uuid_copy(*use_uuid, *uuid);
    queue_metadata_cmd(METADATA_DEL_DIMENSION, use_uuid, NULL);
}

void metaqueue_store_claim_id(uuid_t *host_uuid, uuid_t *claim_uuid)
{
    if (unlikely(!host_uuid))
        return;

    uuid_t *local_host_uuid = mallocz(sizeof(*host_uuid));
    uuid_t *local_claim_uuid = NULL;

    uuid_copy(*local_host_uuid, *host_uuid);
    if (likely(claim_uuid)) {
        local_claim_uuid = mallocz(sizeof(*claim_uuid));
        uuid_copy(*local_claim_uuid, *claim_uuid);
    }
    queue_metadata_cmd(METADATA_STORE_CLAIM_ID, local_host_uuid, local_claim_uuid);
}

void metaqueue_host_update_info(RRDHOST *host)
{
    if (unlikely(!metasync_worker.loop))
        return;
    queue_metadata_cmd(METADATA_ADD_HOST_INFO, host, NULL);
}

void metaqueue_ml_load_models(RRDDIM *rd)
{
    if (unlikely(!metasync_worker.loop))
        return;
    queue_metadata_cmd(METADATA_ML_LOAD_MODELS, rd, NULL);
}

void metadata_queue_load_host_context(RRDHOST *host)
{
    if (unlikely(!metasync_worker.loop))
        return;
    queue_metadata_cmd(METADATA_LOAD_HOST_CONTEXT, host, NULL);
}

//
// unitests
//

static void *unittest_queue_metadata(void *arg) {
    struct thread_unittest *tu = arg;

    struct metadata_cmd cmd;
    cmd.opcode = METADATA_UNITTEST;
    cmd.param[0] = tu;
    cmd.param[1] = NULL;
    cmd.completion = NULL;
    metadata_enq_cmd(&metasync_worker, &cmd);

    do {
        __atomic_fetch_add(&tu->added, 1, __ATOMIC_SEQ_CST);
        metadata_enq_cmd(&metasync_worker, &cmd);
        sleep_usec(10000);
    } while (!__atomic_load_n(&tu->join, __ATOMIC_RELAXED));
    return arg;
}

static void *metadata_unittest_threads(void)
{

    unsigned done;

    struct thread_unittest tu = {
        .join = 0,
        .added = 0,
        .processed = 0,
        .done = &done,
    };

    // Queue messages / Time it
    time_t seconds_to_run = 5;
    int threads_to_create = 4;
    fprintf(
        stderr,
        "\nChecking metadata queue using %d threads for %lld seconds...\n",
        threads_to_create,
        (long long)seconds_to_run);

    netdata_thread_t threads[threads_to_create];
    tu.join = 0;
    for (int i = 0; i < threads_to_create; i++) {
        char buf[100 + 1];
        snprintf(buf, 100, "META[%d]", i);
        netdata_thread_create(
            &threads[i],
            buf,
            NETDATA_THREAD_OPTION_DONT_LOG | NETDATA_THREAD_OPTION_JOINABLE,
            unittest_queue_metadata,
            &tu);
    }
    (void) uv_async_send(&metasync_worker.async);
    sleep_usec(seconds_to_run * USEC_PER_SEC);

    __atomic_store_n(&tu.join, 1, __ATOMIC_RELAXED);
    for (int i = 0; i < threads_to_create; i++) {
        void *retval;
        netdata_thread_join(threads[i], &retval);
    }
    sleep_usec(5 * USEC_PER_SEC);

    fprintf(stderr, "Added %u elements, processed %u\n", tu.added, tu.processed);

    return 0;
}

int metadata_unittest(void)
{
    metadata_sync_init();

    // Queue items for a specific period of time
    metadata_unittest_threads();

    fprintf(stderr, "Items still in queue %u\n", metasync_worker.queue_size);
    metadata_sync_shutdown();

    return 0;
}