Learn how to use Netdata's powerful charts to troubleshoot with real-time, per-second metric data.
Netdata excels in collecting, storing, and organizing metrics in out-of-the-box dashboards. To make sense of all the metrics, Netdata offers an enhanced version of charts that update every second.
These charts provide a lot of useful information, so that you can:
These charts are available on Netdata Cloud's Metrics tab, single node tabs and on your Custom Dashboards.
A Netdata chart looks like this:
With a quick glance you have immediate information available at your disposal:
While Netdata's charts require no configuration and are easy to interact with, they have a lot of underlying complexity. To meaningfully organize charts out of the box based on what's happening in your nodes, Netdata uses the concepts of dimensions, contexts, and families.
Understanding how these work will help you more easily navigate the dashboard, write new alerts, or play around with the API.
A dimension is a value that gets shown on a chart. The value can be raw data or calculated values, such as the
average (the default), minimum, or maximum. These values can then be given any type of unit. For example, CPU
utilization is represented as a percentage, disk I/O as MiB/s
, and available RAM as an absolute value in MiB
or
GiB
.
Beneath every chart (or on the right-side if you configure the dashboard) is a legend of dimensions. When there are multiple dimensions, you'll see a different entry in the legend for each dimension.
The Apps CPU Time chart (with the context apps.cpu
), which visualizes CPU utilization of
different types of processes/services/applications on your node, always provides a vibrant example of a chart with
multiple dimensions.
Dimensions can be hidden to help you focus your attention.
A context is a way of grouping charts by the types of metrics collected and dimensions displayed. It's like a machine-readable naming and organization scheme.
For example, the Apps CPU Time has the context apps.cpu
. A little further down on the dashboard is a similar
chart, Apps Real Memory (w/o shared) with the context apps.mem
. The apps
portion of the context is the type,
whereas anything after the .
is specified either by the chart's developer or by the family.
By default, a chart's type affects where it fits in the menu, while its family creates submenus.
Netdata also relies on contexts for alert configuration (the on
line).
Families are a single instance of a hardware or software resource that needs to be displayed separately from similar instances.
For example, let's look at the Disks section, which contains a number of charts with contexts like disk.io
,
disk.ops
, disk.backlog
, and disk.util
. If your node has multiple disk drives at sda
and sdb
, Netdata creates
a separate family for each.
Netdata now merges the contexts and families to create charts that are grouped by family, following a
[context].[family]
naming scheme, so that you can see the disk.io
and disk.ops
charts for sda
right next to each
other.
Given the four example contexts, and two families of sda
and sdb
, Netdata will create the following charts and their
names:
Context | sda family |
sdb family |
---|---|---|
disk.io |
disk_io.sda |
disk_io.sdb |
disk.ops |
disk_ops.sda |
disk_ops.sdb |
disk.backlog |
disk_backlog.sda |
disk_backlog.sdb |
disk.util |
disk_util.sda |
disk_util.sdb |
When you start interacting with a chart, you'll notice valuable information on the Title bar:
Title bar elements:
Along with viewing chart type, context and units, on this bar you have access to immediate actions over the chart:
Each composite chart has a definition bar to provide information and options about the following:
To help users instantly understand and validate the data they see on charts, we developed the NIDL (Nodes, Instances, Dimensions, Labels) framework. This information is visualized on all charts.
You can explore the in-depth infographic, by clicking on this image and opening it in a new tab, allowing you to zoom in to the different parts of it.
You can rapidly access condensed information for collected metrics, grouped by node, monitored instances, dimension, or any key/value label pair.
At the Definition bar of each chart, there are a few dropdown menus:
These dropdown menus have 2 functions:
The NIDL framework attaches metadata to every metric that is collected to provide for each of them the following consolidated data for the visible time frame:
key:value
pair to the final query, so that you can immediately see how much every label value involved in the query affected the chart.All of these dropdown menus can be used for instantly filtering the information shown, by including or excluding specific nodes, instances, dimensions or labels. Directly from the dropdown menu, without the need to edit a query string and without any additional knowledge of the underlying data.
The "Group by" dropdown menu allows selecting 1 or more groupings to be applied at once on the same dataset.
It supports:
Using this menu, you can slice and dice the data in any possible way, to quickly get different views of it, without the need to edit a query string and without any need to better understand the format of the underlying data.
Tip
A very pertinent example is composite charts over contexts related to cgroups (VMs and containers). You have the means to change the default group by or apply filtering to get a better view into what data your are trying to analyze. For example, if you change the group by to instance you get a view with the data of all the instances (cgroups) that contribute to that chart. Then you can use further filtering tools to focus the data that is important to you and even save the result to your own dashboards.
Tip
Group by instance, dimension to see the time series of every individual collected metric participating in the chart.
Each chart uses an opinionated-but-valuable default aggregate function over the data sources.
For example, the system.cpu
chart shows the average for each dimension from every contributing chart, while the net.net
chart shows the sum for each dimension from every contributing chart, which can also come from multiple networking interfaces.
The following aggregate functions are available for each selected dimension:
out
dimension—-2.1
, -5.5
, -10.2
, -15
, -0.1
—the composite chart displays a
value of −6.58
.-32.9
.In this dropdown, you can view or filter the nodes contributing time-series metrics to the chart.
This menu also provides the contribution of each node to the volume of the chart, and a break down of the anomaly rate of the queried data per node.
If one or more nodes can't contribute to a given chart, the definition bar shows a warning symbol plus the number of
affected nodes, then lists them in the dropdown along with the associated error. Nodes might return errors because of
networking issues, a stopped netdata
service, or because that node does not have any metrics for that context.
In this dropdown, you can view or filter the instances contributing time-series metrics to the chart.
This menu also provides the contribution of each instance to the volume of the chart, and a break down of the anomaly rate of the queried data per instance.
In this dropdown, you can view or filter the original dimensions contributing time-series metrics to the chart.
This menu also presents the contribution of each original dimensions on the chart, and a break down of the anomaly rate of the data per dimension.
In this dropdown, you can view or filter the contributing time-series labels of the chart.
This menu also presents the contribution of each label on the chart,and a break down of the anomaly rate of the data per label.
When the granularity of the data collected is higher than the plotted points on the chart an aggregation function over time is applied.
By default the aggregation applied is average but the user can choose different options from the following:
For more details on each, you can refer to our Agent's HTTP API details on Data Queries - Data Grouping.
Finally, you can reset everything to its defaults by clicking the green "Reset" prompt at the end of the definition bar.
Netdata's unsupervised machine learning algorithm creates a unique model for each metric collected by your Agents, using exclusively the metric's past data. It then uses these unique models during data collection to predict the value that should be collected and check if the collected value is within the range of acceptable values based on past patterns and behavior.
If the value collected is an outlier, it is marked as anomalous.
This unmatched capability of real-time predictions as data is collected allows you to detect anomalies for potentially millions of metrics across your entire infrastructure within a second of occurrence.
The Anomaly Rate ribbon on top of each chart visualizes the combined anomaly rate of all the underlying data, highlighting areas of interest that may not be easily visible to the naked eye.
Hovering over the Anomaly Rate ribbon provides a histogram of the anomaly rates per presented dimension, for the specific point in time.
Anomaly Rate visualization does not make Netdata slower. Anomaly rate is saved in the Netdata database, together with metric values, and due to the smart design of Netdata, it does not even incur a disk footprint penalty.
Hovering over any point in the chart will reveal a more informative overlay. It includes a bar indicating the volume percentage of each time series compared to the total, the anomaly rate, and a notification on if there are data collection issues.
This overlay sorts all dimensions by value, makes bold the closest dimension to the mouse and presents a histogram based on the values of the dimensions.
When hovering the anomaly ribbon, the overlay sorts all dimensions by anomaly rate, and presents a histogram of these anomaly rates.
Additionally, when hovering over the chart, the overlay may display an indication in the "Info" column.
Currently, this column is used to inform users of any data collection issues that might affect the chart. Below each chart, there is an information ribbon. This ribbon currently shows 3 states related to the points presented in the chart:
Partial Data At least one of the dimensions in the chart has partial data, meaning that not all instances available contributed data to this point. This can happen when a container is stopped, or when a node is restarted. This indicator helps to gain confidence of the dataset, in situations when unusual spikes or dives appear due to infrastructure maintenance, or due to failures to part of the infrastructure.
Overflown At least one of the data sources included in the chart has a counter that has overflowed at this point.
Empty Data At least one of the dimensions included in the chart has no data at all for the given points.
All these indicators are also visualized per dimension, in the pop-over that appears when hovering the chart.
Your charts are controlled using the available Time controls. Besides these, when interacting with the chart you can also activate these controls by:
Interaction | Keyboard/mouse | Touchpad/touchscreen | Time control |
---|---|---|---|
Pause a chart | hover |
n/a |
Temporarily Pause |
Stop a chart | click |
tap |
Pause |
Reset a chart | double click |
n/a |
Play |
Note: These interactions are available when the default "Pan" action is used from the Tool Bar.
While exploring the chart, a tool bar will appear. This tool bar is there to support you on this task. The available manipulation tools you can select are:
Drag your mouse/finger to the right to pan backward through time, or drag to the left to pan forward in time. Think of it like pushing the current timeframe off the screen to see what came before or after.
Interaction | Keyboard | Mouse | Touchpad/touchscreen |
---|---|---|---|
Pan | n/a |
click + drag |
touch drag |
Selecting timeframes is useful when you see an interesting spike or change in a chart and want to investigate further by:
Interaction | Keyboard/mouse | Touchpad/touchscreen |
---|---|---|
Highlight a specific timeframe | Alt + mouse selection or ⌘ + mouse selection (macOS) |
n/a |
Note
To clear a highlighted timeframe, simply click on the chart area.
You can zoom to a specific timeframe, either horizontally of vertically, by selecting a timeframe.
Interaction | Keyboard/mouse | Touchpad/touchscreen |
---|---|---|
Zoom to a specific timeframe | Shift + mouse vertical selection |
n/a |
Horizontal Zoom a specific Y-axis area | Shift + mouse horizontal selection |
n/a |
Zooming in helps you see metrics with maximum granularity, which is useful when you're trying to diagnose the root cause of an anomaly or outage.
Zooming out lets you see metrics within the larger context, such as the last hour, day, or week, which is useful in understanding what "normal" looks like, or to identify long-term trends, like a slow creep in memory usage.
Interaction | Keyboard/mouse | Touchpad/touchscreen |
---|---|---|
Zoom in or out | Shift + mouse scrollwheel |
two-finger pinch Shift + two-finger scroll |
The bottom legend where you can see the dimensions of the chart can be ordered by:
Hiding dimensions simplifies the chart and can help you better discover exactly which aspect of your system might be behaving strangely.
Interaction | Keyboard/mouse | Touchpad/touchscreen |
---|---|---|
Show one dimension and hide others | click |
tap |
Toggle (show/hide) one dimension | Shift + click |
n/a |
To resize the chart, click-and-drag the icon on the bottom-right corner of any chart. To restore the chart to its original height, double-click the same icon.