Prometheus integration

Introduced in GitLab 9.0.

GitLab offers powerful integration with Prometheus for monitoring key metrics of your apps, directly within GitLab. Metrics for each environment are retrieved from Prometheus, and then displayed within the GitLab interface.

There are two ways to set up Prometheus integration, depending on where your apps are running:

For deployments on Kubernetes, GitLab can automatically deploy and manage Prometheus.
For other deployment targets, simply specify the Prometheus server.

Once enabled, GitLab will automatically detect metrics from known services in the metric library. You are also able to add your own metrics as well.

Enabling Prometheus Integration

Managed Prometheus on Kubernetes

Introduced in GitLab 10.5.

GitLab can seamlessly deploy and manage Prometheus on a connected Kubernetes cluster, making monitoring of your apps easy.

Requirements

A connected Kubernetes cluster
Helm Tiller installed by GitLab

Getting started

Once you have a connected Kubernetes cluster with Helm installed, deploying a managed Prometheus is as easy as a single click.

Go to the Operations > Kubernetes page to view your connected clusters
Select the cluster you would like to deploy Prometheus to
Click the Install button to deploy Prometheus to the cluster

Getting metrics to display on the Metrics Dashboard

After completing the steps above, you will also need deployments in order to view the Operations > Metrics page. Setting up Auto DevOps will help you to quickly create a deployment:

Navigate to your project's Operations > Kubernetes page, and ensure that, in addition to "Prometheus" and "Helm Tiller", you also have "Runner" and "Ingress" installed. Once "Ingress" is installed, copy its endpoint.
Navigate to your project's Settings > CI/CD page. In the Auto DevOps section, select a deployment strategy and save your changes.
On the same page, in the Variables section, add a variable named KUBE_INGRESS_BASE_DOMAIN with the value of the Ingress endpoint you have copied in the previous step. Leave the type as "Variable".
Navigate to your project's CI/CD > Pipelines page, and run a pipeline on any branch.
When the pipeline has run successfully, graphs will be available on the Operations > Metrics page.

About managed Prometheus deployments

Prometheus is deployed into the gitlab-managed-apps namespace, using the official Helm chart. Prometheus is only accessible within the cluster, with GitLab communicating through the Kubernetes API.

The Prometheus server will automatically detect and monitor nodes, pods, and endpoints. To configure a resource to be monitored by Prometheus, simply set the following Kubernetes annotations:

prometheus.io/scrape to true to enable monitoring of the resource.
prometheus.io/port to define the port of the metrics endpoint.
prometheus.io/path to define the path of the metrics endpoint. Defaults to /metrics.

CPU and Memory consumption is monitored, but requires naming conventions in order to determine the environment. If you are using Auto DevOps, this is handled automatically.

The NGINX Ingress that is deployed by GitLab to clusters, is automatically annotated for monitoring providing key response metrics: latency, throughput, and error rates.

Manual configuration of Prometheus

Requirements

Integration with Prometheus requires the following:

GitLab 9.0 or higher
Prometheus must be configured to collect one of the supported metrics
Each metric must be have a label to indicate the environment
GitLab must have network connectivity to the Prometheus server

Getting started

Installing and configuring Prometheus to monitor applications is fairly straight forward.

Install Prometheus
Set up one of the supported monitoring targets
Configure the Prometheus server to collect their metrics

Configuration in GitLab

The actual configuration of Prometheus integration within GitLab is very simple. All you will need is the DNS or IP address of the Prometheus server you'd like to integrate with.

Navigate to the Integrations page
Click the Prometheus service
Provide the base URL of the your server, for example http://prometheus.example.com/. The Test Settings button can be used to confirm connectivity from GitLab to the Prometheus server.

Monitoring CI/CD Environments

Once configured, GitLab will attempt to retrieve performance metrics for any environment which has had a successful deployment.

GitLab will automatically scan the Prometheus server for metrics from known servers like Kubernetes and NGINX, and attempt to identify individual environment. The supported metrics and scan process is detailed in our Prometheus Metrics Library documentation.

You can view the performance dashboard for an environment by clicking on the monitoring button.

Adding additional metrics (PREMIUM)

Introduced in GitLab Premium 10.6.

Custom metrics can be monitored by adding them on the monitoring dashboard page. Once saved, they will be displayed on the environment performance dashboard provided that either:

A connected Kubernetes cluster with the environment scope of * is used and Prometheus installed on the cluster
Prometheus is manually configured.

A few fields are required:

Name: Chart title
Type: Type of metric. Metrics of the same type will be shown together.
Query: Valid PromQL query.
Y-axis label: Y axis title to display on the dashboard.
Unit label: Query units, for example req / sec. Shown next to the value.

Multiple metrics can be displayed on the same chart if the fields Name, Type, and Y-axis label match between metrics. For example, a metric with Name Requests Rate, Type Business, and Y-axis label rec / sec would display on the same chart as a second metric with the same values. A Legend label is suggested if this feature used.

Query Variables

GitLab supports a limited set of CI variables in the Prometheus query. This is particularly useful for identifying a specific environment, for example with CI_ENVIRONMENT_SLUG. The supported variables are:

CI_ENVIRONMENT_SLUG
KUBE_NAMESPACE

To specify a variable in a query, enclose it in quotation marks with curly braces with a leading percent. For example: "%{ci_environment_slug}".

Defining custom dashboards per project

Introduced in GitLab 12.1.

By default, all projects include a GitLab-defined Prometheus dashboard, which includes a few key metrics, but you can also define your own custom dashboards.

NOTE: Note: The custom metrics as defined below do not support alerts, unlike additional metrics.

Dashboards have several components:

Panel groups, which comprise panels.
Panels, which support one or more metrics.

To configure a custom dashboard:

Create a YAML file with the .yml extension under your repository's root directory inside .gitlab/dashboards/. For example, create .gitlab/dashboards/prom_alerts.yml with the following contents:

dashboard: 'Dashboard Title'
panel_groups:
  - group: 'Group Title'
    panels:
      - type: area-chart
        title: "Chart Title"
        y_label: "Y-Axis"
        metrics:
          - id: metric_of_ages
            query_range: 'http_requests_total'
            label: "Metric of Ages"
            unit: "count"

The above sample dashboard would display a single area chart. Each file should define the layout of the dashboard and the Prometheus queries used to populate data.

Save the file, commit, and push to your repository.
Navigate to your project's Operations > Metrics and choose the custom dashboard from the dropdown.

NOTE: Note: Configuration files nested under subdirectories of .gitlab/dashboards are not supported and will not be available in the UI.

The following tables outline the details of expected properties.

Dashboard properties:

Property	Type	Required	Description
`dashboard`	string	yes	Heading for the dashboard. Only one dashboard should be defined per file.
`panel_groups`	array	yes	The panel groups which should be on the dashboard.

Panel group (panel_groups) properties:

Property	Type	Required	Description
`group`	string	required	Heading for the panel group.
`priority`	number	optional, defaults to order in file	Order to appear on the dashboard. Higher number means higher priority, which will be higher on the page. Numbers do not need to be consecutive.
`panels`	array	required	The panels which should be in the panel group.

Panel (panels) properties:

Property	Type	Required	Description
`type`	enum	no, defaults to `area-chart`	Specifies the chart type to use, can be: `area-chart`, `line-chart` or `anomaly-chart`.
`title`	string	yes	Heading for the panel.
`y_label`	string	no, but highly encouraged	Y-Axis label for the panel.
`weight`	number	no, defaults to order in file	Order to appear within the grouping. Lower number means higher priority, which will be higher on the page. Numbers do not need to be consecutive.
`metrics`	array	yes	The metrics which should be displayed in the panel. Any number of metrics can be displayed when `type` is `area-chart` or `line-chart`, whereas only 3 can be displayed when `type` is `anomaly-chart`.

Metrics (metrics) properties:

Property	Type	Required	Description
`id`	string	no	Used for associating dashboard metrics with database records. Must be unique across dashboard configuration files. Required for alerting (support not yet enabled, see relevant issue).
`unit`	string	yes	Defines the unit of the query's return data.
`label`	string	no, but highly encouraged	Defines the legend-label for the query. Should be unique within the panel's metrics.
`query`	string	yes if `query_range` is not defined	Defines the Prometheus query to be used to populate the chart/panel. If defined, the `query` endpoint of the Prometheus API will be utilized.
`query_range`	string	yes if `query` is not defined	Defines the Prometheus query to be used to populate the chart/panel. If defined, the `query_range` endpoint of the Prometheus API will be utilized.

Panel types for dashboards

The below panel types are supported in monitoring dashboards.

Area or Line Chart

To add an area chart panel type to a dashboard, look at the following sample dashboard file:

dashboard: 'Dashboard Title'
panel_groups:
  - group: 'Group Title'
    panels:
      - type: area-chart # or line-chart
        title: 'Area Chart Title'
        y_label: "Y-Axis"
        metrics:
          - id: area_http_requests_total
            query_range: 'http_requests_total'
            label: "Metric of Ages"
            unit: "count"

Note the following properties:

Property	Type	Required	Description
type	string	no	Type of panel to be rendered. Optional for area panel types
query_range	string	required	For area panel types, you must use a range query

Anomaly chart

Introduced in GitLab 12.5.

To add an anomaly chart panel type to a dashboard, add add a panel with exactly 3 metrics.

The first metric represents the current state, and the second and third metrics represent the upper and lower limit respectively:

dashboard: 'Dashboard Title'
panel_groups:
  - group: 'Group Title'
    panels:
      - type: anomaly-chart
        title: "Chart Title"
        y_label: "Y-Axis"
        metrics:
          - id: anomaly_requests_normal
            query_range: 'http_requests_total'
            label: "# of Requests"
            unit: "count"
        metrics:
          - id: anomaly_requests_upper_limit
            query_range: 10000
            label: "Max # of requests"
            unit: "count"
        metrics:
          - id: anomaly_requests_lower_limit
            query_range: 2000
            label: "Min # of requests"
            unit: "count"

Note the following properties:

Property	Type	Required	Description
type	string	required	Must be `anomaly-chart` for anomaly panel types
query_range	yes	required	For anomaly panel types, you must use a range query in every metric.

Single Stat

To add a single stat panel type to a dashboard, look at the following sample dashboard file:

dashboard: 'Dashboard Title'
panel_groups:
  - group: 'Group Title'
    panels:
      - title: "Single Stat"
        type: "single-stat"
        metrics:
        - id: 10
          query: 'max(go_memstats_alloc_bytes{job="prometheus"})'
          unit: MB
          label: "Total"

Note the following properties:

Property	Type	Required	Description
type	string	yes	Type of panel to be rendered. For single stat panel types, set to `single-stat`
query	string	yes	For single stat panel types, you must use an instant query

Heatmaps

Introduced in GitLab 12.5.

To add a heatmap panel type to a dashboard, look at the following sample dashboard file:

dashboard: 'Dashboard Title'
panel_groups:
  - group: 'Group Title'
    panels:
      - title: "Heatmap"
        type: "heatmap"
        metrics:
        - id: 10
          query: 'sum(rate(nginx_upstream_responses_total{upstream=~"%{kube_namespace}-%{ci_environment_slug}-.*"}[60m])) by (status_code)'
          unit: req/sec
          label: "Status code"

Note the following properties:

Property	Type	Required	Description
type	string	yes	Type of panel to be rendered. For heatmap panel types, set to `heatmap`
query_range	yes	yes	For area panel types, you must use a range query

View and edit the source file of a custom dashboard

Introduced in GitLab 12.5.

When viewing a custom dashboard of a project, you can view the original .yml file by clicking on Edit dashboard button.

Downloading data as CSV

Data from Prometheus charts on the metrics dashboard can be downloaded as CSV.

Setting up alerts for Prometheus metrics (ULTIMATE)

Managed Prometheus instances

Introduced in GitLab Ultimate 11.2 for custom metrics, and 11.3 for library metrics.

For managed Prometheus instances using auto configuration, alerts for metrics can be configured directly in the performance dashboard.

To set an alert:

Click on the ellipsis icon in the top right corner of the metric you want to create the alert for.
Choose Alerts
Set threshold and operator.
Click Add to save and activate the alert.

To remove the alert, click back on the alert icon for the desired metric, and click Delete.

External Prometheus instances

Introduced in GitLab Ultimate 11.8.

For manually configured Prometheus servers, a notify endpoint is provided to use with Prometheus webhooks. If you have manual configuration enabled, an Alerts section is added to Settings > Integrations > Prometheus. This contains the URL and Authorization Key. The Reset Key button will invalidate the key and generate a new one.

To send GitLab alert notifications, copy the URL and Authorization Key into the webhook_configs section of your Prometheus Alertmanager configuration:

receivers:
  name: gitlab
  webhook_configs:
  - http_config:
      bearer_token: 9e1cbfcd546896a9ea8be557caf13a76
    send_resolved: true
    url: http://192.168.178.31:3001/root/manual_prometheus/prometheus/alerts/notify.json
  ...

In order for GitLab to associate your alerts with an environment, you need to configure a gitlab_environment_name label on the alerts you set up in Prometheus. The value of this should match the name of your Environment in GitLab.

Taking action on incidents (ULTIMATE)

Introduced in GitLab Ultimate 11.11.

From GitLab Ultimate 12.5, when GitLab receives a recovery alert, it will automatically close the associated issue.

Alerts can be used to trigger actions, like open an issue automatically (enabled by default since 12.1). To configure the actions:

Navigate to your project's Settings > Operations > Incidents.
Enable the option to create issues.
Choose the issue template to create the issue from.
Optionally, select whether to send an email notification to the developers of the project.
Click Save changes.

Once enabled, an issue will be opened automatically when an alert is triggered which contains values extracted from alert's payload:

Issue author: GitLab Alert Bot
Issue title: Extract from annotations/title, annotations/summary or labels/alertname
Alert Summary: A list of properties
- starts_at: Alert start time via startsAt
- full_query: Alert query extracted from generatorURL
- Optional list of attached annotations extracted from annotations/*
Alert GFM: GitLab Flavored Markdown from annotations/gitlab_incident_markdown

When GitLab receives a Recovery Alert, it will automatically close the associated issue. This action will be recorded as a system message on the issue indicated that it was closed automatically by the GitLab Alert bot.

To further customize the issue, you can add labels, mentions, or any other supported quick action in the selected issue template, which will apply to all incidents. To limit quick actions or other information to only specific types of alerts, use the annotations/gitlab_incident_markdown field.

Since version 12.2, GitLab will tag each incident issue with the incident label automatically. If the label does not yet exist, it will be created automatically as well.

If the metric exceeds the threshold of the alert for over 5 minutes, an email will be sent to all Maintainers and Owners of the project.

Determining the performance impact of a merge

Introduced in GitLab 9.2.

GitLab 9.3 added the numeric comparison of the 30 minute averages.

Developers can view the performance impact of their changes within the merge request workflow.

NOTE: Note: Requires Kubernetes metrics.

When a source branch has been deployed to an environment, a sparkline and numeric comparison of the average memory consumption will appear. On the sparkline, a dot indicates when the current changes were deployed, with up to 30 minutes of performance data displayed before and after. The comparison shows the difference between the 30 minute average before and after the deployment. This information is updated after each commit has been deployed.

Once merged and the target branch has been redeployed, the metrics will switch to show the new environments this revision has been deployed to.

Performance data will be available for the duration it is persisted on the Prometheus server.

Embedding metric charts within GitLab Flavored Markdown

Embedding GitLab-managed Kubernetes metrics

Introduced in GitLab 12.2.

It is possible to display metrics charts within GitLab Flavored Markdown. The maximum number of embeds allowed in a GitLab Flavored Markdown field is 100.

NOTE: Note: Requires Kubernetes metrics.

To display a metric chart, include a link of the form https://<root_url>/<project>/-/environments/<environment_id>/metrics.

A single chart may also be embedded. You can generate a link to the chart via the dropdown located on the right side of the chart:

The following requirements must be met for the metric to unfurl:

The <environment_id> must correspond to a real environment.
Prometheus must be monitoring the environment.
The GitLab instance must be configured to receive data from the environment.
The user must be allowed access to the monitoring dashboard for the environment (Reporter or higher).
The dashboard must have data within the last 8 hours.

If all of the above are true, then the metric will unfurl as seen below:

Embedding metrics in issue templates

It is also possible to embed either the default dashboard metrics or individual metrics in issue templates. For charts to render side-by-side, links to the entire metrics dashboard or individual metrics should be separated by either a comma or a space.

Embedding Grafana charts

Grafana metrics can be embedded in GitLab Flavored Markdown.

Embedding charts via Grafana Rendered Images

It is possible to embed live Grafana charts in issues, as a direct linked rendered image.

The sharing dialog within Grafana provides the link, as highlighted below.

NOTE: Note: For this embed to display correctly the Grafana instance must be available to the target user, either as a public dashboard or on the same network.

Copy the link and add an image tag as inline HTML in your Markdown. You may tweak the query parameters as required. For instance, removing the &from= and &to= parameters will give you a live chart. Here is example markup for a live chart from GitLab's public dashboard:

<img src="https://dashboards.gitlab.com/render/d-solo/RZmbBr7mk/gitlab-triage?orgId=1&refresh=30s&var-env=gprd&var-environment=gprd&var-prometheus=prometheus-01-inf-gprd&var-prometheus_app=prometheus-app-01-inf-gprd&var-backend=All&var-type=All&var-stage=main&panelId=1247&width=1000&height=300"/>

This will render like so:

Embedding charts via integration with Grafana HTTP API

Introduced in GitLab 12.5.

Each project can support integration with one Grafana instance. This configuration allows a user to copy a link to a panel in Grafana, then paste it into a GitLab Markdown field. The chart will be rendered in the GitLab chart format.

Prerequisites for embedding from a Grafana instance:

The datasource must be a Prometheus instance.
The datasource must be proxyable, so the HTTP Access setting should be set to Server.

Setting up the Grafana integration

Generate an Admin-level API Token in Grafana.
In your GitLab project, navigate to Settings > Operations > Grafana Authentication.
To enable the integration, check the "Active" checkbox.
For "Grafana URL", enter the base URL of the Grafana instance.
For "API Token", enter the Admin API Token you just generated.
Click Save Changes.

Generating a link to a chart

In Grafana, navigate to the dashboard you wish to embed a panel from.
In the upper-left corner of the page, select a specific value for each variable required for the queries in the chart.
In Grafana, click on a panel's title, then click Share to open the panel's sharing dialog to the Link tab.
If your Prometheus queries use Grafana's custom template variables, ensure that "Template variables" and "Current time range" options are toggled to On. Of Grafana global template variables, only $__interval, $__from, and $__to are currently supported.
Click Copy to copy the URL to the clipboard.
In GitLab, paste the URL into a Markdown field and save. The chart will take a few moments to render.

Troubleshooting

If the "No data found" screen continues to appear, it could be due to:

No successful deployments have occurred to this environment.
Prometheus does not have performance data for this environment, or the metrics are not labeled correctly. To test this, connect to the Prometheus server and run a query, replacing $CI_ENVIRONMENT_SLUG with the name of your environment.
You may need to re-add the GitLab predefined common metrics. This can be done by running the import common metrics rake task.