GitLab Prometheus metrics
Note: Available since Omnibus GitLab 9.3. For installations from source you'll have to configure it yourself.
To enable the GitLab Prometheus metrics:
- Log into GitLab as an administrator, and go to the Admin area.
- Click on the gear, then click on Settings.
- Find the
Metrics - Prometheussection, and clickEnable Prometheus Metrics - Restart GitLab for the changes to take effect
Collecting the metrics
GitLab monitors its own internal service metrics, and makes them available at the
/-/metrics endpoint. Unlike other Prometheus exporters, in order to access
it, the client IP needs to be included in a whitelist.
For Omnibus and Chart installations, these metrics are automatically enabled and collected as of GitLab 9.4. For source installations or earlier versions, these metrics will need to be enabled manually and collected by a Prometheus server.
Unicorn Metrics available
The following metrics are available:
| Metric | Type | Since | Description |
|---|---|---|---|
| db_ping_timeout | Gauge | 9.4 | Whether or not the last database ping timed out |
| db_ping_success | Gauge | 9.4 | Whether or not the last database ping succeeded |
| db_ping_latency_seconds | Gauge | 9.4 | Round trip time of the database ping |
| filesystem_access_latency_seconds | Gauge | 9.4 | Latency in accessing a specific filesystem |
| filesystem_accessible | Gauge | 9.4 | Whether or not a specific filesystem is accessible |
| filesystem_write_latency_seconds | Gauge | 9.4 | Write latency of a specific filesystem |
| filesystem_writable | Gauge | 9.4 | Whether or not the filesystem is writable |
| filesystem_read_latency_seconds | Gauge | 9.4 | Read latency of a specific filesystem |
| filesystem_readable | Gauge | 9.4 | Whether or not the filesystem is readable |
| http_requests_total | Counter | 9.4 | Rack request count |
| http_request_duration_seconds | Histogram | 9.4 | HTTP response time from rack middleware |
| pipelines_created_total | Counter | 9.4 | Counter of pipelines created |
| rack_uncaught_errors_total | Counter | 9.4 | Rack connections handling uncaught errors count |
| redis_ping_timeout | Gauge | 9.4 | Whether or not the last redis ping timed out |
| redis_ping_success | Gauge | 9.4 | Whether or not the last redis ping succeeded |
| redis_ping_latency_seconds | Gauge | 9.4 | Round trip time of the redis ping |
| user_session_logins_total | Counter | 9.4 | Counter of how many users have logged in |
| upload_file_does_not_exist | Counter | 10.7 in EE, 11.5 in CE | Number of times an upload record could not find its file |
| failed_login_captcha_total | Gauge | 11.0 | Counter of failed CAPTCHA attempts during login |
| successful_login_captcha_total | Gauge | 11.0 | Counter of successful CAPTCHA attempts during login |
| unicorn_active_connections | Gauge | 11.0 | The number of active Unicorn connections (workers) |
| unicorn_queued_connections | Gauge | 11.0 | The number of queued Unicorn connections |
Ruby metrics
Some basic Ruby runtime metrics are available:
| Metric | Type | Since | Description |
|---|---|---|---|
| ruby_gc_duration_seconds_total | Counter | 11.1 | Time spent by Ruby in GC |
| ruby_gc_stat_... | Gauge | 11.1 | Various metrics from GC.stat |
| ruby_file_descriptors | Gauge | 11.1 | File descriptors per process |
| ruby_memory_bytes | Gauge | 11.1 | Memory usage by process |
| ruby_sampler_duration_seconds_total | Counter | 11.1 | Time spent collecting stats |
Metrics shared directory
GitLab's Prometheus client requires a directory to store metrics data shared between multi-process services. Those files are shared among all instances running under Unicorn server. The directory needs to be accessible to all running Unicorn's processes otherwise metrics will not function correctly.
For best performance its advisable that this directory will be located in tmpfs.
Its location is configured using environment variable prometheus_multiproc_dir.
If GitLab is installed using Omnibus and tmpfs is available then metrics
directory will be automatically configured.