Overview
Monitoring reports what your API does in production: which requests fail, which are slow, what changed in the release that broke them, and which commit is the likely cause.
Turning Monitoring on for a project bakes a connection into future generated SDKs. Rebuild and publish, then each developer using the SDK enables reporting in their application; it ships switched off. There is no separate agent to deploy.
The tabs
Every event, newest first. Where you land when something just broke.
Errors grouped into problems, with triage, comments, and suspect commits.
Follow one request across services and see where the time went.
Latency percentiles per endpoint, so a slow tail is visible.
Health per deploy, and what likely caused a regression.
Scheduled checks that call your endpoints and tell you they're down.
Get told about error spikes, new issues and regressions.
The query bar and filter builder every tab shares.
Getting started
- Enable it
Enable Monitoring on the project. This prepares the connection for the next SDK build. See Setup.
- Rebuild your SDKs
The telemetry config is compiled in at build time, so already-published versions do not gain a connection.
- SDK user opts in
The developer using the rebuilt SDK enables
logging.enabledin their application. Browser TypeScript and Python SDKs attach personal context only aftersetLoggingConsent(true). - Upload symbols
Otherwise production stack traces are unreadable. See Readable stack traces.
The overview panel
Beside the tabs, Overview summarizes the selected window: total events, error rate, distinct issues, p95 latency, latency percentiles and top issues.
Two sources of data
| Source | Reports from | Read about it |
|---|---|---|
| Generated SDKs | Your customers' applications | Setup |
| Backend runtimes | Your own servers, or any application | Backend SDKs |
When both run, a failed request shows up as one trace, from the client call to the server code that threw.
Live by default
Events stream to the dashboard as they arrive, so an incident updates in front of you without a refresh.
Credentials are scrubbed on the way out
Every runtime redacts credential-shaped keys and sweeps free text for tokens, JWTs, card numbers and email addresses before an event is sent. It's on by default. See Redaction.
What's kept
Raw events are kept 7 days on Free, 30 on Starter, 90 on Growth, 180 on Business and 365 on Enterprise by default (or the window in your Enterprise order form). Hourly rollups support longer-range charts after individual events expire.
Your plan includes a monitoring storage allowance; usage against it is shown in the dashboard.
Further reading
Background rather than instructions, on the blog.