Logs

The logs sensor reads what a running system wrote about itself, during a test run or after a release, and reports the errors in it, grouped and counted. Loops read it because many failures never reach an exit code: an error is caught and logged, and the run carries on.

What it sees

The errors the code logged, in the runs and releases a query covers. It needs structured logs and a store to query them in. Locally, a store per worktree keeps one change's logs apart from another's; after a release, the version on every line lets the same query compare the new version with the previous one. It cannot see a failure that was never logged, nor a code path that did not run in the period it read.

Reading the signal

• None since the change: no new error appeared in the version and period the query covered.
• A new error: a group that did not occur before the change, with its source and count. The change has to account for it.
• A known error whose count grew: an existing problem the change made more frequent. Compare counts over the same period and the same traffic.

An empty result is not a clean log when the query found no lines at all: the store was unreachable, the run wrote nowhere, or the version was missing from the lines. Check that the query matched the run's own lines before reading silence as a pass.