[bashgeek]

Summary

Last prep before tagging a release.

• It fixes the flaky tests that failed the v1.1.3 and v1.2.0 release runs.
• It adds the cheap database upkeep items from the large-install review.
• It gives Postgres enough shared memory in the shipped compose files.

Type of change

• Bug fix
• New feature
• Refactor / cleanup
• Docs / config only

Changes

Flaky tests

• Fixed sleeps replaced: 13 fixed 300 to 500 ms sleeps in the ingest, transaction and flamegraph tests, and 2 in internal/ingest/txbuffer_test.go, raced the 200 ms ingest flush. They now use waitPersisted, which waits until the buffer reports the rows committed, with a 10 s deadline.
• "Nothing stored" tests no longer wait: the handler drops session-only envelopes, malformed profiles and profiles from projects with profiling off before anything reaches a buffer. Those tests now assert on the buffer's counters instead of sleeping.
• Makefile: COMPOSE ran docker compose version on every make call. That slowed the Makefile tests on CI runners until they hit bun's 5 s default timeout. The variable is now deferred and kept out of the exported environment, so only the db* targets call Docker.
• Bun Makefile tests: they get a 30 s timeout, and a new test puts a stub docker on PATH to check that migrate-create, build, lint and a mistyped target never call it.

Migration 0038 (no table rewrites; every statement can safely run again)

• Drops logs_attributes. No query uses a JSONB operator on log attributes, and its GIN entries were a large share of log insert time and WAL.
• Adds issues_regressed_at for the alert check that runs every 15 s.
• Vacuums and analyzes events, event_tags, transactions, spans, logs and issues at 2% change instead of 20%.

Compose

• shm_size: 1g for Postgres in docker-compose.yml and in the compose file install.sh writes. Docker's default /dev/shm of 64 MB is too small for parallel queries.

Left out on purpose: I also tried replacing the usage_update statement triggers with a row trigger limited to project_id and received_at, so the grouper's per-event update would never fire them. A single UPDATE that moves many rows then fires the row trigger once per row. Each firing updates the same counter row again in one transaction, so the cost grows quadratically; a retention test that moves 205k rows hung. The right fix for that overhead is batching the grouper.

Test plan

• Flaky tests: the affected API and ingest tests passed three times, each in a fresh process.
• Migration 0038: TestMigration0038IsIdempotent re-runs the whole migration. It checks that logs_attributes is gone and issues_regressed_at exists, that the planner uses the new index for regressed_at > $1, and that the autovacuum settings are on all six tables.
• Makefile: make -n db db-stop db-logs still prints the compose commands.
• Compose: the new installer and compose-file tests check shm_size, and docker compose config accepts the file.
• Full suite: go vet, golangci-lint (0 issues), go test -race ./... (all 17 packages) and the four bun script test files (11 tests) pass.

Checklist

• Tests added or updated
• No new dependencies introduced (or justified in the PR description)
• I assign copyright in this contribution to Blendbyte GmbH, as required by the
  Contributor License Agreement

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!