Pull Request Checklist

• Have you added new tests to prevent regressions? (N/A - dev tooling)
• Does yarn test or yarn test-DIALECT pass with this change (including linting)?
• Is a documentation update included (if this change modifies existing APIs, or introduces new ones)? (N/A)
• Did you follow the commit message conventions explained in CONTRIBUTING.md?

Description Of Change

Since #18291 landed, db2 oldest and db2 latest are the most frequent database start failures in CI: 15 failed yarn start-db2-* steps across 9 runs, all with

Timeout of 150s exceeded when waiting for container to be healthy: sequelize-db2-oldest

Examples: db2 oldest (Node 22), db2 latest (Node 24).

Cause

Db2 creates the database on first start. The start step on successful main runs takes 127-213s, and locally on a fast machine the container first reports healthy after 175s. The Db2-only copy of wait-until-healthy.sh polls for 10 × 15s = 150s, which is right at the edge of a normal startup, so a slightly slow runner fails.

That copy also predates #18291: it still has the dead return_code check, prints nothing about the container on failure (which is why the job logs above say nothing useful), and sleeps a blind 15s after the container is healthy.

There is a second, latent problem. The image's setup script restarts the instance twice right after CREATE DATABASE:

21:45:21  SQL1026N  The database manager is already active.   <- db created, connectable
21:45:27  SQL1064N  DB2STOP processing was successful.
21:45:28  SQL1063N  DB2START processing was successful.
21:45:34  SQL1064N  DB2STOP processing was successful.
21:45:36  SQL1063N  DB2START processing was successful.
21:45:36  (*) Setup has completed.

A probe landing in one of those ~6s windows marks the container healthy, after which the instance goes down again and check-connection.ts gets ECONNREFUSED. The old sleep 15 was presumably guarding against this.

Changes

• Delete dev/db2/wait-until-healthy.sh; both start.sh use the shared dev/wait-until-healthy.sh with HEALTHCHECK_TIMEOUT=600, so a slow runner has headroom while a genuinely stuck container still fails in minutes rather than hours, and the health probes + container logs are dumped on failure.
• The healthcheck additionally requires /database/config/.shared-data/setup_complete, the marker the image's setup_db2_instance.sh touches right after the last restart (same path in 11.5.5.1 and 12.1.4.0). This replaces the blind sleep.
• start_period: 300s so the container is not reported unhealthy while it is legitimately still creating the database.

Verification (local, icr.io/db2_community/db2:11.5.5.1)

Note: touches the same start.sh files as #18340, expect a trivial conflict in whichever lands last.

🤖 Generated with Claude Code