The restart test, for any Deployment behind the webmail #39

Merged
jcoffey-dev merged 4 commits from feat/restart-any into main 2026-11-20 18:19:20 +00:00
Owner

lab/chart/webmail-restart-test.sh gets RESTART=, the Deployments to restart under the open tabs, one after another. The default is still inbuxa-webmail.

  • The rule that every live-update stream must end and come back now applies only when the webmail itself restarts.
  • Behind the webmail, the streams just stay open. There it checks that no request failed and no stream was cut off.

Why: K-L10 (#38) saw one webmail request time out in 2 of 4 upgrades, with everything rolling at once. A probe on this branch, since reverted, restarted each Deployment three times under the tabs (webmail rc.10, server rc.34, console rc.22):

Restarted Requests failed
inbuxa-front, inbuxa-core, inbuxa-mta 0 in 9 restarts (~200–230 requests each)
inbuxa-console 0 in 3
inbuxa-webmail 1, 2, 3 (timeouts and refused connections)

So the timeout comes from the webmail, and it's the known one: rc.10 predates the graceful drain (K-W5, webmail #132, merged). It goes away with the next webmail RC in release/suite.yaml. The server's groups lose no requests when they restart.

`lab/chart/webmail-restart-test.sh` gets `RESTART=`, the Deployments to restart under the open tabs, one after another. The default is still `inbuxa-webmail`. - The rule that every live-update stream must end and come back now applies only when the webmail itself restarts. - Behind the webmail, the streams just stay open. There it checks that no request failed and no stream was cut off. **Why:** K-L10 (#38) saw one webmail request time out in 2 of 4 upgrades, with everything rolling at once. A probe on this branch, since reverted, restarted each Deployment three times under the tabs (webmail rc.10, server rc.34, console rc.22): | Restarted | Requests failed | |---|---| | `inbuxa-front`, `inbuxa-core`, `inbuxa-mta` | 0 in 9 restarts (~200–230 requests each) | | `inbuxa-console` | 0 in 3 | | `inbuxa-webmail` | 1, 2, 3 (timeouts and refused connections) | So the timeout comes from the webmail, and it's the known one: rc.10 predates the graceful drain (K-W5, webmail #132, merged). It goes away with the next webmail RC in `release/suite.yaml`. The server's groups lose no requests when they restart.
jcoffey-dev added 4 commits 2026-11-20 18:18:53 +00:00
jcoffey-dev merged commit 5d5ff0ab41 into main 2026-11-20 18:19:20 +00:00
jcoffey-dev deleted branch feat/restart-any 2026-11-20 18:19:20 +00:00
Sign in to join this conversation.