Repository navigation
Queen deploy drain almost stops dispatch for ~30 min (rounds gated during drain) #1705
Description
Why
Measured 2026-10-08 during two Queen deploys. While the old instance drains (about 30 min, because the volume holds the new container back), dispatch nearly stops:
- active bees went 46 -> 31 in the first deploy, and 39 -> 22 in the second;
- in 12 minutes of the second drain, the logs show one
Queen tick decided. A round normally runs every 60 s.
Memory note queen-deploy-does-not-stop-swarm says rounds should keep going through the drain. Something gates them. Every Queen deploy costs about half an hour of dispatch.
Acceptance criteria
- The cause is named from logs or code: which gate stops rounds while the instance drains.
- During a deploy, rounds keep deciding at their normal rate. This is measured by counting
Queen tick decidedper minute over the drain of the next deploy.
Activity
gHashTag commentedon Oct 8, 2026
Correction: the drain is probably not the cause. The logs say:
What changed at 19:00Z is the actor reviewer. Without backoff it visited the 22 rows whose branch cannot be read about 184 times a minute. Each visit records a
waitverdict and publishes events, so the round's queries and its report queue behind that traffic.The wait backoff (#1702, deploying as 6f310933) removes that traffic. I will keep this open until the next measurement shows rounds back at ~30 s, then close it or name the real gate.
chore(queen): where a round's time goes - 'Queen round phases' and 'Q…
chore(queen): where a round's time goes - 'Queen round phases' and 'Q…