after an unplanned promote the new primary refuses writes for 11.7 s while it holds the VIP #118

New issue

Closed

opened 2026-10-09 01:42:52 -03:00 by marcos-mendez · 0 comments

marcos-mendez commented

2026-10-09 01:42:52 -03:00

(Migrated from github.com)

Real nodes, keel 0.23.7 (N2). keel database follow turns read_only off 7.5 s after keel vip promote --old-primary-gone ends. The application gets 70 times ERROR 1290 ... --read-only option between 04:25:06.1 and 04:25:16.9. Unplanned write downtime is 42.2 s, against 35.5 s on 0.23.3. Together with keel#108 (promote takes 30.5 s) this is most of the downtime. Follow must start when the VIP lands on wg0 (the path unit or a direct call from promote) and not wait for a poll.

Real nodes, keel 0.23.7 (N2). `keel database follow` turns read_only off 7.5 s after `keel vip promote --old-primary-gone` ends. The application gets 70 times `ERROR 1290 ... --read-only option` between 04:25:06.1 and 04:25:16.9. Unplanned write downtime is 42.2 s, against 35.5 s on 0.23.3. Together with keel#108 (promote takes 30.5 s) this is most of the downtime. Follow must start when the VIP lands on wg0 (the path unit or a direct call from promote) and not wait for a poll.