feat: use client-go leader election #1128
Conversation
adityathebe
force-pushed
the
feat/leader-election
branch
2 times, most recently
from
6ffe025
to
0942f93
Compare
leader/election.go
Outdated
| AcquireTime: &now, | ||
| RenewTime: &now, | ||
| OnStoppedLeading: func() { | ||
| updateLeaderLabel(ctx, false) |
There was a problem hiding this comment.
I don't think we can rely on this to add a label - if the pod is killed or partitions it will never run - The leader should first add its own label, and then remove the label from other others
There was a problem hiding this comment.
I think we should also stop and start all schedules here
adityathebe
force-pushed
the
feat/leader-election
branch
from
a6d8014
to
a3d9785
Compare
leader/election.go
Outdated
| app string, | ||
| namespace string, | ||
| service string, |
There was a problem hiding this comment.
Why do we need both app and service ?
There was a problem hiding this comment.
app would always be "config-db" which is hardcoded in flanksource/config-db whereas service was meant to be the actual kubernetes service name.
Although, the we could just use the app name as the lease name.
adityathebe
force-pushed
the
feat/leader-election
branch
from
31e08e1
to
5ff4d64
Compare
leader/election.go
Outdated
| AcquireTime: &now, | ||
| RenewTime: &now, | ||
| OnStoppedLeading: func() { | ||
| for _, k := range echo.Crons.Items() { |
There was a problem hiding this comment.
don't we also need to start the crons again when we become leader ?
There was a problem hiding this comment.
starting the jobs had a few pre-processing before the cron was started.
So the operator in config-db would start those jobs in the onLead callback.
Although upon closer inspection, I see that's a one time thing.
adityathebe
force-pushed
the
feat/leader-election
branch
from
24f0782
to
ded2120
Compare
moshloop
force-pushed
the
feat/leader-election
branch
from
b980ffe
to
7d27424
Compare
adityathebe
force-pushed
the
feat/leader-election
branch
from
7d27424
to
9c45330
Compare
No description provided.