fix: ps cmd should ignore dead runners #1050
Conversation
matt2e
requested review from
alecthomas,
dhanji,
wesbillman and
worstell
as code owners
|
Are these runners getting reaped eventually, and if so does the reaping correctly set the module_name to NULL eventually? If not it seems like we should fix that? |
backend/controller/sql/queries.sql
Outdated
| @@ -160,7 +160,7 @@ SELECT d.min_replicas, | |||
| r.endpoint, | |||
| r.labels AS runner_labels | |||
| FROM deployments d | |||
| LEFT JOIN runners r on d.id = r.deployment_id | |||
| LEFT JOIN runners r on d.id = r.deployment_id AND r.state = 'assigned' | |||
There was a problem hiding this comment.
Hmmm actually if we just want to ignore dead runners, this probably isn't the right solution. We probably do want != 'dead' so that it will also pick up runners that are reserved.
There was a problem hiding this comment.
I'm not 100% sure I'm right there, so it would be worth verifying if that does what we want...
There was a problem hiding this comment.
Makes sense, will fix
Looking into why it's not being set to null. |
|
It's the deregister runner code that is setting it to dead in the db without updating But I think the reap case will also end up those columns not nulled out... I'll set this PR back to draft and try recreating that case next week |
matt2e
force-pushed
the
matt2e/dead-runners-in-ps-cmd
branch
from
d6f9b32 to
ef7ba66
Compare
|
Handled the above cases and updated PR description |
matt2e
force-pushed
the
matt2e/dead-runners-in-ps-cmd
branch
from
ef7ba66 to
5cd2373
Compare
fixes #1046
Runners that were found to have died (rather than cleanly being killed) end up with deployment_id & module_name not set to null.
Changes:
pscommanddeployment_idandmodule_nameto null in the following cases: