Open cevich opened 2 years ago
A friendly reminder that this issue had no activity for 30 days.
@vrothberg @mheon WDYT?
A use case would help. In theory it sounds independent of health checks but more related to the restart logic. If a container gets restarted manually by the user or by an on-failure action should not matter.
A use case could be:
A failing health-check on container C could easily be the fault of B or A. Since it uses --requires
, it "knows" to restart A and B due to a failing health check (assuming their own health-checks didn't fire first). Whereas without --requires
, it would simply endlessly/uselessly restart itself until B health-check failures corrected the fault (which could be never).
A friendly reminder that this issue had no activity for 30 days.
@vrothberg followup on this one?
A use case would help. In theory it sounds independent of health checks but more related to the restart logic. If a container gets restarted manually by the user or by an on-failure action should not matter.
I still think that's more a restart thing than a health check one. The health-check action just restarts the container.
Yes, definitely a restart thing, not health-check.
@mheon WDYT about restarting dependent containers?
Code-wise I see some challenges regarding batch restarts where we need to prevent a container from being restarted multiple times.
Could use the graph traversal code in container_graph.go
- guaranteed to go in the correct order and only hit each node once, and because it was originally written for podman pod restart
it already handles pods.
Sorry, it already handles restarting not pods. Need more coffee...
A friendly reminder that this issue had no activity for 30 days.
@mheon @vrothberg @cevich What should we do with this one?
Either close it or leave it open until someone does the work. As mentioned above, it's unrelated to health-check actions but rather a restart-logic feature.
Playing my own devil's advocate here - We could close it with an excuse like: If anyone needs this kind of fancy-stuff, they should use kubernetes not podman.
/kind feature
Description
When a container which depends on another (via
--requires
) fails it's healthcheck with a restart policy, it should honor a single or a chain of--requires
container restarts as well.Steps to reproduce the issue:
Run a container which has a dependency on another. Either or both should have healthchecks enabled with restart policies.
Cause a healthcheck to fail
Describe the results you received:
The single or chain of dependent containers does not restart.
Describe the results you expected:
The single or chain of dependent containers does restart.
Additional information you deem important (e.g. issue happens only occasionally):
If there are a lot of dependencies, or the relationships are complex, perhaps we should provide an opt-out flag.