Closed benley closed 9 years ago
Thanks for reporting, will try to repro this.
The reason might be spray's retries. I couldn't find where marathon configures spray's HostConnector, though. Even after user deletes some urls from subscribers list, post
future could retry to deleted url.
Tried to reproduce this at some length, but we were unable to make it happen on our test cluster. Are you still seeing this problem?
@benley
This may be related to a leadership issue and has probably been fixed with this patch: https://github.com/mesosphere/marathon/pull/953. Do you still see this with current master?
Haven't had a chance to test yet - will get back to you asap.
@benley Any news on this?
Closing this for now, please comment if this comes up again.
Oops, forgot to follow up with this. I am not seeing this issue anymore with marathon 0.7.6, so closing this is good.
Thanks!
This morning I noticed that I had some stale hosts registered in eventSubscriptions, so I DELETEd them:
Output snippet:
After unsubscribing the stale URLs, they were gone from the list:
But in syslog 15 and 30 minutes later, marathon was still printing errors about trying to connect to the dead hosts:
I'm running marathon 0.7.3 in HA mode. There is a haproxy instance in front of marathon's http interface, so sequential http requests may not be sent to the same marathon host - though the non-master instances appear to successfully proxy those requests to the master.
Marathon commandline:
After restarting the master marathon instance, one of the standby instances became master and the stale notifications appear to have stopped.