Marathon: TCP/HTTP health checks not resistant to network partition
Current implementation of health checks does not perform optimally during a network partition. It marks task to be killed and once the network partition heals it kills all of the tasks on the far side of the partition. Can Marathon use TASK_LOST status from Mesos to pause health checks?
medzin
All 5 comments
Marathon treats a TASK_LOST update as a task failure and will launch a new the task, so it wouldn't make sense to pause health checks.
The proper fix would be to add support for Mesos HTTP/TCP health checks once https://issues.apache.org/jira/browse/MESOS-2533 is merged.
gkleiman
on 11 Apr 2016
Fixed in v0.15.x by e8ace92168573927326f64963b73bb6df88dbdd3, we still have to port this fix to 1.x
cc/ @aquamatthias
gkleiman
on 6 Jun 2016
@gkleiman In a perfect world, it seems like this and this are actually better. Doing it this way, would allow not having to have these sort of hacks in marathon proper, and everyone is better off!
What do you think?
SEJeff
on 6 Jun 2016
Support for Mesos health checks is in marathon now.
jasongilanfarr
on 14 Nov 2016
What version will this be in?
SEJeff
on 14 Nov 2016
Related issues
pgkelley4
路
11Comments
sybrandy
路
13Comments
a-nldisr
路
16Comments
balopat
路
5Comments
tobilg
路
9Comments
Most helpful comment
@gkleiman In a perfect world, it seems like this and this are actually better. Doing it this way, would allow not having to have these sort of hacks in marathon proper, and everyone is better off!
What do you think?