Wrong Container ID
Hi, Olivier.
Have found situation, that jobs that was started one by one and ended up on the same computing node have the same container id. So we can't match it properly, to get job metrics from cadvisor.
Comments (12)
-
reporter -
repo owner Will have a look but i don't see how container id can be the same. You use docker containerizer or unified containerizer?
-
repo owner Did you looked at cadvisor web ui directly to see running containers?
-
reporter it is unified containerizer
-
reporter Did you looked at cadvisor web ui directly to see running containers?
yes
It's 5 jobs running on the same node
container.id:"dc379355-0d68-49c2-9736-8dee4984ed2a" correct for godocker job id "id" : 4354,
but doesn't correct for godocker job id "id" : 4380
and godocker job id "id" : 4380 have correct container id 22527b04-5623-424e-b775-4017125760db on mesos (got it from mesos UI) but this one even not found in mongodb
db.jobs.find({"container.id":"22527b04-5623-424e-b775-4017125760db"}).pretty()
-
repo owner are you "replaying" jobs ? or each job is created independently ?
ifnot yet defined, container id is fetched from mesos slave (godocker ask slave the container id for the mesos task). In case of job replay, I just wonder if older id could have been kept (which would be a bug)
-
reporter are you "replaying" jobs ? or each job is created independently ?
I think independently
-
repo owner Which mesos version are you running?
-
reporter v. 1.5.0
-
repo owner - changed status to closed
fix get of container id for unified containerizer with mesos, Closes
#74→ <<cset 027f8f4c1a2c>>
-
repo owner Hi, found the issue, a strange API behavior with mesos which returns all containers on slave even if querying a specific executor. Anyway, I pushed to develop the fix. Already running jobs won't be fixed (will keep invalid container id), this will apply only to new jobs.
-
reporter Lot of thanks!
- Log in to comment