Open robb-j opened 1 year ago
Seeing similar here. The pod itself isn't being actively used although memory seems to leak until it's OOM killed.
Its a shame there are not default values included for the resources requests/limits in the helm chart.
hrm. has this been resolved with newer versions? I set up a goldilocks monitor for it yesterday and am getting this as a resources recommendation:
resources:
requests:
cpu: 163m
memory: 2282M
limits:
cpu: 163m
memory: 2282M
The last couple months: (I recently inherited this cluster)
The variations in the graph are indicative of this sort of status message w/in the dead pod's skeleton:
Message: The node was low on resource: memory. Container connect-api was using 1576284Ki, which exceeds its request of 0. Container connect-sync was using 831820Ki, which exceeds its request of 0.
[...]
connect-sync:
Container ID:
Image: 1password/connect-sync:1.5.6
Image ID:
Port: <none>
Host Port: <none>
State: Terminated
Reason: ContainerStatusUnknown
Message: The container could not be located when the pod was terminated
Exit Code: 137
[...]
👍 , same here. My pod reaches 1GB in two days.
And finally:
The node was low on resource: memory. Threshold quantity: 100Mi, available: 85788Ki. Container connect-api was using 1671000Ki, request is 0, has larger consumption of memory. Container connect-sync was using 926732Ki, request is 0, has larger consumption of memory.
Bueller...
Bueller...
(sigh)
Dear 1password:
Hello. You might remember me from January 4th of this year. (I think I tripped over @robb-j's corpse in the parking lot, but, no matter, I'm here now). This isn't an open-source project in so much as anyone using it is a paying customer or works for a paying customer. If this operator should be considered abandoned, please have the courtesy of letting us know so we can make other arrangements.
Thanks
Second the motion.
Hey @gladiatr72!
Samira here from 1Password 👋🏻 Thanks for bringing this to our attention. We do keep a close eye on all of our repos and this issue in particular hasn't been forgotten. Memory issues tend to be tricky and we've been working on our end to try to figure this one out. Apologies on not keeping the issue updated, but it has not been forgotten.
To alleviate your concerns about this repo being abandoned, I want to stress that this is not the case. We are actively working on it and in fact, we released a new version last month that addressed a few bugs. Here are the release notes on that particular release.
We'll continue making updates to this repo over time and will keep the issue updated with our progress on this memory issue.
Thanks!
I'm starting to feel a bit neglected again...
@edif2008 what's the word?
Hey @gladiatr72, are you familiar with this GitHub issue? There have been a few developments there:
container_memory_usage_bytes
metric that includes memory that is used for "caching" and can be reclaimed by the OS. The assumption is that container_memory_working_set_bytes
is harmless as long as container_memory_rss
does not increase as well. Applying a memory limit for Connect in k8s seems to force the OS to clear and reclaim the memory used for caching which seems to bound the memory consumption. To quickly validate this hypothesis it is possible to use sudo sh -c "sync; echo 3 > /proc/sys/vm/drop_caches"
on the k8s host.Let me know if any of the above is helpful in your case!
Your environment
Operator Version: 1.5.7
Connect Server Version: 1.5.7?
Kubernetes Version: v1.23.14
What happened?
The memory usage of my
onepassword-connect
pod was 1097Mi which seems very high, is there a normal amount of memory this container should stay around?What did you expect to happen?
I thought the container would need a lot less memory
Steps to reproduce
Mi
every few minuteNotes & Logs
before restarting
after restarting
my
values.yml
passed to helm: