nsfs | wait for endpoint startup before namespace monitor registration #8474
Conversation
alphaprinz
force-pushed
the
2284585_nsfs_nsr_rejected
branch
3 times, most recently
from
02e747f to
b4ee9d0
Compare
| client: internal_rpc_client, | ||
| should_monitor: nsr => Boolean(nsr.nsfs_config), | ||
| })); | ||
| setTimeout(() => { |
There was a problem hiding this comment.
why not add a retry in the namespace monitor if the error was ENOENT and start time of the endpoint is lower than 60 seconds? also why 60 seconds?
There was a problem hiding this comment.
Sounds like more code with equivalent solution.
60 seconds to allow pod to stabilize (get to 'ready' or be deleted).
There was a problem hiding this comment.
the reason is understood, but how do you know it takes a minute? is it always a minute? should it be configurable?
There was a problem hiding this comment.
AFAIK, there's no way for node to know status of the pod.
I assume we don't want to make such a dependency.
60 seconds is more than enough on my minikube and is low enough not to bother other deployments in other envs, but I can make it an env variable.
Signed-off-by: Amit Prinz Setter <alphaprinz@gmail.com>
alphaprinz
force-pushed
the
2284585_nsfs_nsr_rejected
branch
from
b4ee9d0 to
7ec7799
Compare
Explain the changes
Wait for endpoint startup before registering namespace resource monitor.
Issues: Fixed #xxx / Gap #xxx
Nsr can enter "Rejected" status if endpoint is deleted by kubernetes before it is in "Ready" state.
https://bugzilla.redhat.com/show_bug.cgi?id=2284585.
Testing Instructions:
Repoduction
Sagie details scenario for ODS in bz.
I've reproduced with this scenario on minikube:
Start with a nsfs nsr on a pvc. A single endpoint A is in "Ready" state.
-Delete nsr. A new endpoint B is being spun.
-While endpoint B is done creating but NOT ready yet (endpoint A is still in "Ready" state), create nsr.
-Kubernetes will delete endpoint B and will leave endpoint A running.
-Endpoint B loads nsr from system store, but nsr is not mounted on it. Endpoint B issues NO_SUCH_BUCKET report on nsr (and then it is deleted by kubernetes).
local_nsfs.yaml.txt