CNI daemon workers indefinitely waiting for locks
Bug #1856107 reported by
Yash Gupta
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
kuryr-kubernetes |
In Progress
|
Undecided
|
Yash Gupta |
Bug Description
Rarely, a CNI worker is not able to acquire a lock (For updating/reading the pod registry), and keeps waiting for the lock.
However, the CNI runner times out and kubelet sends more requests to the CNI. These requests spawn more workers which are waiting on the same lock.
This soon results in growing number of workers waiting for the same lock, and exhausting kuryr-cni pod resources (workers, memory, etc)
However, it could not be determined why the deadlock happened in the first place.
Changed in kuryr-kubernetes: | |
assignee: | nobody → Yash Gupta (yashg2) |
status: | New → In Progress |
To post a comment you must log in.