During volume detach, the following might happen in reconciler
1. Pod is deleting
2. remove volume from reportedAsAttached, so node status updater will
update volumeAttached list
3. detach failed due to some issue
4. volume is added back in reportedAsAttached
5. reconciler loops again the volume, remove volume from
reportedAsAttached
6. detach will not be trigged because exponential back off, detach call
will fail with exponential backoff error
7. another pod is added which using the same volume on the same node
8. reconciler loops and it will NOT try to tigger detach anymore
At this point, volume is still attached and in actual state, but
volumeAttached list in node status does not has this volume anymore, and
will block volume mount from kubelet.
The fix in first round is to add volume back into the volume list that
need to reported as attached at step 6 when detach call failed with
error (exponentical backoff). However this might has some performance
issue if detach fail for a while. During this time, volume will be keep
removing/adding back to node status which will cause a surge of API
calls.
So we changed to logic to check first whether operation is safe to retry which
means no pending operation or it is not in exponentical backoff time
period before calling detach. This way we can avoid keep removing/adding
volume from node status.
Change-Id: I5d4e760c880d72937d34b9d3e904ecad125f802e
It is possible that by the time we check for multiattach
error on node, the reconciler loop may not have processed second
volume and hence we are going to retry for multiattach error
on node before giving up and marking the test as failed.
Only relying on the NewAttacher/Detacher call counts is not enough as they
happen in parallel to the testing/verification code and thus the actual
attaching/detaching may not be done yet, resulting in flaky test results.
Fixes#46244
When the attach/detach controller crashes and a pod with attached PV is deleted
afterwards the controller will never detach the pod's attached volumes. To
prevent this the controller should try to recover the state from the nodes
status.
At master volume reconciler, the information about which volumes are
attached to nodes is cached in actual state of world. However, this
information might be out of date in case that node is terminated (volume
is detached automatically). In this situation, reconciler assume volume
is still attached and will not issue attach operation when node comes
back. Pods created on those nodes will fail to mount.
This PR adds the logic to periodically sync up the truth for attached volumes kept in the actual state cache. If the volume is no longer attached to the node, the actual state will be updated to reflect the truth. In turn, reconciler will take actions if needed.
To avoid issuing many concurrent operations on cloud provider, this PR
tries to add batch operation to check whether a list of volumes are
attached to the node instead of one request per volume.
More details are explained in PR #33760
We had another bug where we confused the hostname with the NodeName.
To avoid this happening again, and to make the code more
self-documenting, we use types.NodeName (a typedef alias for string)
whenever we are referring to the Node.Name.
A tedious but mechanical commit therefore, to change all uses of the
node name to use types.NodeName
Also clean up some of the (many) places where the NodeName is referred
to as a hostname (not true on AWS), or an instanceID (not true on GCE),
etc.
