Deadlock of the pool when destroying a replica

[tiagolobocastro]

Describe the bug Pool lock was taken and never released. This means all grpc for that pool will fail!

To Reproduce Seems like this may happen if we try to delete a replica which is part of a nexus!

Expected behavior Don't lock the pool forever...

Additional context

2024-09-12T06:54:14.717272539+02:00 stdout F [2024-09-12T04:54:14.717133800+00:00  INFO io_engine::grpc::v1::replica:replica.rs:402] DestroyReplicaRequest { uuid: "4711b421-0210-4db5-b88f-c2c55cac52da", pool: Some(PoolName("alex-cloud-sn-2-pool")) }
2024-09-12T06:54:14.719461813+02:00 stdout F [2024-09-12T04:54:14.719375762+00:00  INFO io_engine::lvs::lvs_lvol:lvs_lvol.rs:247] Lvol 'alex-cloud-sn-2-pool/3a6b6004-dbc8-4613-b316-f1f35fce24e0/4711b421-0210-4db5-b88f-c2c55cac52da' [50.00 GiB]: unshared
2024-09-12T06:54:14.719767776+02:00 stdout F [2024-09-12T04:54:14.719701943+00:00  INFO io_engine::bdev::device:device.rs:785] Received SPDK remove event for bdev '4711b421-0210-4db5-b88f-c2c55cac52da'
2024-09-12T06:54:14.719783986+02:00 stdout F [2024-09-12T04:54:14.719730236+00:00  INFO io_engine::bdev::nexus::nexus_bdev_children:nexus_bdev_children.rs:899] Unplugging nexus child device nexus_name="331c0652-0a75-4a2c-8946-3caa0590af06" child_device="4711b421-0210-4db5-b88f-c2c55cac52da"
2024-09-12T06:54:14.719851202+02:00 stdout F [2024-09-12T04:54:14.719744462+00:00  INFO io_engine::bdev::nexus::nexus_child:nexus_child.rs:1113] Child 'bdev:///4711b421-0210-4db5-b88f-c2c55cac52da?uuid=4711b421-0210-4db5-b88f-c2c55cac52da @ 331c0652-0a75-4a2c-8946-3caa0590af06' [open synced]: unplugging child...
2024-09-12T06:54:14.720345689+02:00 stdout F [2024-09-12T04:54:14.719979192+00:00  INFO io_engine::bdev::nexus::nexus_bdev:nexus_bdev.rs:657] Nexus '331c0652-0a75-4a2c-8946-3caa0590af06' [open]: dynamic reconfiguration event: unplug, reconfiguring I/O channels...
2024-09-12T06:54:14.720361719+02:00 stdout F [2024-09-12T04:54:14.720206068+00:00  INFO io_engine::bdev::nexus::nexus_bdev:nexus_bdev.rs:680] Nexus '331c0652-0a75-4a2c-8946-3caa0590af06' [open]: dynamic reconfiguration event: unplug, reconfiguring I/O channels completed with result: Ok
2024-09-12T06:54:14.7203678+02:00 stdout F [2024-09-12T04:54:14.720225935+00:00  INFO io_engine::bdev::nexus::nexus_child:nexus_child.rs:1157] Child 'bdev:///4711b421-0210-4db5-b88f-c2c55cac52da?uuid=4711b421-0210-4db5-b88f-c2c55cac52da @ 331c0652-0a75-4a2c-8946-3caa0590af06' [closed synced]: child successfully unplugged

This was found on another report: #1734

[tiagolobocastro]

Looks like there was a heartbeat failure, which caused control-plane to mark the node as offline. In turn, this means we didn't set the nexus node shutdown request. Nonetheless, we should not have attempted to destroy the replica, because the nexus was not verified as shutdown! Also we could have a check on the dataplane to avoid getting into trouble by ensuring replica is not being used?

[dsharma-dc]

Pool lock was taken and never released

The DestroyReplica call is the one getting starved of lock, but who is holding the pool lock here?

[tiagolobocastro]

Pool lock was taken and never released

The DestroyReplica call is the one getting starved of lock, but who is holding the pool lock here?

The first DestroyReplica call

[tiagolobocastro]

This ticket needs two fixes:

• [ ] Control-plane should not issue replica destroy if replica is still part of the nexus
• [ ] Data-plane should anyway not lockup the pool in case of control-plane bug

[tiagolobocastro]

Control-plane changes: https://github.com/openebs/mayastor-control-plane/pull/862