Journal replay failure after rm -R and re-create on a mount point

Description

To trigger this issue, start an Alluxio cluster and run

1 2 3 alluxio fs mount /foo /tmp alluxio fs rm -R /foo alluxio fs mount /foo /tmp

Then restart Alluxio. The master fails to come up, with the error

1 2 2018-10-24 08:28:07,465 ERROR UfsJournalCheckpointThread - FileSystemMaster: Failed to run journal checkpoint thread, crashing. java.lang.RuntimeException: alluxio.exception.FileAlreadyExistsException: Mount point /foo already exists

This is because when we replay rm -R, we don't remove mount points. So on replay, we see an existing mount point being re-created.

To fix this, we should update the replay path for rm -R so that it removes mount points.

Environment

None

Status

Assignee

Andrew Audibert

Reporter

Andrew Audibert

Labels

None

Components

Affects versions

1.7.1
1.8.1
1.6.0
1.8.0

Priority

Critical
Configure