Issue #1114
closed
If an instance of pulp_resource_manager dies unexpectedly, Pulp incorrectly tries to "cancel all tasks in its queue"
Description
1. Start a clustered pulp installation with two machines in the cluster. Suppose these hostnames are called boxA and boxB. Start all pulp_* and httpd services on boxA.
2. Start a second pulp_resource_manager instance on boxB.
3. Use the /status/ API to verify that you can see both entries. They should show as 'resource_manager@boxA' and 'resource_manager@boxB'.
4. kill -9 the pulp_resource_manager service on boxB.
5. Wait for 6 or 7 minutes
6. Observe a traceback similar to the following in the logs of boxA.
pulp.server.async.scheduler:ERROR: Workers 'resource_manager@boxB' has gone missing, removing from list of workers
pulp.server.async.tasks:ERROR: The worker named resource_manager@boxB is missing. Canceling the tasks in its queue.
Two things are wrong with this, and both of them are located in this section of code.
(1) It should never call _delete_worker(worker.name) which attempts to cancel tasks, log, and clean up reservations, none of which make sense to do for pulp_resource_manager. Instead it should delete the worker record synchronously and continue.
(2) The error message is misleading. I'll suggest it should read something like:
resource_manager@boxB has gone missing.
Related issues
.