Wilfred Spiegelenburg created YUNIKORN-3318:
-----------------------------------------------
Summary: Deadlock in preemption init caused by app reservations
Key: YUNIKORN-3318
URL: https://issues.apache.org/jira/browse/YUNIKORN-3318
Project: Apache YuniKorn
Issue Type: Bug
Components: core - scheduler
Affects Versions: 1.8.0
Reporter: Wilfred Spiegelenburg
Assignee: Wilfred Spiegelenburg
If an application {{A}} holds a reservation {{R}} on a node {{N}} the
preemption initialisation will deadlock.
During the {{initWorkingState()}} call reservations are released from a node if
they are older than the timeout and the priority is lower than the allocation
we are trying to preempt for. This is done as reserved nodes are excluded from
preemption. The initialisation is called by the scheduling cycle which holds
the application lock. The release of the reservation via {{UnReserve()}}
requires that same application lock.
Removing the reservation from the queue also requires the same application lock
as we need to get the queue from the application that holds the reservation.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
