GitHub user tillrohrmann opened a pull request:
https://github.com/apache/flink/pull/318
[FLINK-1376] [runtime] Add proper shared slot release in case of a fatal
TaskManager failure for 0.8 branch
This PR introduces SharedSlots as being a special Slot type and as such
being released properly in case an Instance has been marked dead. This fixes
the problem that a dead instance, which has not been shutdown properly, causes
a job not being removed properly from the system, because it is not aware of
the SubSlots.
Adds test cases where only the heartbeat thread of TaskManager is killed.
Except for the test cases, this is basically the same PR as #317 just
rebased on the 0.8 branch.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/tillrohrmann/flink fixSharedSlotRelease0.8
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/318.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #318
----
commit f5e960e9e4bca79bc2dc138a881446b0668c2c5a
Author: Till Rohrmann <[email protected]>
Date: 2015-01-12T09:58:45Z
[FLINK-1376] [runtime] Add proper shared slot release in case of a fatal
TaskManager failure.
Fixes concurrent modification exception of SharedSlot's subSlots field by
synchronizing all state changing operations through the associated assignment
group. Fixes deadlock where Instance.markDead first acquires InstanceLock and
then by releasing the associated slots the assignment group lockcan block with
a direct releaseSlot call on a SharedSlot which first acquires the assignment
group lock and then the instance lock in order to return the slot to the
instance.
Fixes colocation shared slot releasing. A colocation constraint is now
realized as a SharedSlot in a SharedSlot where the colocated tasks allocate sub
slots.
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
