Umbrella parent issue for grouping all suspected transaction leaks
Description
Environment
Operating System: All
Platform: All
Attachments
- 07 Sep 2017, 10:57 AM
- 04 Sep 2017, 03:51 PM
blocks
is blocked by
Activity
Michael Vorburger February 19, 2018 at 1:44 PM
This is considered Done and complete - until we find the next ones (which https://lf-opendaylight.atlassian.net/browse/NETVIRT-878#icft=NETVIRT-878 should help prevent).
Stephen Kitt September 12, 2017 at 9:37 AM
https://git.opendaylight.org/gerrit/63027 should help with this in general.
Michael Vorburger September 7, 2017 at 10:57 AM
Attachment open-transactions.txt has been added with description: New trace:transactions output re-test of a custom distribution of latest (yesterday's) carbon/stable together with c/62674/5
Michael Vorburger September 7, 2017 at 10:57 AM
Attaching latest trace:transactions - list is already getting smaller - good!
Michael Vorburger September 4, 2017 at 3:51 PM
> new such dump tomorrow from a system after a full Rally load test,
> and will have more interesting Nx in there, which will probably help
> to understand priorities among these.
Attached new open-transactions.txt from a (quote Sridhar, tx!) "single iteration of rally scenario with the following configuration/scenarios" :
Concurrency: 16, 32
times: 300
a. create-list-network
b. create-list-port
c. create-list-router
d. create-list-security-group
e. create-list-subnet
shows 100s and in some VTN-19-2000 non-closed ReadOnlyTransaction and WriteOnlyTransaction obtained directly from DataBroker, and a handful (few) open TransactionChain.
This new parent issue will have all suspected DataBroker transaction leaks we find using the CONTROLLER-1760 tooling linked to it as blocking issues.
Some of the issues I'm about to link here were originally linked to as blocking CONTROLLER-1756 (which is what originally prompted this investigation), but has subsequently been found to not or unlikely be the cause of that particular OOM (in our scenario of stress tests with OpenStack Rally) - but should ideally still all be fixed - thus this new bug.
NB discussion in CONTROLLER-1765 re. possible false positives from CONTROLLER-1760.
NETVIRT-878 will probably need all of these to be fixed.