On 23/08/17 20:21, Michael Vorburger wrote: > On Wed, Aug 23, 2017 at 2:11 PM, Robert Varga <n...@hq.sk > <mailto:n...@hq.sk>> wrote: > > On 23/08/17 13:48, Michael Vorburger wrote: > > > > Robert> Actually no. The backend side of things looks okay, as chains > > are being > > both closed and purged when requested from the frontend. I suspect > > somebody is forgetting to close their transaction chains... > > > > Michael> Is a "transaction chain" the same as a transaction - the > > objects returned from DataBroker's newReadWriteTransaction / > > newReadWriteTransaction / newWriteOnlyTransaction? I did stumble upont > > something - could https://git.opendaylight.org/gerrit/#/c/62196/ > <https://git.opendaylight.org/gerrit/#/c/62196/> fix > > this problem? Or is that unlikely to fix this OOM, but still a Good > > Idea? Or doesn't matter to do cancel() if not submit() ? > > 'the same' in the sense that both are resources which need to be closed. > So yes, it does matter that they are properly closed. > > > Thanks. I spent the afternoon shooting in a couple of directions related > to this (details in https://bugs.opendaylight.org/show_bug.cgi?id=9034#). > > But the problem is what I've found so far is all just "stabs in the > dark". What we really need is a way to more reliably find the origin of > code where transactions (or TransactionChain) are created but never > closed... > > You mentioned on IRC that a datastore.cfg may have some option for that, > but I could not find this - does anyone know any details about that?
nite@nitebug : ~$ cd distribution-karaf-0.6.2-SNAPSHOT/ nite@nitebug : ~/distribution-karaf-0.6.2-SNAPSHOT$ ./bin/karaf opendaylight-user@root>feature:install odl-mdsal-broker nite@nitebug : ~/distribution-karaf-0.6.2-SNAPSHOT$ ls -l etc/org.opendaylight.controller.cluster.datastore.cfg -rw-rw-r--. 1 nite nite 4896 Aug 23 20:47 etc/org.opendaylight.controller.cluster.datastore.cfg for some reason debug-transactions is not documented. https://git.opendaylight.org/gerrit/62224 fixes that. > > Otherwise, an idea I just had in > https://bugs.opendaylight.org/show_bug.cgi?id=9034#c7 would be to see if > I could bolt something onto that mdsal-trace we have (originally > contributed by Josh) to keep track of opened-but-not-yet-closed > transactions. See the idea? Could be very useful to get to the bottom of > this kind of problem, no? No need, as mentioned above. > > > As far as I can see, with my still very limited understanding of > mdsal > > > internals, this does not seem to be the same as our earlier > > > https://bugs.opendaylight.org/show_bug.cgi?id=8941 > <https://bugs.opendaylight.org/show_bug.cgi?id=8941> > > <https://bugs.opendaylight.org/show_bug.cgi?id=8941 > <https://bugs.opendaylight.org/show_bug.cgi?id=8941>> raised by > > Stephen and > > > fixed by Robert (which is being follow-up on in > > > https://bugs.opendaylight.org/show_bug.cgi?id=9028 > <https://bugs.opendaylight.org/show_bug.cgi?id=9028> > > <https://bugs.opendaylight.org/show_bug.cgi?id=9028 > <https://bugs.opendaylight.org/show_bug.cgi?id=9028>> by Robert and > > Tom) - > > > does this initial quick analysis seem accurate to you? > > > > 9034 looks like 8941, except it's not transactions, but chains. I'll > > cook up a prototype. > > > > > > Thanks a lot !!! We'd love to out anything you come up with. > > As mentioned above, this is something else, so no patch coming from me > righ tnow ;) > > > OK, misunderstood because you wrote "I'll cook up a prototype" ... ? Yeah, I checked the code out and responded with "Actually no...", which you quoted above ;-) Bye, Robert
signature.asc
Description: OpenPGP digital signature
_______________________________________________ controller-dev mailing list controller-dev@lists.opendaylight.org https://lists.opendaylight.org/mailman/listinfo/controller-dev
