[CONTROLLER-1760] Tooling to find the real root cause culprit of memory leaks related to non-closed transactions (and tx chains)

Blocks — Wed, 15 Nov 2017 16:53:24 +0000

~~CONTROLLER-1756~~ (and derived ~~OPNFLWPLUG-933~~) made me realize that we really do need some sort of "tooling" to be able to find the real root cause culprit of memory leaks related to non-closed transactions (and tx chains) ...

Originally, I thought that one can find this in HPROF heap dumps, but thinking this through further after Tom's reply on https://lists.opendaylight.org/pipermail/mdsal-dev/2017-August/001323.html and a comment from Robert on IRC, I realized that of course you actually cannot - the "left over" (non closed) transactions piling up on the heap obviously don't have any trace anymore to where they where originally created - thus where the origin of the leak is, that must be bug fixed, is "lost", at that point.

It turns out that the mdsal-trace feature in controller already has a useful basis for such a feature (which is a debug utility, not something we would want in production), and I've started to work on this in:

https://git.opendaylight.org/gerrit/#/c/62288/
https://git.opendaylight.org/gerrit/#/c/62289/
https://git.opendaylight.org/gerrit/#/c/62294/
more coming this week

As this currently stand, there's a chance (hope) that ~~OPNFLWPLUG-933~~ fixes the root cause for ~~CONTROLLER-1756~~ - but if we have this utility, we can be sure that is the one, and we can find other non-closed transaction leaks, today and in the future. Therefore I'd like to create this tool independent of ~~CONTROLLER-1756~~ and ~~OPNFLWPLUG-933~~ (and am thus creating this new separate bugzilla issue).

OpenDaylight JIRA

[CONTROLLER-1760] Tooling to find the real root cause culprit of memory leaks related to non-closed transactions (and tx chains)