On Tue, 27 Dec 2016, Sage Weil wrote: > I'm chasing a pretty reproducible crash of the rocksdb librados > env backend where it asserts on crypto shutdown with this PR applied: > > https://github.com/ceph/ceph/pull/12624 > > This is generally caused by a memory leak of some crypto object (e.g., > CryptoKey). If it run the test in a loop with ms type = simple it's fine, > but with async it crashes after about 10 minutes. On one run valgrind > reported a leak from AsyncConnection, although it didn't make any > sense (recv_buf?). But my suspicion is that the > AsyncMessenger::shutdown() call to stack->drain() isn't sufficient to make > sure all of the connections have actually been destroyed, which means the > cct and crypto shutdown asserts sometimes. > > Does that theory make sense? Any suggestions how to make sure that > the AsyncConnections are reliably reaped during librados shutdown? ...and as soon as I hit send my msg/simple run in a loop also crashed. Nevermind! sage > > Thanks! > sage > > > BTW here's an example of the crash, although it's not particularly > informative in and of itself: > > http://pulpito.ceph.com/sage-2016-12-27_18:31:32-rados-wip-sage-testing---basic-smithi/670919 > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
