Hi, Actualy dead lock detection for pessimistic transactions was introduced in Apache Ignite 1.7 release. But anyway it is good idea to try the same code on the newest release.
In case of trasaction timeout dead lock detection should be triggered and TransactionTimoutException should contain TransactionDeadlockException as cause if dead lock was detected. Is it your case or not? If transaction doesn't release any locks (or non transactional locks) then you will get messages in logs about long running transactions or futures because exchange can't be finished in this case. Do you have something like to this messages? Is it possible to create some reproducer project that we can analyze? On Sat, Feb 18, 2017 at 1:42 AM, bintisepaha <[email protected]> wrote: > Thanks Andrew, > > The same thing happened again today. Clearly the key is locked, we get the > timeout exceptions. But prior update to the same has not thrown any > exceptions. Suddenly one update fails with timeout exceptions and we are > notified due to those exceptions that the key is locked. > > We will upgrade to 1.8, but in the meantime is there a way to free up this > locked key using some code? > > We try killing nodes, but we have one back up and it looks like the lock is > carried over too, which would be the right thing to do. > > Outside the transaction we can read this key (dirty read). This is becoming > an issue for us, since its a production system and the only way to free up > is to restart the cluster. > > Please point us in a direction where we can avoid this or free it up. > > Thanks, > Binti > > > > -- > View this message in context: > http://apache-ignite-users.70518.x6.nabble.com/Pessimistic-TXN-did-not-release-lock-on-a-key-all-subsequent-txns-failed-tp10536p10713.html > Sent from the Apache Ignite Users mailing list archive at Nabble.com.
