Fixing store size calculation on KahaUpdateMessageCommand processing so
that the size won't increase inadvertently if the existing location of
the command in the journal is the same as the new location
(cherry picked from commit a5050a8bc5)
Fixing PList size tracking by moving the null assignment on removal to
after the onRemoval method is called to make sure that the decrement is
done properly.
(cherry picked from commit d635a36940)
Properly nulling out the scheduler service in MessageDatabase after it
is shutdown on a store close so that if the store is restarted the
thread will properly restart.
(cherry picked from commit 7bdcca1bda)
Moving beforeMarshall back to the store implementations because we don't
want all store implementations to marshall (such as memory store)
This reverts commit 32913408a6.
Calling beforeMarshall on messages when they async stored before the
store task is run and before consumer dispatch to prevent two threads
from trying to mutate the message state at the same time.
(cherry picked from commit b9f9f03829)
Expose configuration on the persistence adapter level.
Double the test timeout and increase the compaction frequency to account
for very slow CI boxes.
(cherry picked from commit a9521dcebf)
Computing messageSize for a durable subscription in KahaDB now runs much
faster (n vs n^2) which is noticable when there are a large number of
pending messages for a durable subscription.
(cherry picked from commit 25ff5699f1)
Ensure that when add / remove commands are colocated they don't prevent
the log from being GC'd once it is unreferenced.
(cherry picked from commit 8c4b5f485d)
Fixing the processing of KahaUpdateMessageCommand to add the message to
the store if it doesn't exist, instead of ignoring it. This will
prevent message loss in certain cases when the KahaDB index has to be
rebuilt.
(cherry picked from commit b4aa53d806)
Fixing JournalCorruptionEofIndexRecoveryTest class
Because of the size being stored in the message order index, the test
testRecoveryAftercorruptionCheckSumExistingIndex was no longer passing
as that test was corrupting the size value in the journal, but not the
size value in the index. The test now corrupts both values so that the
expect exception occurs again.
(cherry picked from commit 507d40a79f)
Fixing KahaDB so that the correct marshaller is used for the message
keys inside of the message order index. This will ensure that message
size metrics are accurate.
(cherry picked from commit 246ccb8e04)
Propertly re-setting the storeOpenWireVersion from the BrokerService
on the KahaDB Metadata if a corrupted index is detected and the
Metadata has to be recreated.
(cherry picked from commit 7a7c70ad75)
The journalSize counter should be set to 0 after KahaDB is shut down.
Usually, a store is not restarted again after stopping, but if it ever
was the value would be wrong since it wasn't reset back to 0.
Christopher L. Shannon (cshannon)
Timothy Bish
gtully
Torsten Mielke
Dejan Bosanac