Description
when looking at the speed of broker.stop with kahadb and the scheduler store. There is a full gc run, which can be expensive as the whole index needs to be traversed.
Fast stop/restart is important for fast failover. Leaving gc for runtime, where it has an effect on latency in the normal way, rather than availability, is better.
I am wondering if there is a use case for gc only at shutdown if the cleanupInterval <= 0, indicating that there were no gc at runtime. The alternative is adding another boolean to the config or adding that back in if the need arises.
I am leaning towards just removing the gc call during shutdown.
Note: matching the indexCacheSize to the index file size, trading off with memory, does help to speed up the index (read) traversal.
Attachments
Issue Links
- is related to
-
AMQ-3646 Allow KahaDB to run without disk syncs, higher through put without the jms persistence guarantee
-
- Resolved
-