A big benefit of cqrs is to be able to replay events during development and debugging.
Currently in my application a replay of 90k events takes 5 mins. Besides optimizing the Readmodel for performance, what would you do to make replay faster? I already tried the async cluster but on my 8-core machine there was no improvement
thx ro
What version of axon you use? Jdbc or jpa event store?
We read 3.5M events in approx 5 minutes. A have a quite fast mysql server. I guess it's the xml deserialiser that slows us down now...
I use the MongoDB Eventstore,
But do you tell me that you replay events only on your dedicated mysql server? or that you have an optimized config for mysql?
The xml deserialiser, can I tun em somehow?
Shared mysql server for writes and reads. Slow when the server is cold, approx 30 mins, 5 mins when it's warm.
We use jdbc event store, that is pretty fast in reading events.
I don't know if the xml deserialiser can be optimized...
Hi,
the XStream serializer can be tuned a bit, but don’t expect to much improvement in the area of performance. By default, it uses the XML Pull Parser, which is reasonably good.
Something you might do t improve performance is keep your DBObjects in memory as long as possible. That allows you to remove much of the I/O towards MongoDB. Of course, that’s only possible if your event handlers don’t do update queries.
Did you try profiling? That should pinpoint the #1 performance issue.
Cheers,
Allard
“MongoDB writes to RAM in our days”, but there are findings (see link) that read operations could be improved by switching to Redis.
At least for the read side this could be a viable way.
https://stackoverflow.com/questions/5252577/how-much-faster-is-redis-than-mongodb
Btw: Profiling shows that the “last modified” date that I update for each event of an aggregate is consuming 20% of the overall replay time. This indicates that on the other hand the read side has no big computational bottleneck, its really mongo that’s slow.
thanks, ro