What kind of listener you used? Could you share your configuration or
link to a reproducer?
I've checked write behind feature with mysql instance and don't find any
performance loss (i receive cache put events instantly via registered
listener but they appear in DB with configured time skew/batch size.
Ignite cluster and mysql server were on same computer)
Note that if you use local listener cache PUTS operation that happens on
remote node (due to affinity function distribution) will not be visible
on local node.
Concerning flush buffer location it is located in heap memory. Check out
*org.apache.ignite.internal.processors.cache.store.GridCacheWriteBehindStore*
class.
This component is responsible for spreading data among so called
flushers. Flusher is a worker-like component that has queue of data
chunks prepared for transmission to target storage (e.g. mysql).
04.06.2021 01:32, gigabot пишет:
I observed that turning on writeThrough with writeBehind slows down
performance(i.e. time data reaches listeners) greatly. I tried setting the
flush frequency to 0 and the flush size to 1 billion, and so the flushes are
not even happening yet, and still the performance is slow. I am wondering
where the slowdown is occurring? Where is the flush buffer located
(heap/off-heap/disk/etc.) and it is configurable?
--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
