Ok, so the documentation lacks and the examples are incorrect.
I my situation Native Persistence doesn't suit me, I have a read-through
cache to a 3rd party persistence (CacheLoader API) - no write-through. I
don't expect data to be recoverable if partitions are lost, so swap would
suit me just fine, all I need is to not unnecessarily have to fetch data
from persistence.
This is my situation:
On a physical server with 128GB RAM I am running multiple ignite server
nodes each with 3.5GB heap, 1GB DirectMemory and 256MB MetaDataSpace. With
swapPath defined on the data region, when is swap supposed to kick in?
Ideally my cluster nodes should consume no more than ~5GB of ram each, but
it looks like it is consuming as much memory as it wants.
Metrics for local node
...
^-- Heap [used=2246MB, free=35.82%, comm=3500MB]
^-- Off-heap memory [used=1406MB, free=80.5%, allocated=1772MB]
Limited DirectMemory using -XX:MaxDirectMemorySize=2g is not adhered to,
data region keep trying to expand in memory.
I would expect a node started with -Xmx2g and -XX:MaxDirectMemorySize=2g not
to exceed 4GB of memory regardless of data region size, ie. 4GB. I would
expect swap (if configured) to be used before exceeding 4GB, and probably
expect it to crash with an OOM if swap was not configured.
What am I missing?
ibelyakov wrote
> Yes, you're correct, there is a mistake in the documentation, according to
> the example a data region max size will be 5Gb not 500Mb.
>
> Also you can use Native Persistence for your case:
> https://www.gridgain.com/docs/latest/developers-guide/persistence/native-persistence
>
> You can specify data region with maxSize=1Gb and enabled persistence then
> your data will be stored on disk and GridGain will keep up to 1Gb of the
> "hot" data in the RAM.
>
> Igor
--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
