Looks like you're running a third party company's distribution of HBase.
I'd recommend you start by engaging them for support.
Architecturally, HBase does very little when there is no client load
applied to the system. If you're experiencing OOME's when the system is
idle, that sounds like you might have some kind of basic misconfiguration.
On 8/3/21 8:56 AM, Anshuman Singh wrote:
Hi,
I have a 7 node Hbase 2.0.0.3.1 cluster with 12 TB of data, 128MB memstore
flush size and 20GB regions.
At a time writing happens only on 7 regions and I’m not performing any reads.
I’m seeing OOM errors with 32 GB heap even if I stop the ingestions.
I followed this blog post related to Hbase RegionServer sizing
http://hadoop-hbase.blogspot.com/2013/01/hbase-region-server-memory-sizing.html.
But here according to the explanation only the active regions ( on which new
data is being written) will impact the heap/memstore but the heap usage even
without any ingestion is touching 32GB. Even the memstore size as shown on the
Hbase UI is 0 and the block cache is totally free.
I took heap dumps and it shows the majority of heap being used by byte[]
objects of regionserverHStore and regionserverHRegionServer.
Can anyone point me in the right direction about what may be causing high heap
usage even without reads and writes?
Regards,
Anshuman