I selected 2 GB as the HEAP usage on the nodes usually dips below 50%
for HEAP usage. After this large collection has completed, I plan on
reducing it down, I would not expect this memory to be used except
during full reindexes. Our normal weekly adds are < 60_000.
Presuming the change has taken effect after the reload as the cause,
I've not seen any indications of the prior slow downs we were
experiencing. IO Wait is now near or at 0 a good portion of the time
with occasional spikes.
On 11/4/21 10:03, Shawn Heisey wrote:
On 11/4/21 7:43 AM, Michael Conrad wrote:
ok,
will set the following to try and fix the issues with full reindexing.
<ramBufferSizeMB>2048</ramBufferSizeMB>
<maxBufferedDocs>1000000</maxBufferedDocs>
Setting 2GB of index buffer memory for each core is probably going to
lead to Solr being killed by its OOM script. Your max heap is only
5GB. The default buffer size has been 100MB for a while, and was 32MB
in much older versions. Experiments done in Solr's early days by
people a lot smarter than me determined that there was little benefit
for most users in setting that value above 128.
Thanks,
Shawn
