You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@cassandra.apache.org by Mikael Taina <mi...@giosg.com> on 2018/01/11 08:01:48 UTC

High memory usage when running repair with 3.11.1

Hi I'm having a problem after upgrading Cassandra 3.0.14 to 3.11.1.

When running a repair the node I'm issuing the repair starts to use use all
available memory and then the GC grinds it to death. I'm seeing with
nodetool compactionstats that there is a lot of (> 100) active concurrent
compaction jobs typed 'Validation'.

I'm running the repair with `nodetool repair -pr -seq keyspace table`

Cassandra has 18G max heap memory, but that doesn't seem to be even close
to enough.

With visualvm I can see that big portion of the live heap is used on the
validation's objects:
https://imgur.com/U0WUWEs
MerkleTree$Inner: 5G
MerkleTree$Leaf: 3.2G
Murmur3Partitioner$LongToken: 2.5G

Any ideas?