Hi there,

I've been trying to upgrade a Cassandra 3.9 cluster to Cassandra 3.11.
Whenever I try to add a new Cassandra 3.11 node to the main datacenter,
using `-Dcassandra.force_3_0_protocol_version=true` on the new node, this
new node uses almost 100% of its CPU. Checking `nodetool compactionstats
-H`, I can see there is always pending compacting tasks.

t0:
```
[aladdin@ip-172-16-1-10 ~]$ nodetool compactionstats -H
pending tasks: 11
- x.morning: 9
- y.afternoon: 2

id                                   compaction type keyspace      table
 completed  total     unit  progress
b0f174d0-8671-10e8-8429-114a220c107d Compaction      x morning 170.15 MiB
23.64 GiB bytes 0.70%
38b02640-8439-10e8-8429-114a220c107d Compaction      y   afternoon   2.55
GiB   40.17 GiB bytes 6.35%
Active compaction remaining time :   1h05m09s
```

t1:
```
[aladdin@ip-172-16-1-10 ~]$ nodetool compactionstats -H
pending tasks: 11
- x.morning: 9
- y.afternoon: 2

id                                   compaction type keyspace      table
 completed total     unit  progress
b0f174d0-8671-10e8-8429-114a220c107d Compaction      x morning 404.3 MiB
23.64 GiB bytes 30.20%
38b02640-8439-10e8-8429-114a220c107d Compaction      y   afternoon   2.55
GiB  40.17 GiB bytes 6.35%
Active compaction remaining time :   1h04m55s
```

t2:
```
[aladdin@ip-172-16-1-10 ~]$ nodetool compactionstats -H

pending tasks: 8
- system.size_estimates: 1
- x.morning: 4
- y.afternoon: 3

id                                   compaction type keyspace   table
 completed  total     unit  progress
38b02640-8439-10e8-8429-114a220c107d Compaction      y   afternoon   2.55
GiB   40.17 GiB bytes 6.35%
06f3a660-867b-10e8-8429-114a220c107d Compaction      x morning 538.04 MiB
11.15 GiB bytes 4.71%
```


`38b02640-8439-10e8-8429-114a220c107d` seems to be stuck at 6.35% for the
last 4 days and y. morning keeps compacting forever, always with a
different id.

Nodes are configured to run via `GossipingPropertyFileSnitch` on a i3.large
(AWS) and this problem seems to happen only on the main datacenter. The
secondary datacenter has one node running `Cassandra 3.9` and one node
running `Cassandra 3.11` without any further issues.

Have you guys faced this problem before?

Thanks in advance!
Igor

Reply via email to