Hello Everyone,
Thanks or reading.
Question related to DSE to OSS C* migration.
Quick Backgroud: Successfully using DSE 5.1.11 with 600+ nodes, 3 DCs with
close to 1000+ tables response time (MAX) less than 1 sec. 99th P level: 1 to 8
milli sec. GC tuned to its fullest. GC=64 GB with G1. Avg GC pasue around
300ms. nodes are 540 GB memory 64 cores.
Now due to data, time and hardware constraints we had to do an inplace
migration (used /data folder as is). These are the steps follows:
1. Exact available configurations copied from DSE to OSS C*.
2. Altered few keyspace/tables
alter keyspace dse_system with replication = {'class':
'NetworkTopologyStrategy', 'DC3': '3'}; /DC1,DC2=OSS C*
alter table cfs_archive.sblocks with compaction = {'class':
'org.apache.cassandra.db.compaction.SizeTieredCompactionStrategy',
'max_threshold': '32', 'min_threshold': '4'};
alter table cfs.sblocks with compaction = {'class':
'org.apache.cassandra.db.compaction.SizeTieredCompactionStrategy',
'max_threshold': '32', 'min_threshold': '4'};
3.
auto_bootstrap: false
JVM_OPTS="$JVM_OPTS -Dcassandra.allow_unsafe_replace=true"
JVM_OPTS="$JVM_OPTS -Dcassandra.replace_address=..."
(Other usual options didn't work, other than replace address)
After migration of two DCs to Cassandra 3.11.3 (one DC still in DSE), we have
started seeing very high GC pause (1000ms+), high read latency and many hints,
read repairs.
Quesiton we have- Is there a possibility of data corruption with conversion?
Does DSE 5.1 have a different codebase than OSS 3.11? Any one faced this issue
before?
Thanks in advance
---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscr...@cassandra.apache.org
For additional commands, e-mail: user-h...@cassandra.apache.org
