Memory only strategy will throw mlock error if disk_access_mode set incorrectly

Summary

When using in-memory tables in DSE. The setting of disk_access_mode in the cassandra.yaml must be mmap

Applies to

• DSE 5.1 and earlier

Symptoms

If a a table with the compaction strategy org.apache.cassandra.db.compaction.MemoryOnlyStrategy is created and the node has disk_access_mode: standard set in the cassandra.yaml, DSE can fail to start with the following error:

ERROR [main] 2019-01-22 12:15:38,368 MemoryOnlyStrategy.java:401 - There was a problem mlocking a segment for file BigTableReader(path='/var/lib/cassandra/data/mykeyspace/mytable-81718281fa3511e7b2f67198b740403f/.mms_hier_lfd/mc-13-big-Data.db') 
ERROR [main] 2018-06-22 12:15:38,368 MemoryOnlyStrategy.java:402 - Exception: java.lang.IllegalArgumentException: SSTable Index is not mmapped, is compression turned off? BufferedSegmentedFile(path='/var/lib/cassandra/data/mykeyspace/mytable -81718281fa3511e7b2f67198b740403f/.mms_hier_lfd/mc-13-big-Index.db', length=355)

Cause

Memory only SSTables have to be mapped into off-heap memory. Therefore, the disk_access_mode setting must use a mode that supports this.

Solution

Ensure the disk_access_mode setting in cassandra.yaml is set to mmap.

See also

• DataStax Enterprise documentation cassandra.yaml
• Support KB article FAQ - Use of disk_access_mode in DSE 5.1 and earlier
• Support KB article Increased memory use on nodes after upgrading to DSE 5.0 or DSE 5.1