5 Cassandra Message Store Pre-Installation Tasks

Linux Tuning Settings

To optimize Cassandra on Linux, see the DataStax recommendations at:

https://docs.datastax.com/en/landing_page/doc/landing_page/recommendedSettingsLinux.html

dse File

For DC_INDEX nodes, which run Solr, make the following change to the /etc/default/dse file:

SOLR_ENABLED=1

dse.yaml File

For DC_INDEX (Solr) nodes, make the following changes to the /etc/dse/cassandra/dse.yaml file to improve performance:

max_solr_concurrency_per_core: 6
back_pressure_threshold_per_core: 5000
cql_slow_log_option:
     enabled: false

cassandra.yaml File

Make the changes in this section to the /etc/dse/cassandra/cassandra.yaml file.

For all nodes, to enable separate clusters for better performance, specify cluster_name.

Make the following changes to the num_tokens setting:

• DC_MSG, DC_META, and DC_CACHE nodes:

  num_tokens: 256
  

• DC_INDEX (Solr) nodes:

  num_tokens: 16
  allocate_tokens_for_local_replication_factor 
  

On DC_INDEX (Solr) nodes, make the following change to the allocate_tokens_for_local_replication_factor setting:

allocate_tokens_for_local_replication_factor replication_factor

where replication_factor is derived from the store.cassolrrf configuration option, and by default has a value of 2.

To improve performance, locate data on SSD drives:

• data_file_directories:

  /var/lib/cassandra/data
  

• commitlog_directory:

  /var/lib/cassandra/commitlog
  

• saved_caches_directory:

  /var/lib/cassandra/saved_caches
  

• hints_directory:

  /var/lib/cassandra/hints
  

To support large mailbox and message, increase the commitlog size:

commitlog_segment_size_in_mb: 256

To specify seed nodes, you must use two nodes from each data center in the cluster, preferably located on different racks, so that each cluster has different seeds, for example:

• DC_MSG cluster:

  seeds: "192.0.2.12,192.0.2.24"
  

• DC_META/DC_INDEX cluster:

  seeds: "192.0.2.1,192.0.2.2,192.0.2.10,192.0.2.3"
  

• DC_CACHE cluster:

  seeds: "192.0.2.14,192.0.2.7"
  

For DC_INDEX nodes, make the following changes to improve performance:

memtable_heap_space_in_mb: 2048

For all nodes, make the following change to improve performance:

memtable_flush_writers: 8

For all nodes, to specify listen_address, rpc_address, and so on, make the following changes:

listen_address: 10.128.128.12
rpc_address: 10.128.128.12

cassandra-env.sh File

For all nodes, to specify the location of the heap dump, make the following change to the /etc/dse/cassandra/cassandra-env.sh file:

export CASSANDRA_HEAPDUMP_DIR=/scratch/heapdump

jvm.options File

For DC_MSGDC_META, DC_INDEX, and DC_CACHE nodes, make the following heap size changes:

-Xms32G
-Xmx32G

For DC_META nodes, to improve performance, make the following heap size changes to improve performance:

-Xms16G
-Xmx16G

For all nodes, make the following changes to improve performance:

-XX:InitiatingHeapOccupancyPercent=70
-XX:ParallelGCThreads=12
-XX:ConcGCThreads=12 

For all nodes, print garbage collection measurements, which are useful for monitoring system performance:

-XX:+PrintGCDetails
-XX:+PrintGCDateStamps
-XX:+PrintHeapAtGC
-XX:+PrintTenuringDistribution
-XX:+PrintGCApplicationStoppedTime
-XX:+PrintPromotionFailure
-XX:PrintFLSStatistics=1
-Xloggc:/var/log/cassandra/gc.log
-XX:+UseGCLogFileRotation
-XX:NumberOfGCLogFiles=10
-XX:GCLogFileSize=10M

cassandra-rackdc.properties File

For all nodes, make the following changes to the cassandra-rackdc.properties file.

1. Configure the endpoint snitch:

   endpoint_snitch: GossipingPropertyFileSnitch
   

2. Set the data center and rack names as appropriate:

   dc=mydc
   rack=myrac
   

   For example, for a node in DC_CACHE in a physical rack in one location, set dc=DC_CACHE and rack=RAC1. And, for another node in DC_CACHE in a physical rack in another location, set dc=DC_CACHE and rack=RAC2.

   Note:

   Data center and rack names are case sensitive.