how many partition is suitable per each node in cassandra

0
votes

I'm testing cassandra. before setting my cassandra nodes at server.

I wonder how many partitions are suitable per each node .

  • i'm planning to save 0~18000000 data per each partition.
  • Q: partitions?????? / 1node ==> stable
    CREATE TABLE if not exists  access_log
 (  time_boundary int, --yyyymmddhh
    user_id TEXT,  --user01
    timestamp bigint, --yyyymmddhhmmssms
    url TEXT, -- /file/file/blabla~~
    menu_id TEXT, 
    ip TEXT,
    params map, -- 
    PRIMARY KEY((time_boundary),user_id,timestamp)
 ) ;
1
cassandranosqlpartition
and i'm planning to create 25000 partitions per 1 node – yeon
please show your table schema, and other information – Alex Ott
i added my table schema for sample. – yeon

1 Answers

1
votes

I consider a safe rule of thumb (depends heavily on schema and write rates) to be about 1-4 billion partitions within a table (per node) you start to get issues with repair over streaming with the 2^20 max merkle tree depth.

With incremental repair you can go higher (only recommended on latest 3.11.x, and even there test heavily), providing you are using LeveledCompactionStrategy. With SizeTieredCompactionStrategy ALTER TABLE yourtable WITH min_index_interval = 4096 and max_index_interval = 4096 (or higher if you see issues) to prevent an OOM when the compactions preallocate the index summaries.