Cassandra Caching Questions

(6 posts) (3 voices)

• Started 1 year ago by
• Latest reply from joaquin
• This topic is not a support question

Tags:

No tags yet.

1. 

   Anonymous
   Unregistered

   Hi friends,

   I am currently doing read performance tuning and had a few questions on Cassandra caching:

   From describe keyspace, I see these two settings under the ColumnFamily section:
   Row cache size / save period in seconds / keys to save : 0.0/0/all
   Key cache size / save period in seconds: 200000.0/14400

   1. I understand the row/key cache sizes can be set, how is the save period in seconds set?
   2. If there is x rows in cache, will these rows get flushed regardless after however many seconds set above in the save period (even if the cache doesn’t fill up)?
   3. These settings are cluster level and not node level, right?
   4. When inserting into the Cassandra cluster, before the MemTable is flushed to the SSTable, if we make a query to data that is currently in the MemTable, will it fetch it directly from memory (MemTable) then? Will the MemTable memory (for inserts) share the same configuration (row/key cache size) and space as the cache memory for reads?
   5. How does the caching mechanism work? Least recently read, first to leave the cache?

   I know this is quite a few questions, partial responses would be awesome too. Thanks so much in advance!

   Regards,
   Alvin

   Posted 1 year ago #
2. 

   khahn
   Member

   To answer your first question, you can use CQLsh 2.0 to set these column family attributes. For example, ALTER TABLE addamsFamily WITH comment = 'A most excellent and useful column family' AND row_cache_save_period_in_seconds = 1;

   Here is more information from our docs:

   key_cache_save_period_in_seconds
   Sets number of seconds between saving key caches: the key caches can be saved
   periodically, and if one exists on startup it will be loaded. This property is called key_cache_save_period in the cassandra-cli client.

   row_cache_save_period_in_seconds
   Sets the number of seconds between saving row caches: the row caches can be saved periodically, and if one exists on startup it will be loaded.

   Sorry, I can only provide a partial response. Hope it helps.

   Posted 1 year ago #
3. 

   khahn
   Member

   Correction: Use CLI. For example, http://comments.gmane.org/gmane.comp.db.cassandra.user/21967

   Posted 1 year ago #
4. 

   khahn
   Member

   Regarding your question 5, please see http://www.datastax.com/docs/1.0/operations/tuning.

   Posted 1 year ago #
5. 

   khahn
   Member

   Another correction: The row_cache_save_period_in_seconds and key_cache_save_period_in_seconds are obsolete. They no longer do anything. Sorry for the confusion.

   Posted 1 year ago #
6. 

   joaquin
   Administrator

   To answer your other questions:
   3. These settings are cluster level and not node level, right?

   Correct, almost all settings that can be changed via CQL and the cassandra-cli are cluster level.
   The cassandra.yaml and cassandra-env.sh file are, for the most part, node level. If set incorrectly according to cluster settings, you will error out and not be able to join the cluster as a safety measure.

   4. When inserting into the Cassandra cluster, before the MemTable is flushed to the SSTable, if we make a query to data that is currently in the MemTable, will it fetch it directly from memory (MemTable) then? Will the MemTable memory (for inserts) share the same configuration (row/key cache size) and space as the cache memory for reads?

   Yes, queries will check the MemTables first, then cache, then disk.

   No, these are two different spaces. Memtables can be configured in the cassandra.yaml, but the default of 1/3 of your total heap space is typically a great setting.

   Thanks,
   Joaquin

   Posted 1 year ago #

RSS feed for this topic