Ariel Weisberg created CASSANDRA-20210:
------------------------------------------
Summary: Improve performance of range reads on Accord
Key: CASSANDRA-20210
URL: https://issues.apache.org/jira/browse/CASSANDRA-20210
Project: Apache Cassandra
Issue Type: Improvement
Components: Accord, Consistency/Coordination
Reporter: Ariel Weisberg
Currently Accord range reads split the ranges provided by
{{RangeCommandIterator}} in a transaction into a read command per intersecting
command store that execute in parallel in a single transaction and then the
results are merged and the limit is re-applied. In Cassandra this command would
execute as a single read in a single thread and would only read up to the limit
once.
This could be improved on by having command stores read serially so we can
avoid reading and discarding extra data in the common case where there is a lot
data to be returned from the first command store.
The simplest approach would be to run a txn per command store from
{{RangeCommandIterator}}, but that requires the latency of scheduling an entire
transaction per command store and gets slower as you increase the number of
command stores per node so it doesn’t seem like a viable approach.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
