Eugene Kirpichov created BEAM-3424:
--------------------------------------
Summary: CassandraIO uses 1 split if can't estimate size
Key: BEAM-3424
URL: https://issues.apache.org/jira/browse/BEAM-3424
Project: Beam
Issue Type: Bug
Components: sdk-java-extensions
Reporter: Eugene Kirpichov
Assignee: Jean-Baptiste Onofré
See
https://stackoverflow.com/questions/48090668/how-to-increase-dataflow-read-parallelism-from-cassandra?noredirect=1#comment83227824_48090668
. When CassandraIO can't estimate size, it falls back to a single split:
https://github.com/apache/beam/blob/master/sdks/java/io/cassandra/src/main/java/org/apache/beam/sdk/io/cassandra/CassandraServiceImpl.java#L196
A single split is very poor for performance. We should fall back to a different
value. Not sure what a good value would be; probably the largest value that
still doesn't introduce too much per-split overhead? E.g. would there be any
downside to just changing that number to 100?
Alternatively/additionally, like in DatastoreIO, CassandraIO could accept
requested number of splits as a parameter.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
