Ariel Weisberg created CASSANDRA-20588:
------------------------------------------
Summary: Fix single token batch atomicity with Accord/non-Accord
batches
Key: CASSANDRA-20588
URL: https://issues.apache.org/jira/browse/CASSANDRA-20588
Project: Apache Cassandra
Issue Type: Bug
Components: Accord
Reporter: Ariel Weisberg
Single token batches are currently 100% atomic without Accord because they are
sent to a single node and can be applied to multiple tables out of the commit
log reliably. Currently if that happens and one table is running on Accord and
one is not then the batch will be split it into a Cassandra write and an Accord
write and lose atomicity.
The most straightforward fix is probably to route the entire batch through
Accord if any of it touches an Accord table. It’s fine to use Accord to apply
writes and this won’t impact the latency since Accord was being used anyways.
This will require some modifications to TxnUpdate and maybe TxnWrite to allow
the entire contents of the batch even for tables Accord isn’t managing to tag
along and be applied at the same time as the involved Accord tables.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
