aweisberg commented on code in PR #3694:
URL: https://github.com/apache/cassandra/pull/3694#discussion_r1848867807
##########
src/java/org/apache/cassandra/service/accord/interop/AccordInteropExecution.java:
##########
@@ -290,6 +291,66 @@ private AsyncChain<Data> readChains()
}));
});
});
+ return results;
+ }
+
+ private List<AsyncChain<Data>> rangeReadChains(long nowInSeconds,
Dispatcher.RequestTime requestTime)
+ {
+ TxnRangeRead read = (TxnRangeRead) txn.read();
+ Seekables<?, ?> keys = txn.read().keys();
+ List<AsyncChain<Data>> results = new ArrayList<>();
+ keys.forEach(key -> {
+ TokenRange range = (TokenRange)key;
+ PartitionRangeReadCommand command = read.commandForSubrange(range,
nowInSeconds);
+
+ // This should only rarely occur when coordinators start a
transaction in a migrating range
+ // because they haven't yet updated their cluster metadata.
+ // It would be harmless to do the read, because it will be
rejected in `TxnQuery` anyways,
+ // but it's faster to skip the read
+ // TODO (required): To make migration work we need to validate
that the range is all on Accord
+ // if any part isn't we should reject the read
Review Comment:
We can't forcibly sequence things running on the correct system because
Accord might run the transaction in a later epoch from when it was submitted
and we only find out when we go to execute it.
Here we need to validate like keys do that every key being read is readable
by Accord.
Users don't see the error because it's automatically retried again on the
correct system.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
