[
https://issues.apache.org/jira/browse/FLINK-4123?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15361147#comment-15361147
]
ASF GitHub Bot commented on FLINK-4123:
---------------------------------------
Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/2183#discussion_r69440065
--- Diff:
flink-streaming-connectors/flink-connector-cassandra/src/main/java/org/apache/flink/streaming/connectors/cassandra/CassandraTupleWriteAheadSink.java
---
@@ -132,11 +132,15 @@ protected void sendValues(Iterable<IN> values, long
timestamp) throws Exception
}
try {
while (updatesSent != updatesConfirmed.get()) {
+ if (exception.get() != null) { // verify that
no query failed until now
+ LOG.warn("Sending a value failed.",
exception.get());
+ break;
+ }
Thread.sleep(100);
}
--- End diff --
Could we try to replace the busy waiting loop with a wait-notify construct?
That would decrease latency and wouldn't burn unnecessary CPU cycles. We could
sync on the `updateConfirmed` variable.
> CassandraWriteAheadSink can hang on cassandra failure
> -----------------------------------------------------
>
> Key: FLINK-4123
> URL: https://issues.apache.org/jira/browse/FLINK-4123
> Project: Flink
> Issue Type: Bug
> Components: Cassandra Connector
> Affects Versions: 1.1.0
> Reporter: Chesnay Schepler
> Assignee: Chesnay Schepler
> Priority: Blocker
> Fix For: 1.1.0
>
>
> The CassandraWriteAheadSink verifies that all writes send to cassandra have
> been applied by counting how many were sent and how many callbacks were
> activated. Once all writes were sent the sink enters into a loop that is only
> exited once both counts are equal.
> Thus, should cassandra crash after all writes were sent, without having
> acknowledged all writes, the sink will deadlock in the loop.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
