StefanRRichter commented on a change in pull request #7009: [FLINK-10712]
Support to restore state when using RestartPipelinedRegionStrategy
URL: https://github.com/apache/flink/pull/7009#discussion_r245659768
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/StateAssignmentOperation.java
##########
@@ -54,6 +55,7 @@
private static final Logger LOG =
LoggerFactory.getLogger(StateAssignmentOperation.class);
private final Map<JobVertexID, ExecutionJobVertex> tasks;
+ private final Map<JobVertexID, BitSet> taskIndices;
Review comment:
In my previous suggestion, what I meant was really using something like the
interface `Iterable<Integer>`, not directly the implementation of `BitSet`.
Then we could have two implementations of the iterable, one that delegates to a
bitset internally, and one for the general case that just takes a number
(parallelism) and generates a sequence from 0..parallelism. This would keep the
general code path more memory friendly and does not tie us to `BitSet`. Instead
of `Iterable`, `Collection` could be considered if you find that knowing the
size opens up optimizations.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
