zachjsh commented on code in PR #12443:
URL: https://github.com/apache/druid/pull/12443#discussion_r869920252
##########
indexing-service/src/main/java/org/apache/druid/indexing/common/task/batch/parallel/ParallelIndexSupervisorTask.java:
##########
@@ -901,13 +906,43 @@ public static Map<Interval, Integer>
determineNumShardsFromCardinalityReport(
{
// aggregate all the sub-reports
Map<Interval, Union> finalCollectors = mergeCardinalityReports(reports);
+ return computeIntervalToNumShards(maxRowsPerSegment, finalCollectors);
+ }
+ @Nonnull
+ @VisibleForTesting
+ static Map<Interval, Integer> computeIntervalToNumShards(
+ int maxRowsPerSegment,
+ Map<Interval, Union> finalCollectors
+ )
+ {
return CollectionUtils.mapValues(
finalCollectors,
union -> {
final double estimatedCardinality = union.getEstimate();
- // determine numShards based on maxRowsPerSegment and the cardinality
- final long estimatedNumShards = Math.round(estimatedCardinality /
maxRowsPerSegment);
+ final long estimatedNumShards;
+ if (estimatedCardinality <= 0) {
+ // I don't think we can use the estimate in any way being
negative, seven sounds like a nice prime number
+ // it is ok if we end up not filling them all, the ingestion code
handles that
+ // Seven on the other hand will at least create some shards rather
than potentially a single huge one
+ estimatedNumShards = 7L;
Review Comment:
But in the case that we get into this situation, should the fall back here
be configurable, so that if 7 results in a bad estimate / default value, the
job can be rerun with a different value and produce potentially better results?
--
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]
