Github user mridulm commented on a diff in the pull request:
https://github.com/apache/spark/pull/17276#discussion_r108103956
--- Diff:
core/src/main/java/org/apache/spark/shuffle/sort/BypassMergeSortShuffleWriter.java
---
@@ -169,6 +173,36 @@ public void write(Iterator<Product2<K, V>> records)
throws IOException {
}
}
mapStatus = MapStatus$.MODULE$.apply(blockManager.shuffleServerId(),
partitionLengths);
+ if (mapStatus instanceof HighlyCompressedMapStatus) {
+ HighlyCompressedMapStatus hc = (HighlyCompressedMapStatus) mapStatus;
+ long underestimatedBlocksSize = 0L;
+ for (int i = 0; i < partitionLengths.length; i++) {
+ if (partitionLengths[i] > mapStatus.getSizeForBlock(i)) {
+ underestimatedBlocksSize += partitionLengths[i];
+ }
+ }
+ writeMetrics.incUnderestimatedBlocksSize(underestimatedBlocksSize);
+ if (logger.isDebugEnabled() && partitionLengths.length > 0) {
+ int underestimatedBlocksNum = 0;
+ // Distribution of sizes in MapStatus.
+ double[] cp = new double[partitionLengths.length];
+ for (int i = 0; i < partitionLengths.length; i++) {
+ cp[i] = partitionLengths[i];
+ if (partitionLengths[i] > mapStatus.getSizeForBlock(i)) {
+ underestimatedBlocksNum++;
+ }
+ }
+ Distribution distribution = new Distribution(cp, 0, cp.length);
+ double[] probabilities = {0.0, 0.25, 0.5, 0.75, 1.0};
+ String distributionStr =
distribution.getQuantiles(probabilities).mkString(", ");
+ logger.debug("For task {}.{} in stage {} (TID {}), the block sizes
in MapStatus are " +
+ "inaccurate (average is {}, {} blocks underestimated, size of
underestimated is {})," +
+ " distribution at the given probabilities(0, 0.25, 0.5, 0.75,
1.0) is {}.",
+ taskContext.partitionId(), taskContext.attemptNumber(),
taskContext.stageId(),
+ taskContext.taskAttemptId(), hc.getAvgSize(),
+ underestimatedBlocksNum, underestimatedBlocksSize,
distributionStr);
+ }
+ }
--- End diff --
The value is not accurate - it is a 1og 1.1 'compression' which converts
the long size to a byte : and caps the value at 255.
So there are two errors introduced; it over-estimates the actual block size
when compressed value < 255 [1] (which is something this PR currently ignores),
when block size goes above 34k mb or so, it under estimates the block size
(which is higher than what spark currently supports due to 2G limitation).
[1] I did not realize it always over-estimates; if the current PR is
targetting only blocks which are under estimated; I would agree that not
handling `CompressedMapStatus` for time being might be ok - though would be
good to add a comment to that effect on 'why' we dont need to handle it.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
