[
https://issues.apache.org/jira/browse/TEZ-4521?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
László Bodor updated TEZ-4521:
------------------------------
Fix Version/s: 0.10.3
> Partition stats should be always uncompressed size
> --------------------------------------------------
>
> Key: TEZ-4521
> URL: https://issues.apache.org/jira/browse/TEZ-4521
> Project: Apache Tez
> Issue Type: Sub-task
> Affects Versions: 0.10.2
> Reporter: okumin
> Assignee: okumin
> Priority: Major
> Fix For: 0.10.3
>
> Time Spent: 1h 10m
> Remaining Estimate: 0h
>
> We always put compressed size in
> [ExternalSorter#partitionStats|https://github.com/apache/tez/blob/rel/release-0.10.2/tez-runtime-library/src/main/java/org/apache/tez/runtime/library/common/sort/impl/ExternalSorter.java#L134]
> while we put uncompressed size in
> [UnorderedPartitionedKVWriter#sizePerPartition|https://github.com/apache/tez/blob/rel/release-0.10.2/tez-runtime-library/src/main/java/org/apache/tez/runtime/library/common/writers/UnorderedPartitionedKVWriter.java#L130-L131].
> Those should have consistent semantics.
>
> As far as I know, the uncompressed size is preferable because of some reasons.
> # The stats are used in FairShuffleVertexManager to configure the
> parallelism. The normal ShuffleVertexManager which is broadly used computes
> parallelism based on uncompressed size. Otherwise, we need to tune
> `tez.fair-shuffle-vertex-manager.desired-task-input-size` based on compressed
> size though `tez.shuffle-vertex-manager.desired-task-input-size` must be
> based on decompressed size
> # Ming pointed out we should use uncompressed size in TEZ-3206. Looks like,
> we missed creating a follow-up ticket
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
