Github user liancheng commented on a diff in the pull request:
https://github.com/apache/spark/pull/8687#discussion_r39155809
--- Diff: core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala
---
@@ -979,8 +976,9 @@ class PairRDDFunctions[K, V](self: RDD[(K, V)])
hadoopConf.set("mapred.output.compression.type",
CompressionType.BLOCK.toString)
}
- // Use configured output committer if already set
- if (conf.getOutputCommitter == null) {
+ // Use configured output committer if already set and speculation is
not enabled.
+ val speculationEnabled = self.conf.getBoolean("spark.speculation",
false)
+ if (speculationEnabled || conf.getOutputCommitter == null) {
--- End diff --
This is also somewhat tricky. Although rare, some output formats may set
their own output committers. And if they do, they use the same configuration
key (i.e. `mapred.output.committer.class`). One of the example is [Parquet]
[1], which uses a specialized committer to write summary files. So unlike the
case in Spark SQL, here we cannot distinguish a real user-defined committer
(which is probably a direct committer), or a default committer bundled with the
output format. I'm not quite sure forcing all output formats to use
`FileOutputCommitter` when speculation is enabled is a good idea. On the other
hand, I don't know any other output formats other than Parquet that use
customized output committers, and it should be safe to assert that Spark users
don't use `saveAsHadoopFile()` to write Parquet files.
[1]:
https://github.com/apache/parquet-mr/blob/apache-parquet-1.7.0/parquet-hadoop/src/main/java/org/apache/parquet/hadoop/mapred/DeprecatedParquetOutputFormat.java#L60
---
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]
