You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
MAPREDUCE-7474. Task save failure stats in job stats + more
Statistics Collection and Printing
* New statistic task_stage_save_summary_file to distinguish from
other saving operations (job success/report file)
* After a failure to save a task attempt, the iostats of the manifest
are rebuilt so the stats on failures are updated.
This will get into the final job _SUCCESS statistics so we can see
if anything happened
* Make the manifest print command something which can be invoked from
the commandline: mapred successfile
This is covered in the docs.
The failure stats regeneration is nice; works by passing down a
lambda-expression of the logic to (re)generate the manifest, and invoking
this on every attempt. As this is where the stats are aggregated,
it includes details on the previous failing attempts.
Directory size for deletion
* Optionally pass down directory count under job dir to cleanup stage
* This is determined in job commit from aggregate statistics;
unknown elsewhere (app abort etc.).
* It is currently only logged; it may be possible to support an option
of when to skip the initial serial delete, though it will depend on
abfs login mechanism.
Testing
* More fault injection scenarios.
* Ability to assert that iostats do not contain specific non-zero stats.
This is used in ITestAbfsTerasort to assert no task save or rename failures.
The stats before this change imply this did happen in a job commit; no
other details, hence the new probe.
* Log manifest committer at debug in mapred-core
Note: if there's a retry process which means the operation can take minutes,
the initial operation will block progress() callbacks so mapreduce jobs
will fail. Spark is unaffected
Change-Id: Id423267de89c7f31e4b1283f9c433b729ff0d87b
Copy file name to clipboardExpand all lines: hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/lib/output/committer/manifest/ManifestCommitterStatisticNames.java
+6Lines changed: 6 additions & 0 deletions
Original file line number
Diff line number
Diff line change
@@ -187,6 +187,12 @@ public final class ManifestCommitterStatisticNames {
Copy file name to clipboardExpand all lines: hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/lib/output/committer/manifest/files/ManifestPrinter.java
Copy file name to clipboardExpand all lines: hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/lib/output/committer/manifest/impl/InternalConstants.java
Copy file name to clipboardExpand all lines: hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/lib/output/committer/manifest/stages/AbstractJobOrTaskStage.java
Copy file name to clipboardExpand all lines: hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/lib/output/committer/manifest/stages/CleanupJobStage.java
+37-9Lines changed: 37 additions & 9 deletions
Original file line number
Diff line number
Diff line change
@@ -160,10 +160,18 @@ protected Result executeStage(
Copy file name to clipboardExpand all lines: hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/lib/output/committer/manifest/stages/CommitJobStage.java
0 commit comments