[jira] [Commented] (HDFS-14869) Data loss in case of distcp using snapshot diff. Replication should include rename records if file was skipped in the previous iteration
[
https://issues.apache.org/jira/browse/HDFS-14869?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16989699#comment-16989699
]
Hudson commented on HDFS-14869:
---
SUCCESS: Integrated in Jenkins build Hadoop-trunk-Commit #17732 (See
[https://builds.apache.org/job/Hadoop-trunk-Commit/17732/])
HDFS-14869 Copy renamed files which are not excluded anymore by filter
(shashikant: rev fc97034b29243a0509633849de55aa734859)
* (edit)
hadoop-tools/hadoop-distcp/src/main/java/org/apache/hadoop/tools/DistCpSync.java
* (edit)
hadoop-tools/hadoop-distcp/src/main/java/org/apache/hadoop/tools/DistCp.java
* (edit)
hadoop-tools/hadoop-distcp/src/test/java/org/apache/hadoop/tools/TestDistCpSync.java
> Data loss in case of distcp using snapshot diff. Replication should include
> rename records if file was skipped in the previous iteration
>
>
> Key: HDFS-14869
> URL: https://issues.apache.org/jira/browse/HDFS-14869
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: distcp
>Reporter: Aasha Medhi
>Assignee: Aasha Medhi
>Priority: Major
> Fix For: 3.1.4
>
>
> This issue arises when a directory or file is excluded by exclusion filter
> during distcp replication. Later on if the directory is renamed later to a
> name which is not excluded by the filter, the snapshot diff reports only a
> rename operation. The directory is never copied to target even though its
> not excluded now. This also doesn't throw any error so there is no way to
> find the issue.
> Steps to reproduce
> * Create a directory in hdfs to copy using distcp.
> * Include a staging folder in the directory.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop fs -ls
> /tmp/tocopy
> Found 4 items
> -rw-r--r-- 3 hdfs hdfs 16 2019-09-12 10:32 /tmp/tocopy/.b.txt
> drwxr-xr-x - hdfs hdfs 0 2019-09-23 09:18 /tmp/tocopy/.staging
> -rw-r--r-- 3 hdfs hdfs 12 2019-09-12 10:32 /tmp/tocopy/a.txt
> -rw-r--r-- 3 hdfs hdfs 4 2019-09-20 08:23 /tmp/tocopy/foo.txt{code}
> * The exclusion filter is set to exclude any staging directory
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ cat
> /tmp/filter
> .*\.Trash.*
> .*\.staging.*{code}
> * Do a copy using distcp snapshots, the staging directory is not replicated.
> {code:java}
> hadoop jar hadoop-distcp-3.3.0-SNAPSHOT.jar
> -Dmapreduce.job.user.classpath.first=true -filters /tmp/filter
> /tmp/tocopy/.snapshot/s1 /tmp/target
> [hdfs@ctr-e141-1563959304486-33995-01-03 root]$ hadoop fs -ls /tmp/target
> Found 3 items
> -rw-r--r-- 3 hdfs hdfs 16 2019-09-24 06:56 /tmp/target/.b.txt
> -rw-r--r-- 3 hdfs hdfs 12 2019-09-24 06:56 /tmp/target/a.txt
> -rw-r--r-- 3 hdfs hdfs 4 2019-09-24 06:56 /tmp/target/foo.txt{code}
> * Rename the staging directory to final
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop fs -mv
> /tmp/tocopy/.staging /tmp/tocopy/final{code}
> * Do a copy using snapshot diff.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hdfs
> snapshotDiff /tmp/tocopy s1 s2[hdfs@ctr-e141-1563959304486-33995-01-03
> hadoop-mapreduce]$ hdfs snapshotDiff /tmp/tocopy s1 s2Difference between
> snapshot s1 and snapshot s2 under directory /tmp/tocopy:M .R ./.staging ->
> ./final
> {code}
> * The diff report just has a rename record and the new final directory is
> never copied.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop jar
> hadoop-distcp-3.3.0-SNAPSHOT.jar -Dmapreduce.job.user.classpath.first=true
> -filters /tmp/filter -diff s1 s2 -update /tmp/tocopy /tmp/target
> 19/09/24 07:05:32 INFO tools.DistCp: Input Options:
> DistCpOptions{atomicCommit=false, syncFolder=true, deleteMissing=false,
> ignoreFailures=false, overwrite=false, append=false, useDiff=true,
> useRdiff=false, fromSnapshot=s1, toSnapshot=s2, skipCRC=false, blocking=true,
> numListstatusThreads=0, maxMaps=20, mapBandwidth=0.0,
> copyStrategy='uniformsize', preserveStatus=[BLOCKSIZE], atomicWorkPath=null,
> logPath=null, sourceFileListing=null, sourcePaths=[/tmp/tocopy],
> targetPath=/tmp/target, filtersFile='/tmp/filter', blocksPerChunk=0,
> copyBufferSize=8192, verboseLog=false, directWrite=false},
> sourcePaths=[/tmp/tocopy], targetPathExists=true, preserveRawXattrsfalse
> 19/09/24 07:05:32 INFO client.RMProxy: Connecting to ResourceManager at
> ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128:8050
> 19/09/24 07:05:33 INFO client.AHSProxy: Connecting to Application History
> server at ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128
[jira] [Commented] (HDFS-14869) Data loss in case of distcp using snapshot diff. Replication should include rename records if file was skipped in the previous iteration
[
https://issues.apache.org/jira/browse/HDFS-14869?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16948384#comment-16948384
]
Steve Loughran commented on HDFS-14869:
---
as I said on the PR, I do not know enough about distCp to be safe to review
this.
I've had a quick glance at the diff and suggested changes to the test based on
what I like to see in tests, but someone else is going to need to have to look
at the core algorithm.
Sorry
> Data loss in case of distcp using snapshot diff. Replication should include
> rename records if file was skipped in the previous iteration
>
>
> Key: HDFS-14869
> URL: https://issues.apache.org/jira/browse/HDFS-14869
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: distcp
>Reporter: Aasha Medhi
>Assignee: Aasha Medhi
>Priority: Major
>
> This issue arises when a directory or file is excluded by exclusion filter
> during distcp replication. Later on if the directory is renamed later to a
> name which is not excluded by the filter, the snapshot diff reports only a
> rename operation. The directory is never copied to target even though its
> not excluded now. This also doesn't throw any error so there is no way to
> find the issue.
> Steps to reproduce
> * Create a directory in hdfs to copy using distcp.
> * Include a staging folder in the directory.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop fs -ls
> /tmp/tocopy
> Found 4 items
> -rw-r--r-- 3 hdfs hdfs 16 2019-09-12 10:32 /tmp/tocopy/.b.txt
> drwxr-xr-x - hdfs hdfs 0 2019-09-23 09:18 /tmp/tocopy/.staging
> -rw-r--r-- 3 hdfs hdfs 12 2019-09-12 10:32 /tmp/tocopy/a.txt
> -rw-r--r-- 3 hdfs hdfs 4 2019-09-20 08:23 /tmp/tocopy/foo.txt{code}
> * The exclusion filter is set to exclude any staging directory
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ cat
> /tmp/filter
> .*\.Trash.*
> .*\.staging.*{code}
> * Do a copy using distcp snapshots, the staging directory is not replicated.
> {code:java}
> hadoop jar hadoop-distcp-3.3.0-SNAPSHOT.jar
> -Dmapreduce.job.user.classpath.first=true -filters /tmp/filter
> /tmp/tocopy/.snapshot/s1 /tmp/target
> [hdfs@ctr-e141-1563959304486-33995-01-03 root]$ hadoop fs -ls /tmp/target
> Found 3 items
> -rw-r--r-- 3 hdfs hdfs 16 2019-09-24 06:56 /tmp/target/.b.txt
> -rw-r--r-- 3 hdfs hdfs 12 2019-09-24 06:56 /tmp/target/a.txt
> -rw-r--r-- 3 hdfs hdfs 4 2019-09-24 06:56 /tmp/target/foo.txt{code}
> * Rename the staging directory to final
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop fs -mv
> /tmp/tocopy/.staging /tmp/tocopy/final{code}
> * Do a copy using snapshot diff.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hdfs
> snapshotDiff /tmp/tocopy s1 s2[hdfs@ctr-e141-1563959304486-33995-01-03
> hadoop-mapreduce]$ hdfs snapshotDiff /tmp/tocopy s1 s2Difference between
> snapshot s1 and snapshot s2 under directory /tmp/tocopy:M .R ./.staging ->
> ./final
> {code}
> * The diff report just has a rename record and the new final directory is
> never copied.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop jar
> hadoop-distcp-3.3.0-SNAPSHOT.jar -Dmapreduce.job.user.classpath.first=true
> -filters /tmp/filter -diff s1 s2 -update /tmp/tocopy /tmp/target
> 19/09/24 07:05:32 INFO tools.DistCp: Input Options:
> DistCpOptions{atomicCommit=false, syncFolder=true, deleteMissing=false,
> ignoreFailures=false, overwrite=false, append=false, useDiff=true,
> useRdiff=false, fromSnapshot=s1, toSnapshot=s2, skipCRC=false, blocking=true,
> numListstatusThreads=0, maxMaps=20, mapBandwidth=0.0,
> copyStrategy='uniformsize', preserveStatus=[BLOCKSIZE], atomicWorkPath=null,
> logPath=null, sourceFileListing=null, sourcePaths=[/tmp/tocopy],
> targetPath=/tmp/target, filtersFile='/tmp/filter', blocksPerChunk=0,
> copyBufferSize=8192, verboseLog=false, directWrite=false},
> sourcePaths=[/tmp/tocopy], targetPathExists=true, preserveRawXattrsfalse
> 19/09/24 07:05:32 INFO client.RMProxy: Connecting to ResourceManager at
> ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128:8050
> 19/09/24 07:05:33 INFO client.AHSProxy: Connecting to Application History
> server at ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128:10200
> 19/09/24 07:05:33 INFO tools.DistCp: Number of paths in the copy list: 0
> 19/09/24 07:05:33 INFO client.RMProxy: Connecting to ResourceManager at
> ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128:8050
> 19/09/24 07:05:33 INFO client.AHSProxy: Connecti
[jira] [Commented] (HDFS-14869) Data loss in case of distcp using snapshot diff. Replication should include rename records if file was skipped in the previous iteration
[
https://issues.apache.org/jira/browse/HDFS-14869?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16948318#comment-16948318
]
Shashikant Banerjee commented on HDFS-14869:
[~ste...@apache.org], can you please have a look at this?
> Data loss in case of distcp using snapshot diff. Replication should include
> rename records if file was skipped in the previous iteration
>
>
> Key: HDFS-14869
> URL: https://issues.apache.org/jira/browse/HDFS-14869
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: distcp
>Reporter: Aasha Medhi
>Assignee: Aasha Medhi
>Priority: Major
>
> This issue arises when a directory or file is excluded by exclusion filter
> during distcp replication. Later on if the directory is renamed later to a
> name which is not excluded by the filter, the snapshot diff reports only a
> rename operation. The directory is never copied to target even though its
> not excluded now. This also doesn't throw any error so there is no way to
> find the issue.
> Steps to reproduce
> * Create a directory in hdfs to copy using distcp.
> * Include a staging folder in the directory.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop fs -ls
> /tmp/tocopy
> Found 4 items
> -rw-r--r-- 3 hdfs hdfs 16 2019-09-12 10:32 /tmp/tocopy/.b.txt
> drwxr-xr-x - hdfs hdfs 0 2019-09-23 09:18 /tmp/tocopy/.staging
> -rw-r--r-- 3 hdfs hdfs 12 2019-09-12 10:32 /tmp/tocopy/a.txt
> -rw-r--r-- 3 hdfs hdfs 4 2019-09-20 08:23 /tmp/tocopy/foo.txt{code}
> * The exclusion filter is set to exclude any staging directory
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ cat
> /tmp/filter
> .*\.Trash.*
> .*\.staging.*{code}
> * Do a copy using distcp snapshots, the staging directory is not replicated.
> {code:java}
> hadoop jar hadoop-distcp-3.3.0-SNAPSHOT.jar
> -Dmapreduce.job.user.classpath.first=true -filters /tmp/filter
> /tmp/tocopy/.snapshot/s1 /tmp/target
> [hdfs@ctr-e141-1563959304486-33995-01-03 root]$ hadoop fs -ls /tmp/target
> Found 3 items
> -rw-r--r-- 3 hdfs hdfs 16 2019-09-24 06:56 /tmp/target/.b.txt
> -rw-r--r-- 3 hdfs hdfs 12 2019-09-24 06:56 /tmp/target/a.txt
> -rw-r--r-- 3 hdfs hdfs 4 2019-09-24 06:56 /tmp/target/foo.txt{code}
> * Rename the staging directory to final
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop fs -mv
> /tmp/tocopy/.staging /tmp/tocopy/final{code}
> * Do a copy using snapshot diff.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hdfs
> snapshotDiff /tmp/tocopy s1 s2[hdfs@ctr-e141-1563959304486-33995-01-03
> hadoop-mapreduce]$ hdfs snapshotDiff /tmp/tocopy s1 s2Difference between
> snapshot s1 and snapshot s2 under directory /tmp/tocopy:M .R ./.staging ->
> ./final
> {code}
> * The diff report just has a rename record and the new final directory is
> never copied.
> {code:java}
> [hdfs@ctr-e141-1563959304486-33995-01-03 hadoop-mapreduce]$ hadoop jar
> hadoop-distcp-3.3.0-SNAPSHOT.jar -Dmapreduce.job.user.classpath.first=true
> -filters /tmp/filter -diff s1 s2 -update /tmp/tocopy /tmp/target
> 19/09/24 07:05:32 INFO tools.DistCp: Input Options:
> DistCpOptions{atomicCommit=false, syncFolder=true, deleteMissing=false,
> ignoreFailures=false, overwrite=false, append=false, useDiff=true,
> useRdiff=false, fromSnapshot=s1, toSnapshot=s2, skipCRC=false, blocking=true,
> numListstatusThreads=0, maxMaps=20, mapBandwidth=0.0,
> copyStrategy='uniformsize', preserveStatus=[BLOCKSIZE], atomicWorkPath=null,
> logPath=null, sourceFileListing=null, sourcePaths=[/tmp/tocopy],
> targetPath=/tmp/target, filtersFile='/tmp/filter', blocksPerChunk=0,
> copyBufferSize=8192, verboseLog=false, directWrite=false},
> sourcePaths=[/tmp/tocopy], targetPathExists=true, preserveRawXattrsfalse
> 19/09/24 07:05:32 INFO client.RMProxy: Connecting to ResourceManager at
> ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128:8050
> 19/09/24 07:05:33 INFO client.AHSProxy: Connecting to Application History
> server at ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128:10200
> 19/09/24 07:05:33 INFO tools.DistCp: Number of paths in the copy list: 0
> 19/09/24 07:05:33 INFO client.RMProxy: Connecting to ResourceManager at
> ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128:8050
> 19/09/24 07:05:33 INFO client.AHSProxy: Connecting to Application History
> server at ctr-e141-1563959304486-33995-01-03.hwx.site/172.27.68.128:10200
> 19/09/24 07:05:33 INFO mapreduce.JobResourceUploader: Disabling Erasure
> Coding for path: /user
