[Zebra] Zebra does not support concurrent deletions of column groups now.
-------------------------------------------------------------------------
Key: PIG-1057
URL: https://issues.apache.org/jira/browse/PIG-1057
Project: Pig
Issue Type: Bug
Affects Versions: 0.4.0
Reporter: Chao Wang
Assignee: Chao Wang
Fix For: 0.6.0
Zebra does not support concurrent deletions of column groups now. As a result,
the TestDropColumnGroup testcase can sometimes fail due to this.
In this testcase, multiple threads will be launched together, with each one
deleting one particular column group. The following exception can be thrown
(with callstack):
/*************************************************************************************************************************/
...
java.io.FileNotFoundException: File
/.../pig-trunk/build/contrib/zebra/test/data/DropCGTest/CG02 does not exist.
at
org.apache.hadoop.fs.RawLocalFileSystem.getFileStatus(RawLocalFileSystem.java:361)
at
org.apache.hadoop.fs.RawLocalFileSystem.listStatus(RawLocalFileSystem.java:290)
at org.apache.hadoop.fs.FileSystem.listStatus(FileSystem.java:716)
at org.apache.hadoop.fs.FileSystem.listStatus(FileSystem.java:741)
at
org.apache.hadoop.fs.ChecksumFileSystem.listStatus(ChecksumFileSystem.java:465)
at
org.apache.hadoop.zebra.io.BasicTable$SchemaFile.setCGDeletedFlags(BasicTable.java:1610)
at
org.apache.hadoop.zebra.io.BasicTable$SchemaFile.readSchemaFile(BasicTable.java:1593)
at
org.apache.hadoop.zebra.io.BasicTable$SchemaFile.<init>(BasicTable.java:1416)
at org.apache.hadoop.zebra.io.BasicTable.dropColumnGroup(BasicTable.java:133)
at
org.apache.hadoop.zebra.io.TestDropColumnGroup$DropThread.run(TestDropColumnGroup.java:772)
...
/*************************************************************************************************************************/
We plan to fix this in Zebra to support concurrent deletions of column groups.
The root cause is that a thread or process reads in some stale file system
information (e.g., it sees /CG0 first) and then can fail later on (it tries to
access /CG0, however /CG0 may be deleted by another thread or process).
Therefore, we plan to adopt a retry logic to resolve this issue. More detailed,
we allow a dropping column group thread to retry n times when doing its
deleting job - n is the total number of column groups.
Note that here we do NOT try to resolve the more general concurrent column
group deletions + reads issue. If a process is reading some data that could be
deleted by another process, it can fail as we expect.
Here we only try to resolve the concurrent column group deletions issue. If you
have multiple threads or processes to delete column groups, they should succeed.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
