You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues-all@impala.apache.org by "ASF subversion and git services (Jira)" <ji...@apache.org> on 2020/09/10 20:48:00 UTC

[jira] [Commented] (IMPALA-9403) Allow TSAN to be set on codegen

    [ https://issues.apache.org/jira/browse/IMPALA-9403?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17193847#comment-17193847 ] 

ASF subversion and git services commented on IMPALA-9403:
---------------------------------------------------------

Commit f7dbd4939903b1dbb1994423f24a2f4159daf48a in impala's branch refs/heads/master from Sahil Takiar
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=f7dbd49 ]

IMPALA-9740, IMPALA-9403: Fix remaining custom cluster TSAN errors

This patch fixes the remaining TSAN errors reported while running custom
cluster tests. After this patch, TSAN can be enabled for custom cluster
tests (currently it is only run for be tests).

Adds a data race suppression for
HdfsColumnarScanner::ProcessScratchBatchCodegenOrInterpret, which
usually calls a codegen function. TSAN currently does not support
codegen functions, so this warning needs to be suppressed. The call
stack of this warning is:

    #0 kudu::BlockBloomFilter::Find(unsigned int) const kudu/util/block_bloom_filter.cc:257:7
    #1 <null> <null> (0x7f19af1c74cd)
    #2 impala::HdfsColumnarScanner::ProcessScratchBatchCodegenOrInterpret(impala::RowBatch*) exec/hdfs-columnar-scanner.cc:106:10
    #3 impala::HdfsColumnarScanner::TransferScratchTuples(impala::RowBatch*) exec/hdfs-columnar-scanner.cc:66:34

Fixes a data race in DmlExecState::FinalizeHdfsInsert where a local
HdfsFsCache::HdfsFsMap is unsafely passed between threads of a
HdfsOperationSet. HdfsOperationSet instances are run in a
HdfsOpThreadPool and each operation is run in one of the threads from
the pool. Each operation uses HdfsFsCache::GetConnection to get a hdfsFs
instance. GetConnection can take in a 'local_cache' of hdfsFs instances
before using the global map. The race condition is that the same local
cache is used for all operations in HdfsOperationSet.

Testing:
* Re-ran TSAN tests and confirmed the data races have disappeared

Change-Id: If1658a9b56d220e2cfd1f8b958604edcdf7757f4
Reviewed-on: http://gerrit.cloudera.org:8080/16426
Reviewed-by: Impala Public Jenkins <im...@cloudera.com>
Tested-by: Impala Public Jenkins <im...@cloudera.com>


> Allow TSAN to be set on codegen
> -------------------------------
>
>                 Key: IMPALA-9403
>                 URL: https://issues.apache.org/jira/browse/IMPALA-9403
>             Project: IMPALA
>          Issue Type: Sub-task
>          Components: Backend
>            Reporter: Sahil Takiar
>            Priority: Major
>
> Similar to this commit, but for TSAN. Requires adding the {{-fsanitize=thread}} flag to {{CLANG_IR_CXX_FLAGS}}.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org