[jira] [Commented] (HBASE-6725) Check and Put can fail when using locks
[ https://issues.apache.org/jira/browse/HBASE-6725?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13452235#comment-13452235 ] Nicolas Thiébaud commented on HBASE-6725: - The bug comes from the fact that checkAndMutate uses getLock(lockId, get.getRow(), true) and resuses the provided lock. Since several client threads use the same lock many are allowed to mutate. I believe there are 2 solutions: a. Use some sort of internal lock, which requires a specific lock set for check and mutate b. Consider CAP with locks as bad usage of locks and deprecate the feature. I'm willing to contribute for this patch, but I'm not sure how to go on from now. Check and Put can fail when using locks --- Key: HBASE-6725 URL: https://issues.apache.org/jira/browse/HBASE-6725 Project: HBase Issue Type: Bug Components: master Affects Versions: 0.90.4 Environment: tested on mac os x Reporter: Nicolas Thiébaud Attachments: CAPwithLocks.zip, TestCase_HBASE_6725.patch When multiple threads race using CAP with and a lock on the same row, several instances may be allowed to update the cell with the new value (although the expected value is different). If all threads race with a wrong expected value and a lock, none will be able to update. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (HBASE-6725) Check and Put can fail when using locks
[ https://issues.apache.org/jira/browse/HBASE-6725?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13452246#comment-13452246 ] Ted Yu commented on HBASE-6725: --- @Nicolas: Thanks for the finding. For point b, can you write a summary on dev@ list to poll people's opinion ? Check and Put can fail when using locks --- Key: HBASE-6725 URL: https://issues.apache.org/jira/browse/HBASE-6725 Project: HBase Issue Type: Bug Components: master Affects Versions: 0.90.4 Environment: tested on mac os x Reporter: Nicolas Thiébaud Attachments: CAPwithLocks.zip, TestCase_HBASE_6725.patch When multiple threads race using CAP with and a lock on the same row, several instances may be allowed to update the cell with the new value (although the expected value is different). If all threads race with a wrong expected value and a lock, none will be able to update. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (HBASE-6725) Check and Put can fail when using locks
[ https://issues.apache.org/jira/browse/HBASE-6725?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13452513#comment-13452513 ] Nicolas Thiébaud commented on HBASE-6725: - Done. Check and Put can fail when using locks --- Key: HBASE-6725 URL: https://issues.apache.org/jira/browse/HBASE-6725 Project: HBase Issue Type: Bug Components: master Affects Versions: 0.90.4 Environment: tested on mac os x Reporter: Nicolas Thiébaud Attachments: CAPwithLocks.zip, TestCase_HBASE_6725.patch When multiple threads race using CAP with and a lock on the same row, several instances may be allowed to update the cell with the new value (although the expected value is different). If all threads race with a wrong expected value and a lock, none will be able to update. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (HBASE-6725) Check and Put can fail when using locks
[ https://issues.apache.org/jira/browse/HBASE-6725?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13451272#comment-13451272 ] Nicolas Thiébaud commented on HBASE-6725: - It also fails with non-empty expected value. I'm going to try to find the bug and provide a patch. Check and Put can fail when using locks --- Key: HBASE-6725 URL: https://issues.apache.org/jira/browse/HBASE-6725 Project: HBase Issue Type: Bug Components: master Affects Versions: 0.90.4 Environment: tested on mac os x Reporter: Nicolas Thiébaud Attachments: CAPwithLocks.zip, TestCase_HBASE_6725.patch When using CAP with an empty array as expected value and using a lock on the corresponding row, hbase may allow a Put even if the value was already there. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (HBASE-6725) Check and Put can fail when using locks
[
https://issues.apache.org/jira/browse/HBASE-6725?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13450781#comment-13450781
]
Ted Yu commented on HBASE-6725:
---
I tried the test against 0.92 and trunk:
{code}
Failed tests:
testConcurrentCAPWithLocks(org.apache.hadoop.hbase.client.TestConcurrentCAPWithLocks):
expected:5 but was:1
{code}
Check and Put can fail when using locks
---
Key: HBASE-6725
URL: https://issues.apache.org/jira/browse/HBASE-6725
Project: HBase
Issue Type: Bug
Components: master
Affects Versions: 0.90.4
Environment: tested on mac os x
Reporter: Nicolas Thiébaud
Attachments: CAPwithLocks.zip, TestCase_HBASE_6725.patch
When using CAP with an empty array as expected value and using a lock on the
corresponding row, hbase may allow a Put even if the value was already there.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (HBASE-6725) Check and Put can fail when using locks
[
https://issues.apache.org/jira/browse/HBASE-6725?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13450819#comment-13450819
]
Ted Yu commented on HBASE-6725:
---
nit:
{code}
for(int c : counts) Assert.assertEquals(c, 1);
{code}
c and 1 should be exchanged so that the assertion failure is clear to the user.
Does this only fail with an empty array as expected value ?
It would be nice if you can provide patch for this issue.
Check and Put can fail when using locks
---
Key: HBASE-6725
URL: https://issues.apache.org/jira/browse/HBASE-6725
Project: HBase
Issue Type: Bug
Components: master
Affects Versions: 0.90.4
Environment: tested on mac os x
Reporter: Nicolas Thiébaud
Attachments: CAPwithLocks.zip, TestCase_HBASE_6725.patch
When using CAP with an empty array as expected value and using a lock on the
corresponding row, hbase may allow a Put even if the value was already there.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (HBASE-6725) Check and Put can fail when using locks
[ https://issues.apache.org/jira/browse/HBASE-6725?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13449875#comment-13449875 ] Ted Yu commented on HBASE-6725: --- @Nicolas: Can you format ConcurrentCAPWithLocks class as a proper test so that we can easily verify this bug ? Thanks Check and Put can fail when using locks --- Key: HBASE-6725 URL: https://issues.apache.org/jira/browse/HBASE-6725 Project: HBase Issue Type: Bug Components: master Affects Versions: 0.90.4 Environment: tested on mac os x Reporter: Nicolas Thiébaud Attachments: CAPwithLocks.zip When using CAP with an empty array as expected value and using a lock on the corresponding row, hbase may allow a Put even if the value was already there. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira
