[
https://issues.apache.org/jira/browse/HBASE-4645?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13132894#comment-13132894
]
[email protected] commented on HBASE-4645:
------------------------------------------------------
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/2524/#review2756
-----------------------------------------------------------
Amit-- code changes look solid.
It is unfortunate that there was a nice test TestWALReplay.java to test this
exact case-- but was broken in catching this issue.
Furthermore, the test writes to multiple CFs in separate puts (transactions).
It will be good to enhance it to do cross-CF puts (in a single txn).
src/test/java/org/apache/hadoop/hbase/regionserver/wal/TestWALReplay.java
<https://reviews.apache.org/r/2524/#comment6197>
looks like the original test is busted. It expects to flush only one region
(in line 271); but here region.close() will cause it to flush all stores :)
- Kannan
On 2011-10-21 18:06:16, Amitanand Aiyer wrote:
bq.
bq. -----------------------------------------------------------
bq. This is an automatically generated e-mail. To reply, visit:
bq. https://reviews.apache.org/r/2524/
bq. -----------------------------------------------------------
bq.
bq. (Updated 2011-10-21 18:06:16)
bq.
bq.
bq. Review request for Ted Yu, Michael Stack, Jonathan Gray, Lars Hofhansl,
Amitanand Aiyer, Kannan Muthukkaruppan, Karthik Ranganathan, and Nicolas
Spiegelberg.
bq.
bq.
bq. Summary
bq. -------
bq.
bq. There is a data loss happening (for some of the column families) when we
do the replay logs.
bq.
bq. The bug seems to be from the fact that during replay-logs we only choose
to replay
bq. the logs from the maximumSequenceID across ALL the stores. This is wrong.
If a
bq. column family is ahead of others (because the crash happened before all
the column
bq. families were flushed), then we lose data for the column families that
have not yet
bq. caught up.
bq.
bq. The correct logic for replay should begin the replay from the minimum
across the
bq. maximum in each store.
bq.
bq.
bq. This addresses bug hbase-4645.
bq. https://issues.apache.org/jira/browse/hbase-4645
bq.
bq.
bq. Diffs
bq. -----
bq.
bq. src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java 8c32839
bq.
src/test/java/org/apache/hadoop/hbase/regionserver/wal/TestWALReplay.java
966262b
bq.
bq. Diff: https://reviews.apache.org/r/2524/diff
bq.
bq.
bq. Testing
bq. -------
bq.
bq. Initial patch. v1.
bq.
bq. mvn test (running).
bq.
bq. TBD: add a test case to repro the issue and make sure it fixes.
bq.
bq.
bq. Thanks,
bq.
bq. Amitanand
bq.
bq.
> Edits Log recovery losing data across column families
> -----------------------------------------------------
>
> Key: HBASE-4645
> URL: https://issues.apache.org/jira/browse/HBASE-4645
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.89.20100924, 0.92.0
> Reporter: Amitanand Aiyer
> Assignee: Amitanand Aiyer
>
> There is a data loss happening (for some of the column families) when we do
> the replay logs.
> The bug seems to be from the fact that during replay-logs we only choose to
> replay
> the logs from the maximumSequenceID across *ALL* the stores. This is wrong.
> If a
> column family is ahead of others (because the crash happened before all the
> column
> families were flushed), then we lose data for the column families that have
> not yet
> caught up.
> The correct logic for replay should begin the replay from the minimum across
> the
> maximum in each store.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
