[
https://issues.apache.org/jira/browse/HBASE-14004?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15050322#comment-15050322
]
Yu Li commented on HBASE-14004:
-------------------------------
bq. Yu, is it still a problem after we not using hsync?
My concern was mainly about whether it's possible to have duplicated entries in
WAL of *different* RS.
Think about the fail over case, replication queue will be transfered to some
other RS then entries of the failed RS WAL will be replicated, meanwhile the
same WAL will be split, replayed and its entries written into WAL of the new RS
serving the same region. In this situation we add a {{isReplay}} flag in
WALEdit to avoid duplicated replication (see below code segment in
{{FSHlog#append}})
{code}
if (entry.getEdit().isReplay()) {
// Set replication scope null so that this won't be replicated
entry.getKey().setScopes(null);
}
{code}
I could see a similar situation here:
# hflush timeout due to network failure but actually persisted
# WAL logger tries to re-write the buffered entries to a new WAL but new WAL
creation failed due to the same network failure, and returns fail to client
# region got reassigned due to balance or hbase shell command
# client retry writing to the same region served by a new RS and succeed
In this case we have duplicated entries in different RS
Feel free to correct me if the assumption is wrong, but if it's possible, then
we need to handle it in HBASE-14949 [~chenheng]
> [Replication] Inconsistency between Memstore and WAL may result in data in
> remote cluster that is not in the origin
> -------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-14004
> URL: https://issues.apache.org/jira/browse/HBASE-14004
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Reporter: He Liangliang
> Priority: Critical
> Labels: replication, wal
>
> Looks like the current write path can cause inconsistency between
> memstore/hfile and WAL which cause the slave cluster has more data than the
> master cluster.
> The simplified write path looks like:
> 1. insert record into Memstore
> 2. write record to WAL
> 3. sync WAL
> 4. rollback Memstore if 3 fails
> It's possible that the HDFS sync RPC call fails, but the data is already
> (may partially) transported to the DNs which finally get persisted. As a
> result, the handler will rollback the Memstore and the later flushed HFile
> will also skip this record.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
