[ https://issues.apache.org/jira/browse/HBASE-21838?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16761166#comment-16761166 ]
Sergey Shelukhin commented on HBASE-21838: ------------------------------------------ I'm not familiar with replication enough. There's a comment says "cells have already been decoded". As far as I see when splitting WAL, cells are for the most part not really decoded and are written as is from source to destination, as byte arrays (which is a good thing, less overhead). The only few places where parts of cells are decoded can be seen in the parent JIRA - we check column family to filter meta cells and during append... is there code in replication that does something similar, or more? Otherwise LGTM. > Create a special ReplicationEndpoint just for verifying the WAL entries are > fine > -------------------------------------------------------------------------------- > > Key: HBASE-21838 > URL: https://issues.apache.org/jira/browse/HBASE-21838 > Project: HBase > Issue Type: Sub-task > Reporter: Duo Zhang > Assignee: Duo Zhang > Priority: Major > Fix For: 3.0.0, 2.2.0, 2.3.0 > > Attachments: HBASE-21838.patch > > > This is a missing part in our ITBLL. We can config a dummy replication > endpoint which replicates nothing but only reads and verifies that all the > wal files are readable and the entries are all fine. > With this I think it will be much easier to catch the problem in the parent > issue. -- This message was sent by Atlassian JIRA (v7.6.3#76005)