[
https://issues.apache.org/jira/browse/HBASE-2667?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
stack updated HBASE-2667:
-------------------------
Attachment: 2667.txt
The test checked sequenceid ordering in the produced split. They were being
written in the reverse order -- newest first. The reconstruction log over in
HStore expects to read the edits oldest through to newest so changed how we
make the recovered edits files.
Here is the change:
{code}
@@ -1552,10 +1552,11 @@ public class HLog implements HConstants, Syncable {
try {
int editsCount = 0;
WriterAndPath wap = logWriters.get(region);
- for (ListIterator<Entry> iterator = entries.listIterator();
- iterator.hasNext();) {
- Entry logEntry = iterator.next();
-
+ // We put edits onto the Stack ordered oldest sequence id to newest.
+ // Pop them off starting with the oldest.
+ for (ListIterator<Entry> iterator =
entries.listIterator(entries.size());
+ iterator.hasPrevious();) {
+ Entry logEntry = iterator.previous();
{code}
> TestHLog.testSplit failing in trunk
> -----------------------------------
>
> Key: HBASE-2667
> URL: https://issues.apache.org/jira/browse/HBASE-2667
> Project: HBase
> Issue Type: Bug
> Reporter: stack
> Assignee: stack
> Attachments: 2667-prelim.txt, 2667.txt
>
>
> Build is broke because this test won't pass.
> Seems to be two issues at least:
> 1. we need to change the test so it keeps logs in one subdir, archives in
> another and split products in yet another as per regionserver. New split
> code does cleanup of old logs dir if successful. In this test it was
> removing the created splits so when verify went to run, the log no longer
> existed
> 2. The verify of splits tests that sequence numbers are in order. They don't
> seem to be (the assertion failure was masked by a close reader exception).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
