[
https://issues.apache.org/jira/browse/HIVE-21213?focusedWorklogId=354252&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-354252
]
ASF GitHub Bot logged work on HIVE-21213:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 05/Dec/19 12:36
Start Date: 05/Dec/19 12:36
Worklog Time Spent: 10m
Work Description: ashutosh-bapat commented on pull request #587:
HIVE-21213 : Acid table bootstrap replication needs to handle directory created
by compaction with txn id
URL: https://github.com/apache/hive/pull/587#discussion_r354277500
##########
File path: ql/src/java/org/apache/hadoop/hive/ql/parse/repl/CopyUtils.java
##########
@@ -463,7 +464,29 @@ public static Path
getCopyDestination(ReplChangeManager.FileInfo fileInfo, Path
String[] subDirs = fileInfo.getSubDir().split(Path.SEPARATOR);
Path destination = destRoot;
for (String subDir: subDirs) {
- destination = new Path(destination, subDir);
+ // If the directory is created by compactor, then the directory will
have the transaction id also.
+ // In case of replication, the same txn id can not be used at target, as
the txn with same id might be a
+ // aborted or live txn at target.
+ // In case of bootstrap load, we copy only the committed data, so the
directory with only write id
+ // can be created. The validity txn id can be removed from the directory
name.
+ // TODO : Support for incremental load flow. This can be done once
replication of compaction is decided.
+ if (AcidUtils.getVisibilityTxnId(subDir) > 0) {
+ if (subDir.startsWith(AcidUtils.BASE_PREFIX)) {
+ AcidUtils.ParsedBase pb = AcidUtils.ParsedBase.parseBase(new
Path(subDir));
+ destination = new Path(destination,
AcidUtils.baseDir(pb.getWriteId()));
Review comment:
Looks like, we are removing the transaction id from the compactor and just
leaving writeId there. Are we replicating the corresponding transaction? How do
we know whether this writeId is visible. I am assuming that a failed compaction
will leave a directory behind but its writeId won't be visible.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
Issue Time Tracking
-------------------
Worklog Id: (was: 354252)
Time Spent: 0.5h (was: 20m)
> Acid table bootstrap replication needs to handle directory created by
> compaction with txn id
> --------------------------------------------------------------------------------------------
>
> Key: HIVE-21213
> URL: https://issues.apache.org/jira/browse/HIVE-21213
> Project: Hive
> Issue Type: Bug
> Components: Hive, HiveServer2, repl
> Reporter: mahesh kumar behera
> Assignee: mahesh kumar behera
> Priority: Major
> Labels: pull-request-available
> Attachments: HIVE-21213.01.patch, HIVE-21213.02.patch,
> HIVE-21213.03.patch
>
> Time Spent: 0.5h
> Remaining Estimate: 0h
>
> The current implementation of compaction uses the txn id in the directory
> name. This is used to isolate the queries from reading the directory until
> compaction has finished and to avoid the compactor marking used earlier. In
> case of replication, during bootstrap , directory is copied as it is with the
> same name from source to destination cluster. But the directory created by
> compaction with txn id can not be copied as the txn list at target may be
> different from source. The txn id which is valid at source may be an aborted
> txn at target. So conversion logic is required to create a new directory with
> valid txn at target and dump the data to the newly created directory.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
