[jira] [Work logged] (HIVE-21109) Stats replication for ACID tables.

["ASF GitHub Bot (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HIVE-21109?focusedWorklogId=221109&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-221109 ]

ASF GitHub Bot logged work on HIVE-21109:
-----------------------------------------

                Author: ASF GitHub Bot
            Created on: 01/Apr/19 06:30
            Start Date: 01/Apr/19 06:30
    Worklog Time Spent: 10m 
      Work Description: ashutosh-bapat commented on pull request #579: HIVE-21109 : Support stats replication for ACID tables.
URL: https://github.com/apache/hive/pull/579#discussion_r270728648
 
 

 ##########
 File path: ql/src/java/org/apache/hadoop/hive/ql/exec/ColumnStatsUpdateTask.java
 ##########
 @@ -297,21 +303,34 @@ private ColumnStatisticsDesc getColumnStatsDesc(String dbName,
 
   private int persistColumnStats(Hive db) throws HiveException, MetaException, IOException {
     ColumnStatistics colStats = constructColumnStatsFromInput();
-    ColumnStatisticsDesc colStatsDesc = colStats.getStatsDesc();
-    // We do not support stats replication for a transactional table yet. If we are converting
-    // a non-transactional table to a transactional table during replication, we might get
-    // column statistics but we shouldn't update those.
-    if (work.getColStats() != null &&
-        AcidUtils.isTransactionalTable(getHive().getTable(colStatsDesc.getDbName(),
-                                                          colStatsDesc.getTableName()))) {
-      LOG.debug("Skipped updating column stats for table " +
-                TableName.getDbTable(colStatsDesc.getDbName(), colStatsDesc.getTableName()) +
-                " because it is converted to a transactional table during replication.");
-      return 0;
-    }
-
     SetPartitionsStatsRequest request =
             new SetPartitionsStatsRequest(Collections.singletonList(colStats));
+
+    // Set writeId and validWriteId list for replicated statistics.
+    if (work.getColStats() != null) {
+      String dbName = colStats.getStatsDesc().getDbName();
+      String tblName = colStats.getStatsDesc().getTableName();
+      Table tbl = db.getTable(dbName, tblName);
+      long writeId = work.getWriteId();
+      // If it's a transactional table on source and target, we will get a valid writeId
+      // associated with it. Otherwise it's a non-transactional table on source migrated to a
+      // transactional table on target, we need to craft a valid writeId here.
+      if (AcidUtils.isTransactionalTable(tbl)) {
+        ValidWriteIdList writeIds;
+        if (writeId <= 0) {
 
 Review comment:
   We can not set writeId in the ColumnStatsUpdateWork because the writeId for migration is available only after a transaction is opened for migration, which doesn't happen at the load time (when the work is created). Going by the gist of your suggestion, I have set a flag in work to indicate that the writeId should be the one used for migration.
 
----------------------------------------------------------------
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:
users@infra.apache.org


Issue Time Tracking
-------------------

    Worklog Id:     (was: 221109)
    Time Spent: 8h 40m  (was: 8.5h)

> Stats replication for ACID tables.
> ----------------------------------
>
>                 Key: HIVE-21109
>                 URL: https://issues.apache.org/jira/browse/HIVE-21109
>             Project: Hive
>          Issue Type: Sub-task
>            Reporter: Ashutosh Bapat
>            Assignee: Ashutosh Bapat
>            Priority: Major
>              Labels: pull-request-available
>         Attachments: HIVE-21109.01.patch, HIVE-21109.02.patch, HIVE-21109.03.patch, HIVE-21109.04.patch, HIVE-21109.05.patch, HIVE-21109.06.patch, HIVE-21109.07.patch, HIVE-21109.08.patch
>
>          Time Spent: 8h 40m
>  Remaining Estimate: 0h
>
> Transactional tables require a writeID associated with the stats update. This writeId needs to be in sync with the writeId on the source and hence needs to be replicated from the source.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)