title | description | services | documentationcenter | author | manager | ms.service | ms.custom | ms.devlang | ms.topic | ms.tgt_pltfrm | ms.workload | ms.date | ms.author |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Common issues during cluster upgrade | Microsoft Docs |
Diagnosing and fixing issue during cluster upgrade |
hdinsight |
duoxu |
ashitg |
hdinsight |
hdinsightactive |
na |
article |
na |
big-data |
11/10/2017 |
duoxu |
We have a customer upgrading their cluster from 3.4 to 3.5 (HDP 2.4 to HDP 2.5), there is one view contains no data after new cluster was created.
i see the view in sqlline... i see no data in the view in sqlline though
0: jdbc:phoenix:zk1-profil> select count(*) from table1_view;
+-----------+
| COUNT(1) |
+-----------+
| 1 |
+-----------+
however in the table that the view is backed by...
0: jdbc:phoenix:zk1-profil> select count(*) from table1;
+-----------+
| COUNT(1) |
+-----------+
| 1823616 |
+-----------+
HBCK returned fine, so there are no offline regions.
The root cause is actually that index table for views (all indexes for view are stored in a single physical HBase table) is truncated during upgrade and it’s expected that user will recreate them.
In phoenix-core/src/main/java/org/apache/phoenix/util/UpgradeUtil.java, there are comments that
// Disable the view index and truncate the underlying hbase table.
// Users would need to rebuild the view indexes.
Users should drop and recreate all indexes after upgrade.