[ https://issues.apache.org/jira/browse/HAWQ-471?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15188846#comment-15188846 ]
Ming LI commented on HAWQ-471: ------------------------------ The root cause of this defect is: during recovery pass2, user issues command 'hawq stop cluster -M smart', this command will wait until pass 2 finished, and also it will change the flag in pg_control file, so next time hawq start, it can't find it need recovery pass 3. If we change the command from 'hawq stop cluster -M smart' to 'killall -9 postgres', all problem disappear. The solution is to check recovery during hawq stop. Which is already fixed in https://issues.apache.org/jira/browse/HAWQ-467. So I will close this defect. > Reindex bug > ----------- > > Key: HAWQ-471 > URL: https://issues.apache.org/jira/browse/HAWQ-471 > Project: Apache HAWQ > Issue Type: Bug > Components: Catalog > Reporter: Dong Li > Assignee: Ming LI > > cr_workload=# reindex index pg_class_relname_nsp_index; > ERROR: index "pg_class_relname_nsp_index" contains unexpected zero page at > block 20 > HINT: Please REINDEX it. > The index of pg_class_relname_nsp has problem and error out "index > "pg_class_relname_nsp_index" contains unexpected zero page at block 20". > But if we reindex it, it errors it out again. > So it is a recurse bug and the index can never be fixed. -- This message was sent by Atlassian JIRA (v6.3.4#6332)