[HAWQ-1777] Database reset due to transaction ID assignment problem

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: Core
• Assignee: Ruilong Huo
• Reporter: LyndonYang
• Created: 13/Jan/21 06:34
• Updated: 13/Jan/21 06:34

Description

AtAbort_AppendOnly

        while ((pending = (AppendOnlyHashEntryPendingCleanup *) hash_seq_search(&status)) != NULL)

        {

            if (pending->nestedLevel >= currentLevel)

            {

                /*

                 * the sub transaction which include the truncate/alter operator has been aborted,

                 * so its aoentry is not staled anymore.

                 */

 

                aoentry = AORelLookupHashEntry(pending->relid);

 

                if (aoentry)

               

{                     Insist(aoentry->staleTid == CurrentXid);                     aoentry->staleTid = InvalidTransactionId; //clear flag stale                 }

 

                hash_search(AppendOnlyHashEntryPendingDeleteCleanup,

                        (void *) &pending->relid,

                        HASH_REMOVE,

                        &found);

            }

        }

If an error occurs during execution, aoentry->staleTid is not assigned，abort will failed and 

AOSegFileLock cannot free, which maybe cause when abort the entire transaction cannot get the lock