Bug #44099 Scan can timeout incorrectly if low selectivity

• Submitted: 5 Apr 2009 20:55
• Modified: 15 Apr 2009 3:27
• Reporter: Jonas Oreland
• Status: Closed
• Category: MySQL Cluster: Cluster (NDB) storage engine
• Severity: S3 (Non-critical)
• Version: *
• OS: Any
• Assigned to: Jonas Oreland
• CPU Architecture: Any

[5 Apr 2009 20:55] Jonas Oreland

Description:
1) deadlock detection is based solely on timeouts
2) if doing a scan, it will report to TC (to avoid timeout)
   on a regular basis, when it is not hanging on a lock
3) ACC-scan had a hard-coded value of checking this once every 100ms
   which mades scan timeout incorrectly if using
   TransactionDeadlockDetectionTimeout <= 100
4) TUP/TUX-scans did not have any such code, making them very easy to timeout.

---

Solution is to move this code from ACC into LQH (which handles ACC/TUP/TUX scans)
and remove the hard-coded value.

How to repeat:
create table with ~10M rows
run SQL-scan statement (with engine_condition_pushdown enabled) that
  will only return 1 (or very few rows)
set TransactionDeadlockDetectionTimeout to a low value.

Suggested fix:
see above

[5 Apr 2009 20:57] Bugs System

A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/71408

2899 Jonas Oreland	2009-04-05
      ndb - bug#44099 - move SCAN_HB handling to LQH, and check it once per row returned from ACC_SCAN

[5 Apr 2009 21:04] Bugs System

Pushed into 5.1.32-ndb-6.2.18 (revid:[e-mail]-20090405205643-d7rplq1dym8gsoea) (version source revid:[e-mail]-20090405205643-d7rplq1dym8gsoea) (merge vers: 5.1.32-ndb-6.2.18) (pib:6)

[5 Apr 2009 21:05] Bugs System

Pushed into 5.1.32-ndb-6.3.24 (revid:[e-mail]-20090405210010-idfu6505qfrnngp3) (version source revid:[e-mail]-20090405210010-idfu6505qfrnngp3) (merge vers: 5.1.32-ndb-6.3.24) (pib:6)

[5 Apr 2009 21:05] Bugs System

Pushed into 5.1.32-ndb-7.0.5 (revid:[e-mail]-20090405210306-0285cjs2pn7wplo0) (version source revid:[e-mail]-20090405210306-0285cjs2pn7wplo0) (merge vers: 5.1.32-ndb-7.0.5) (pib:6)

[15 Apr 2009 3:27] Jon Stephens

Documented bugfix in the NDB-6.2.18, 6.3.24, and 7.0.5 changelogs as follows:

        TransactionDeadlockDetectionTimeout values less than 100 were treated
        as 100. This could cause scans to time out unexpectedly.

Also noted behaviour (and change) as appropriate in 4.1/5.0/5.1 versions of configuration section of Cluster chapter in Manual.