Bug #47832 MT DD crash during SR in Dbtup::disk_page_free

• Submitted: 5 Oct 2009 10:45
• Modified: 2 Nov 2009 21:16
• Reporter: Pekka Nousiainen
• Status: Closed
• Category: MySQL Cluster: Disk Data
• Severity: S2 (Serious)
• Version: mysql-5.1-telco-7.0
• OS: Any
• Assigned to: Pekka Nousiainen
• CPU Architecture: Any

[5 Oct 2009 10:45] Pekka Nousiainen

Description:
In MT DD it is still possible to get a crash
during System Restart.  Current line# 1197.

DbtupDiskAlloc.cpp
disk_page_free
  const Uint32 *src= ((Fix_page*)pagePtr.p)->get_ptr(page_idx, 0);
> ndbassert(* (src + 1) != Tup_fixsize_page::FREE_RECORD);
  lsn= disk_page_undo_free(pagePtr.p, key,

Other defects with similar symptoms were fixed under bug#41915
(which is closed).  This bug# takes over the rest.

How to repeat:
testSystemRestart -v -n SR_DD_2b

[19 Oct 2009 14:06] Pekka Nousiainen

Above crash is caused (at least) by LCP of fragment
which has become empty (scan pos was not initialized).
So a better test case was:

testSystemRestart -n SR_DD_2b_LCP -r 1 D1

[20 Oct 2009 15:12] 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/87488

3023 Pekka Nousiainen	2009-10-20
      bug#47832 01_lcp.diff
      Fix un-initialized scan position of LCP scan of empty fragment and
      make the scan state a bit more explicit.

[2 Nov 2009 19:58] Jonas Oreland

pushed to 6.2.19, 6.3.28 and 7.0.9

[2 Nov 2009 21:16] Jon Stephens

Documented bugfix in the NDB-6.2.19, 6.3.28, and 7.0.9 changelogs, as follows:

        A local checkpoint of an empty fragment could cause a data node
        crash.

Closed.

[3 Nov 2009 7:15] Jon Stephens

Per discussion with Pekka, revised changelog entry to read

        A local checkpoint of an empty fragment could cause a crash
        during a system restart which was based on that LCP.