Bug #43632 MT+DD use of uninitalized table object during UNDO-execution, if table dropped

• Submitted: 13 Mar 2009 13:46
• Modified: 15 Apr 2009 2:28
• Reporter: Jonas Oreland
• Status: Closed
• Category: MySQL Cluster: Cluster (NDB) storage engine
• Severity: S3 (Non-critical)
• Version: 6.4
• OS: Any
• Assigned to: Jonas Oreland
• CPU Architecture: Any

[13 Mar 2009 13:46] Jonas Oreland

Description:
DbtupProxy uses getInstanceKey, wo/ knowing if table is dropped or not
This leads to crash in DIH

2009-03-13 14:42:32 [ndbd] INFO     -- dbdih/DbdihMain.cpp
2009-03-13 14:42:32 [ndbd] INFO     -- DBDIH (Line: 8419) 0x00000002
2009-03-13 14:42:32 [ndbd] INFO     -- Error handler startup shutting down system
2009-03-13 14:42:32 [ndbd] INFO     -- Error handler shutdown completed - aborting

How to repeat:
run follwing on 1-node ndbmtd

create_tab D1 D2
hugoLoad -r 1000 D1; hugoLoad -r 1000 D2

echo "checkpoint"
ndb_mgm -e "all dump 7099"
sleep 10

ndb_delete_all D1 D2
ndb_drop_table D1 D2

ndb_mgm -e "all restart -a"

Suggested fix:
why not store instance key on diskpage ?
if that deemed to important,
dbtup-proxy must keep state of which tables are created/dropped
(which it can since it's proxying create-drop-table)

[13 Mar 2009 13:48] Jonas Oreland

this also affects autotest
testSystemRestart -n basic T1

[16 Mar 2009 17:04] 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/69326

2952 Pekka Nousiainen	2009-03-16
      bug#43632 01.diff
      TUP proxy: for UNDO execution keep track of which tables exist
      modified:
        storage/ndb/src/kernel/blocks/LocalProxy.cpp
        storage/ndb/src/kernel/blocks/LocalProxy.hpp
        storage/ndb/src/kernel/blocks/dblqh/DblqhProxy.cpp
        storage/ndb/src/kernel/blocks/dbtup/DbtupProxy.cpp
        storage/ndb/src/kernel/blocks/dbtup/DbtupProxy.hpp

[16 Mar 2009 17:07] Bugs System

Pushed into 5.1.32-ndb-7.0.4 (revid:[e-mail]-20090316170507-kp3n7um1txenx487) (version source revid:[e-mail]-20090316170507-kp3n7um1txenx487) (merge vers: 5.1.32-ndb-7.0.4) (pib:6)

[19 Mar 2009 6:20] Jon Stephens

Documented bugfix in the NDB-7.0.4 changelog as follows:

        When using multi-threaded data nodes, dropping a Disk Data 
        table followed by a data node restart led to a crash.

[5 Apr 2009 12:44] Jonas Oreland

reopen, cause testSystemRestart -n basic T1 still crashes
with same backtrace

[8 Apr 2009 13:13] Jonas Oreland

new testcase

create_tab D1 D2; reorg_tab -p 4 D1 D2; hugoLoad -r 1000 D1 D2; ndb_mgm -e "all dump 7099"; sleep 10; ndb_drop_table D1 D2; create_tab D1 D2; ndb_mgm -e "all restart -a"

key concepts:
Proxy does not react to create/drop in undo-log, causing it to potentially
make incorrect getInstanceKey calls.

Patch (upcoming) changes this...

[8 Apr 2009 13:45] 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/71652

2886 Jonas Oreland	2009-04-08
      ndb - bug#43632 - make DbtupProxy also react to create/drop table in UNDO log

[8 Apr 2009 14:55] Bugs System

Pushed into 5.1.32-ndb-7.0.5 (revid:[e-mail]-20090408145040-hhbuq1zvc7xj0tx4) (version source revid:[e-mail]-20090408134427-1ch4i8l7odzpmml1) (merge vers: 5.1.32-ndb-7.0.5) (pib:6)

[15 Apr 2009 2:28] Jon Stephens

Also documented in the NDB-7.0.5 changelog, with additional note:

        This fix completes one that was made for this issue in MySQL
        Cluster NDB-7.0.4, which did not rectify the problem in all
        cases.