Bug #54785 Cluster fails on large ALTER TABLESPACE ADD DATAFILE Windows
Submitted: 24 Jun 2010 17:55 Modified: 26 Dec 2010 15:55
Reporter: Steven Davis Email Updates:
Status: Verified Impact on me:
None 
Category:MySQL Cluster: Disk Data Severity:S2 (Serious)
Version:mysql-5.1-telco-7.1 OS:Microsoft Windows (2008 Server Std x64)
Assigned to: CPU Architecture:Any
Tags: mysql-5.1.44 ndb-7.1.4
Triage: Triaged: D2 (Serious) / R6 (Needs Assessment) / E6 (Needs Assessment)

[24 Jun 2010 17:55] Steven Davis
Description:
Adding 10 gigabytes of datafiles to a tablespace cause the data nodes to fail.
Reporting adding the 10 gigabytes as two 5 GB files.

mysql> alter TABLESPACE TestNDB_TS ADD DATAFILE 'TestNDBTS_2.dat' INITIAL_SIZE=5
G ENGINE=NDBCLUSTER;
ERROR 1533 (HY000): Failed to alter:  CREATE DATAFILE

How to repeat:
Nodes: 1 mgm, 2 mysqld on hardware as 2 ndbd
Data Node 15 drive has 17,374,150,656 bytes free
Data Node 17 drive has 22,499,024,896 bytes free
LOGFILE GROUP ExplicitNDB_LogGroup
TABLESPACE TestNDB_TS uses LOGFILE GROUP ExplicitNDB_LogGroup
TestNDB_TS has no data files at start of test.
mysql> alter TABLESPACE TestNDB_TS ADD DATAFILE 'TestNDBTS_1.dat' INITIAL_SIZE=5
G ENGINE=NDBCLUSTER;
Query OK, 0 rows affected (3 min 25.44 sec)
mysql> alter TABLESPACE TestNDB_TS ADD DATAFILE 'TestNDBTS_2.dat' INITIAL_SIZE=5
G ENGINE=NDBCLUSTER;
Both data nodes get very busy
The non-master data node gets shutdown by the arbiter, then the master data node gets a Windows has stopped working dialog box.
[25 Jun 2010 15:04] Hartmut Holzgraefe
Both data nodes were missing heartbeat signals, probably due to I/O overload while initializing the table space files ...

Combining several smaller data files into a larger table space does indeed look like the best workaround for now