MySQL Bugs: #8774: LOAD DATA FROM MASTER creates errors in logfiles

Bug #8774	LOAD DATA FROM MASTER creates errors in logfiles
Submitted:	24 Feb 2005 11:10	Modified:	29 Mar 2005 12:15
Reporter:	Conrad Micallef	Email Updates:
Status:	Duplicate	Impact on me:	None
Category:	MySQL Server: Replication	Severity:	S2 (Serious)
Version:	4.0.22	OS:	Linux (linux gentoo)
Assigned to:		CPU Architecture:	Any

Description:
I have one database containing a bunch of tables. Database size is around 50MB. I have one master and one slave.
When testing replication everything was working fine. I normally use slave stop;LOAD DATA FROM MASTER;slave start when i simulate replication errors.

When the master has load from other clients, i have very strange symptoms when issuing the LOAD DATA FROM MASTER on the slave, and I get a number of errors in mysqld.log
050224 11:38:37 Delete link points outside datafile at 3656
050224 11:38:37 Delete link points outside datafile at 3748
050224 11:38:37 Delete link points outside datafile at 68

I had particular instances where the replication does not even recover and the only way to recover it was to delete a particular table which was always problematic.

During this problematic table, i stopped the master and did an isamcheck but it revealed nothing. I tried this using the same master and a number of different slaves on different platforms too and the thing is repeatable. This particular table contained HUGE BLOBS and when failing the problem reported what that it did not have enough memory. All blobs I store are around 64KB.

I tried to find a similar bug reported but none to date.

The network connection between master and slave is a 100MB ethernet directly connected via an ethernet switch, no firewalls in between. The LOAD DATA FROM MASTER process takes about 41 seconds to execute.

It seems that data is transferred "corrupted" during the table load.

How to repeat:
To repeat the anomaly on my systems i just have to execute LOAD DATA FROM MASTER on the slave.

Hi,

Thank you for the report, but I wasn't able reproduce it with 4.0.24-debug-log.
Is corruption repeated only if master is updated by other clients during LOAD DATA FROM MASTER? Do this clients work with those tables with BLOBs?

i have 3 users connected to the mysql. Both connect to the master and to the slave.

The writes are sent to the master and the reads are sent to the slave.

The clients use all tables, but the least used ones are the ones with blob.

I modified clients to do all read/write operations on the master and not to use the slave.

when clients are using the tables on the master and i execute LOAD DATA FROM MASTER on the slave the problem happens.

No feedback was provided for this bug for over a month, so it is
being suspended automatically. If you are able to provide the
information that was originally requested, please do so and change
the status of the bug back to "Open".

Hi Conrad,
I'm marking this one as duplicate of your BUG#9125. Feel free to comment here if this is not the case.