MySQL Bugs: #26888: func_in fails on many different platforms, but with low probability

Bug #26888	func_in fails on many different platforms, but with low probability
Submitted:	6 Mar 2007 18:09	Modified:	30 Oct 2008 18:57
Reporter:	Daniel Fischer	Email Updates:
Status:	Verified	Impact on me:	None
Category:	MySQL Server: Tests	Severity:	S3 (Non-critical)
Version:	5.0 + 5.1 + 5.5	OS:	Any
Assigned to:		CPU Architecture:	Any

Description:
Test case func_in fails on winx64.

func_in                        [ fail ]
  
Errors are (from /PATH/mysqltest-time) :
mysqltest: At line NNN: query 'select count(*) from (^M
select @str:=concat(@str, @cnt:=@cnt+1, ",") ^M
from t1 A, t1 B, t1 C, t1 D) Z' failed: 126: Incorrect key file for table '7-winbuild\mysql-5.0.37-winx64\mysql-test\var\tmp\#sql_c84_0.MYI'; try to repair it
(the last lines may be the most important ones)
Result from queries before failure can be found in r/func_in.log
  
Stopping All Servers
Restoring snapshot of databases
Resuming Tests  

How to repeat:
Build 5.0.37 on winx64 and run test.

This happens in many release builds, but often was not reported.

Scanning the isolated failures in the archive
(which is incomplete, because for some builds this isolation wasn't done)
I find these occurrences:

5.0.21     Linux/x86(RPM)
5.0.26     Windows-64
5.0.30sp1  Linux/IA64, Linux/s390
5.0.32     Linux/s390
5.0.34     Linux/IA64, Linux/s390
5.0.36     Linux/s390
5.0.37     Windows-64
5.0.42     Linux/x86_64, Linux/s390
5.0.44     Linux/s390
5.0.50     Linux/s390
5.0.50sp1a Linux/s390
5.0.56     Linux/s390
5.0.58     Linux/s390
5.0.62     Windows-64
5.0.64     Windows-64
5.0.66     Windows-64
5.0.68     Linux/s390
5.0.72     Linux/x86_64/icc, Solaris-10/x86, Solaris-10/x86_64, AIX 5.2(32)

5.1.21     OS-X/x86
5.1.22     Linux/x86_64(RPM)
5.1.23     Linux/x86, Linux/x86_64, Solaris-8/Sparc, Solaris-10/x86
5.1.26     Linux/x86_64
5.1.27     Linux/x86_64

So we have it in many releases, on many different platforms (using different compilers), some occurrences in optimized builds, some in debug,
but always with a low probability only (the test passes in the majority of runs).

There is some interesting observation when using
xref on main.func_in:
1. Output pattern "%free space%"
   - ~34 failures with this pattern
   - ~32 of these ~34 are like
     At line 230: query 'select count(*) from (
     select @str:=concat(@str, @cnt:=@cnt+1, ",") 
     from t1 A, t1 B, t1 C, t1 D) Z' failed: 
     20: Disk is full writing '/dev/shm/pbtmp-n_mix-120/#sql_3424_0.MAD'
     (Errcode: 28). Waiting for someone to free space... Retry in 60 secs
2. Output pattern "%line 230%"
   - ~179 failures with this pattern
   - ~32 of these ~179 with "Waiting for someone to free space"
   - ~30 of these ~179 with "2013: Lost connection to MySQL server ..",
            but only Dec 2007 and Maria tree
   - most remaining of the ~179 are with
     126: Incorrect key file for table '/dev/shm/pbtmp-embedded-
           112/#sql_751b_0.MYI'; try to repair it
Could it be that the effect at the beginning of bug report
  At line 230: query 'select count(*) from (.
  ... 
  failed: 126: Incorrect key file for table
  ... 
  try to repair it
is caused by file system full?

Still happens in 5.5.2-m2