You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by "Lee, Kevin" <kd...@qualcomm.com> on 2004/01/22 22:47:08 UTC

RE: test failure on .36

This appears to be a problem due to running on a multiprocessor machine.
The errors are intermittent due to task scheduling.  SVN attempts to
change a file to read-only, but the file does not exist yet because a
different thread (or process?) has not completed creating the file yet.

 

I hacked a simple change to the svn_is_set_file_read_only (io.c file)
where it checks to see if apr_file_attrs_set returns an error, it goes
to sleep for a second and then tries again.  This is only a hack, but it
makes the regression tests all pass now for both file and http access
whereas before I would get random failures.

 

Is this something that needs to be fixed?  SVN seems to function
correctly even though it reports the error.  Are there other potential
problems when running on a multi-proc machine?

 

Thanks!  Kevin

 

-----Original Message-----
From: Lee, Kevin [mailto:kdlee@qualcomm.com] 
Sent: Tuesday, January 20, 2004 9:10 AM
To: users@subversion.tigris.org
Subject: test failure on .36

 

I compiled 0.36 from source on redhat 9.  Everything seems to be working
fine, but when I run the tests (make check from the top), I get test
failures of the type:

 

CMD: svn "checkout" "--username" "jrandom" "--password" "rayjandom"
"file:///home/kdlee/workarea/subversion0.36.0/subversion/tests/clients/c
mdline/repositories/externals_tests-4"
"working_copies/externals_tests-4" "--config-dir"
"/home/kdlee/workarea/subversion-0.36.0/subversion/tests/clients/cmdline
/local_tmp/config" <TIME = 4.576458>

 

EXPECTED STDERR:

ACTUAL STDERR:

subversion/libsvn_subr/io.c:723: (apr_err=2)

svn: Can't set file 'working_copies/externals_tests-4/A/D/.svn/entries'
read-only: No such file or directory

EXCEPTION: SVNLineUnequal

FAIL:  externals_tests.py 4: update change to an unmodified external
module

 

The weird thing is if I run the tests again, I get the same kind of
failure, but in a different test.  It always has trouble setting one of
the entries files to read-only because it thinks it does not exist.  But
each time it is a different test and file that fails.

 

Is it possible that I screwed something up when I compiled the code?  I
just wanted to make sure that this is not a real problem.

 

Thanks!  Kevin


Re: test failure on .36

Posted by Mark Benedetto King <mb...@lowlatency.com>.
On Thu, Jan 22, 2004 at 02:47:08PM -0800, Lee, Kevin wrote:
> This appears to be a problem due to running on a multiprocessor machine.
> The errors are intermittent due to task scheduling.  SVN attempts to
> change a file to read-only, but the file does not exist yet because a
> different thread (or process?) has not completed creating the file yet.
> 
>  
> 
> I hacked a simple change to the svn_is_set_file_read_only (io.c file)
> where it checks to see if apr_file_attrs_set returns an error, it goes
> to sleep for a second and then tries again.  This is only a hack, but it
> makes the regression tests all pass now for both file and http access
> whereas before I would get random failures.
> 
>  
> 
> Is this something that needs to be fixed?  SVN seems to function
> correctly even though it reports the error.  Are there other potential
> problems when running on a multi-proc machine?
> 

What OS and filesystem is this occuring on?

--ben

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: test failure on .36

Posted by Branko Čibej <br...@xbc.nu>.
Lee, Kevin wrote:

> This appears to be a problem due to running on a multiprocessor
> machine.  The errors are intermittent due to task scheduling.  SVN
> attempts to change a file to read-only, but the file does not exist
> yet because a different thread (or process?) has not completed
> creating the file yet.
>
I've been developing, running and testing SVN on a 2-cpu WinXP box for
quite a while now, and haven't seen anything similar. Besides, the
client and the tests run in a single thread; there can be no races there.

(Note that there can be a race between the client and the server, and
these show up on Windows in the test cleanup processing; we have code in
the test suite that "fixes" that.)

-- 
Brane Čibej   <br...@xbc.nu>   http://www.xbc.nu/brane/

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: test failure on .36

Posted by Branko Čibej <br...@xbc.nu>.
Lee, Kevin wrote:

> This appears to be a problem due to running on a multiprocessor
> machine.  The errors are intermittent due to task scheduling.  SVN
> attempts to change a file to read-only, but the file does not exist
> yet because a different thread (or process?) has not completed
> creating the file yet.
>
I've been developing, running and testing SVN on a 2-cpu WinXP box for
quite a while now, and haven't seen anything similar. Besides, the
client and the tests run in a single thread; there can be no races there.

(Note that there can be a race between the client and the server, and
these show up on Windows in the test cleanup processing; we have code in
the test suite that "fixes" that.)

-- 
Brane Čibej   <br...@xbc.nu>   http://www.xbc.nu/brane/

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: test failure on .36

Posted by Mark Benedetto King <mb...@lowlatency.com>.
On Thu, Jan 22, 2004 at 02:47:08PM -0800, Lee, Kevin wrote:
> This appears to be a problem due to running on a multiprocessor machine.
> The errors are intermittent due to task scheduling.  SVN attempts to
> change a file to read-only, but the file does not exist yet because a
> different thread (or process?) has not completed creating the file yet.
> 
>  
> 
> I hacked a simple change to the svn_is_set_file_read_only (io.c file)
> where it checks to see if apr_file_attrs_set returns an error, it goes
> to sleep for a second and then tries again.  This is only a hack, but it
> makes the regression tests all pass now for both file and http access
> whereas before I would get random failures.
> 
>  
> 
> Is this something that needs to be fixed?  SVN seems to function
> correctly even though it reports the error.  Are there other potential
> problems when running on a multi-proc machine?
> 

What OS and filesystem is this occuring on?

--ben

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org