RE: Intermittent CVS file corruption with 1.11.5

[Joseph Lung]

I'm pretty sure that the corruption occurred when 1.11.5 was in place
because the timestamp of the files with corruption (including all
problem files identified by check_cvs) indicate modification after 1.11.5
was installed.

(clarification: the server is Redhat 6.2)

Thanks.  We'll investigate your suggestions.


-----Original Message-----
From: Larry Jones [mailto:lawrence.jones@eds.com]
Sent: Tuesday, March 11, 2003 9:33 AM
To: Joseph Lung
Cc: bug-cvs@gnu.org
Subject: Re: Intermittent CVS file corruption with 1.11.5


Joseph Lung writes [in very long lines]:
> 

> We upgraded the CVS server from 1.10.8 to 1.11.5 and then started
> encountering intermittent file corruption.

Are you sure that's when the corruption started occurring and not just
when you started noticing it?  1.11.5 does a lot more integrity checking
than 1.10.8 did.  I suggest you run the contrib/check_cvs script on your
CVS server if you haven't already to ensure that you don't have more
corrupted files that you don't know about.

>    Server:  Linux 6.2
> 
>    Repository:  NFS mounted from a NetApps 6.2R2
>       (the only place this file system is mounted from is the CVS Server 
> machine.)

Using an NFS-mounted repository is asking for trouble, particularly when
the NFS client and server systems are different, and even more so when
the NFS client system is a Linux system.  I strongly urge you to move
the repository to a locally mounted disk if at all possible.  If not,
you might try commenting out the definition of HAVE_MMAP in your
config.h file and rebuilding CVS -- it may be that there is some
conflict with mmap() that doesn't occur with normal file I/O (CVS 1.10.8
didn't use mmap).  If that turns out to be the problem, you need to
report it to your Linux supplier.

-Larry Jones

Even if lives DID hang in the balance, it would depend on whose they were.
-- Calvin