Re: [Evolution] [Fwd: Re: evolution locking up]

[Brian L Scipioni <brian scipioni janusresearch com>]

In case anyone's interested here is what I have found:

My problem started in the first place due to a few runaway processes on
a server (the mail machine also).  The saturated CPUs resulted in very
slow response and NFS timeouts.  The issue is how this is handled by a
multi-threaded volution.  In this case it created a corrupted downloaded
message.

I seems to me that when, for whatever reason, Evolution has trouble
getting a response, it easily leads to either race conditions or
deadlocks.  I can't prove it, but the symptoms are there.  It has
happened to me many times.  Usually I get a deadlock and have to kill
the process - sometimes remotely.  This time I got a race condition
while reading/writing my Inbox file.  It resulted in a single email
repeated with over 3 million lines - a 180 MB message.  When I cat-ed
out the good parts before and after the corrupted message - I was able
to recover all the mail I had lost in the Inbox.  (There was nothing
wrong with the original message on the email server which I had accessed
many times).

It took a long time for me to figure out what was wrong, since whenever
I brought up Evolution, it crashed immediately while trying to read its
folders.  My advice for others in this situation - copy all the email
files to somewhere else and add them back to .evolution/mail/local one
or several at a time until you find the bad one.

The developers might want to validate downloaded message formats to
prevent this.

Also they might want to make thread synchronization more robust to
prevent these race/deadlocks.

Often, after I add/update a Contact in my LDAP server through Evolution,
that thread locks up/dies while the email editor seems to continue on
its merry way as if nothing has happened.  I have to kill the program to
repair my LDAP connection.

thanks,
- 
Brian L Scipioni <brian scipioni janusresearch com>



On Fri, 2007-08-03 at 17:25 +0200, LLLActive GMX Net wrote:
> > Anyone know how to validate/reset the .evolution tree without
> trashing
> > it?
>
> I'd also like to know ...
-- 
Brian L Scipioni <brian scipioni janusresearch com>