On Fri, 2006-09-29 at 14:10 -0400, Bill Boebel wrote:

On Fri, September 29, 2006 11:46 am, Dominic Marks <dom@helenmarks.co.uk> said:

...

...

Actually, I found in the archives that I should increase login_process_size if I'm not running with login_process_per_connection. I have done this and that seems to have taken care of the malloc failure. However I am still getting a ton of these every few seconds:

Sep 29 11:30:06 director4 dovecot: imap-login: SSL_read() syscall failed: EOF [70.171.61.190] Sep 29 11:30:06 director4 dovecot: imap-login: SSL_read() syscall failed: EOF [24.236.204.154] Sep 29 11:30:06 director4 dovecot: imap-login: SSL_read() syscall failed: EOF [70.171.61.190] Sep 29 11:30:08 director4 dovecot: imap-login: SSL_read() syscall failed: EOF [65.242.19.3]

These are client disconnects, I believe.

Yep, all of my customers who had IMAP SSL problems are reporting that the problem is fixed now that I have increased login_process_size to 64 MB. In just two hours though the imap-login process is up to 33 MB, and continuing to grow. Is that normal? This particular server handles 1200 IMAP connections and 800 IMAPS connections concurrently on average.

I don't know how much memory each SSL connection takes, but I'd suggest increasing the memory limit even higher, maybe even 256MB. If it continues to grow up to it then I guess there's a memory leak somewhere.

The EOF errors are pointless. It just means that the other side closed the connection without sending a proper SSL BYE message. I don't think verbose_ssl is all that useful for anything.

As for the "Bad file descriptor" errors, that is a bug.. How often do you see them? I guess I'll anyway look at the code if I can figure out why it could happen.