At 18.24 12/07/2006, you wrote:
The vpopmail module to authdaemon is might be missing
the vpopmail vclose() function.

I'm giving a fast look, but I don't see either close and open :-( .

If so, then every connection to the authdaemon would open a new mysql connection,
which allocates memory. Then without the vclose the memory never gets released
and eventually runs out of memory.

This is what authdaemon should not do. It open a first time for each child and then use the sames connection always (apart from the eight hours problem). So it must be something after the "connect" phase, something called thousands times.

I'm trying to check, but I'll do better next week.



Ken Jones

tonix (Antonio Nati) wrote:
I don't know if this is a limit of this useful patch (I don't think so), or more likely of the vpopmail library (I feel could be), or yet of courier authdaemon, but after more than ten days of successfull running, I've got this error:
Jul 11 16:24:33 myserver authdaemond: ^GOut of memory (Needed 8164 bytes)
Jul 11 16:24:33 myserver authdaemond: vmysql: sql error[3]:
Jul 11 16:24:33 myserver authdaemond: Attempting to rebuild connection to SQL server
Jul 11 16:24:33 myserver authdaemond: ^GOut of memory (Needed 8164 bytes)
Jul 11 16:24:33 myserver authdaemond: vmysql: connection rebuild failed:
So, I'm adding an authdaemon restart each morning (early) to solve the problem. Note: after stopping and restarting courier authdaemon, 1200 MB of system memory have been released (here I have 4GB of system memory). I feel like vpopmail calls just alloc() memory and never release it, as they have been probably designed for one shot program. When they are used within a "permanent" service, calling them dozen thousands times, they waste memory and cause problems.
Any one else if watching his courier authdaemon growing?
At 21.20 29/06/2006, you wrote:
This patch is in the upcoming 5.4.17 release.


On May 29, 2006, at 7:05 AM, Ron Gage wrote:

I don't know if anyone in here is encountering this problem (yet), but it has
been affecting me for the past few weeks - ever since I upgraded my MySQL
server to 5.0.19. It took quite a bit of digging, but I believe I have found
the problem.

To describe the problem: when you run vpopmail in MySQL mode, with
courier-authdaemond and MySQL v5.0 or later, you will find that for the first 8 hours, everything works just fine, but after 8 hours, nobody will be able to authenticate to the email server and you will see "MySQL server has gone away"
errors in the maillog.

The cause of the problem is that in MySQL 5.0 (and probably some 4.1 releases),
MySQL implements a new timeout definition for connections, a timeout that
ignores traffic. This timeout will shut down the socket thread from the MySQL
side.  The problem is that the client (vchkpw and friends) do not
know/understand about this timeout and socket termination so they continue on in ignorate bliss until they try to send to the socket and find that it's no
longer valid - literally "the server has gone away".

The fix is to simply destroy the internal flags and file handles related to that
socket, rebuild a new one, and try again.

The included patch (inline and attached) implements this fix. Please note that there doesn't appear to be any way at this time to disable the timeout feature
in MySQL.

Please feel free to comment, tear apart, beat up, or otherwise rip to shreads my

Ron Gage
(LPIC1 MCP A+ Net+)
Westland, Michigan

--- vmysql.c~   2006-05-29 10:17:20.000000000 -0400
+++ vmysql.c    2006-05-29 10:17:20.000000000 -0400
@@ -465,7 +465,31 @@
     if (mysql_query(&mysql_read,SqlBufRead)) {
         fprintf(stderr, "vmysql: sql error[3]: %s\n",
-        return(NULL);
+ /* Ron Gage - May 29, 2006 - With newer versions of MySQL, there is
such a thing
+ as a connection timeout regardless of activity. By default under MySQL
5, this
+ timeout is 28800 seconds (8 hours). If your vpopmail system runs fine
for the
+ first 8 hours, then stops authenticating, this timeout is your problem
+        under authdaemond).
+ What this code does is when an error is encountered, it first tries to
drop and
+ rebuild a connection to the SQL server and tries again. If this second
+ fails, then something other than the connection timeout is the problem.
 This fix
+ need to be implemented in other places but in my setup (Slackware 10.2,
+ vpopmail, courier-authdaemond, courier-imapd and a few others), this is
always where
+        the auth attempt died with a "SQL server has gone away" error.
+        */
+ fprintf(stderr, "Attempting to rebuild connection to SQL server\n");
+        vclose();
+        verrori = 0;
+        if ( (err=vauth_open_read()) != 0 ) {
+          verrori = err;
+          return(NULL);
+        }
+        if (mysql_query(&mysql_read, SqlBufRead)) {
+          fprintf (stderr, "vmysql: connection rebuild failed: %s\n",
+          return(NULL);
+        }

     if (!(res_read = mysql_store_result(&mysql_read))) {

This message was sent using IMP, the Internet Messaging Program.

Reply via email to