You could accuse me of thinking the same - first stuck connection 20/06 @ 16:26, since then I have 19 in total.

Hello all...

Something else for you to try, not sure if this has anything to do with the stuck connections, but worth a try.

The other day I decided for no reason at all to uncheck the "Enable Activity Log" under the Activity Log tab. Since then I don't seem to of had any stuck connections?

Again this may be totally unrelated but thought I'd let you know so you can try it (if you don't already have the activity log disabled) to see if it makes any difference.

Regards,
Lyndon.

Everyone,

When a connection is disconnected, the following occurs, in order, one right after the other:

1. The "Current inbound connections" counter is decreased.
2. The IP is added to the blacklist cache if it is spam
3. The "Disconnect" log entry is added to the activity logfile
4. The connection is removed from the Connections tab.

What we're seing in the logs we've received is that, for the connections that "appear" stuck, only events (1) and (2) happen. On the other hand, (3) and (4) are never executed, so the disconnect event is never logged to the logfile (and on the screen), and the connection is not removed from the list. Lyndon, you were right on the money on this one i believe, as I do think it is indeed tied to the activity log.
We've narrowed this down a lot thanks to all your help. Now that we know what happens, we're trying to figure out "why" and how to prevent it. We may have a new build later today.

I would say that the issue has been fixed with that .578 build.   All my stuck connections and very high CPU usage has stopped.   My servers are running back to normal.

Thanks, Roberto for the quick fix.

Dan

Everything is possible, especially since we were never able to replicate this behavior to begin with. I think we know "where" this was occurring in the code, but not "why".
...and without knowing the "why" it's very hard to fix the issue, as we do not know what the issue is...!

Are the stuck connections still present in 578? If you still have the two ini files that exibit the two different behaviors, I'd like to compare them to see if there's any clues.

R,

Yes, the stuck connections were still present with 578 until I replaced the ini file with spamserver4's ini file.

Dan

 

Today, I have noticed that I always have 2+ incoming connections (from the status bar) however, when I look at the connections screen, those first 2 are not shown.   For example, when no connections show on the grid, the counter says 2.  When I have 5 active connections on the grid, the counter shows 7.

Any way that plays into your current thoughts with this stuck connections issue?

I am running the 3.0.2.578.

jerbo128

Mine seems to be +9

LogSat wrote: thanks for pointing that out, we'll look into that aspect right now. As we're still not able to replicate this, we're open to any suggestions / hints you all have.

I thought I would at least share some information with you guys, perhaps
related or perhaps not... since you have not cracked the case on this issue
yet.

I had spamfilter running for some time, no issues.

Then I started implementing REGEX filter of my own creation.

All of a sudden I started having issues much like everyone is describing
here. Stuck connections, and in my case 100% CPU usage as well.
Obviously the number of incoming connections would start to build up.

The problem could only be sorted out by shutting things down and
bringing it all back up again.

The problem went on for days and days, I was getting quite stressed.
Eventually I removed the REGEX filters I had put in place and bang the
problem was gone.

For fun I put the REGEX back online, and the problem came back right
away.

At that point I just pulled the REGEX stuff out and have not bothered with
them again.

The server would process thousands of messages and then start acting
up when the regex and some combination of email happened.

Something to consider anyway, just try removing the regex filters and see
if it goes away or not.

This also explains why not everyone has the problem, it may only show
itself with some combination of regex filters.

The stuck connections on the connections tab problem has appeared to be solved, with us anyway.  However, the past couple of months, several times we have noticed the opposite problem as before.  The connections tab shows nothing, however the "current inbound connections" counter shows that there are incoming connections.  I have watched the counter / connections tab for quite a period of time, and it appears that the counter does not ever correct without a restart of the service.  Some times these connections appear right after a restart, other times the service has been running for a week or more before they show up.  I am running 3.1.3.597

Jerbo128

My stuck connections is still going full bore. We have been having this problem for a few weeks and only a restart fixes it. We finally wrote a script to monitor the log file and alert us when we start getting 'Too Many Connections' we know that the number of connection has hit 40 (our setting in config for max connections) and its time to restart SF because of no free connections  If we look in SF connections we see 'stuck' connection about 30-40 connections. Eventually mail stops flowing because there are 0 connections freed. Talk about po'd customers. Probably have to find a way to automatically restart the service since this can happen any time. Was hoping this issue would be resolved by now but I guess not.

After restart if we perioically check the connections we see the condition you describing, connections page showing less then the number of connections listed at the bottom of the screen in the status area.

We removed all out regex filters per someones suggestion on this forum but it has made no difference. Some type of spam is triggering this but I don't see any way of figuring out what it is.

Are other still seeing these issues or is it maybe my environment.

Running 3.1.3.597

Edited by WebGuyz

Everyone,

While we've never been able to reproduce the problem, we *may* have found a possible way to avoid it. We've been working a bit in the dark as we could only guess what could be happening. Build 3.1.3.600 (containing the possible fix) was being tested internally to see if there were any unwanted side effects.

As there have not been any so far, and as apparently there are several of you experiencing this symptom, we've just released this verison in the registered user area.

We'd appreciate any feedback (yeah/nea) to let us know how it works.

I installed .600 this weekend and it looked optimistic Saturday and Sunday but his morning I'm 6 connections short and I'm sure it will climb as the traffic drags on.

Update. We've made another attempt to change the code that counts the connections, and uploaded build 601 in the registered user area. Initial tests in an installation that hade the "stuck connection" issue look promising.