Atomicorp

I've never had the opportunity to physically test exactly what happens when max_load or instances are exceeded, particularly for smtp_psa

I had assumed that nothing drastic would occur - the sending server would get no connection and try again later, when the system was not under an abnormally high load. Certainly this is what I would want to happen.

But I've just had a chat with another forumite who reports that at least one sender got an immediate bounce when they sent a message during a period when his system was under heavy load and max_load had been exceeded.

I forget the exact nature of the bounce reason but involved a 5.5.0 and something about a socket timeout.

This doesn't seem like what we want to happen, does it?

Now in this particular instance, the server was under very heavy load and all sorts of services were effected, not just email. So the bounce could have been the result of a combination of things as opposed to max_load being exceeded.

Nevertheless, does anybody have any real experience of what happens when max_load or instances are exceeded and could shed some light on the matter? It isn't exactly something that's easy to test without potentially causing unnecessary problems for customers.

faris wrote:Nevertheless, does anybody have any real experience of what happens when max_load or instances are exceeded and could shed some light on the matter? It isn't exactly something that's easy to test without potentially causing unnecessary problems for customers.

Real world experience with these limits here. If a limit is set and reached, the SMTP server will simply not reply (the client will experience this as a time-out). The same as when a mail server is unreachable. The sending MTA should try the next (fallback) MX, or queue and retry later. It should not result in an immediate bounce. That is also our experience, we have never seen it causing bounces.

Do also note that if the limit is reached, your local applications (webmail, mailer scripts) can also not connect to the mail server via SMTP.

prupert wrote:Do also note that if the limit is reached, your local applications (webmail, mailer scripts) can also not connect to the mail server via SMTP.

And those probably fail immediately and don't retry, so maybe that's what you're referring to?

Thanks guys.

The bounce was an external mailserver sending mail to the overloaded server. But given prupert's experience, it would seem that under normal circumstances everything happens as I'd expect it.

I wonder what was going on in this case then. I just can't figure out what would cause an immediate bounce. No matter what was going on, I'd expect the sending server not to assume any kind of error was a permanent rejection.

I do remember issues in the past when clamd was down or unresponsive that caused problems, but I don't remember if they were tempfails or full rejections. Again I would not expect a permanent rejection, but .....

Anyway..thanks the for input both of you. My mind is more at rest now that I know things SHOULD happen as I expect them to happen with instances and max_load, even if they might not always if something else is also going on on an overloaded server:-)