shibd Aborted, No Error in Logs

[Cantor, Scott]

I can promise after all this time and from running SPs for months on end that there's no deadlock or internal issue that would cause a crash. A bug, sure, but that would show up in a stack crash.

If it dies in select(), something else is terminating it.

If it's resource exhaustion, then you could be causing too many sockets to get opened by using Apache prefork. Absent prefork, there's no reason it would open enough of them to be a problem, even heavily loaded Apache using worker or event doesn't result in crazy numbers of threads or sockets in my experience.

You should have the thread count there from the core, and that would indicate roughly how many sockets were open, though not one to one.

The SP's major problem is that while it scales to a point, it could never scale *that* well, and has no effective clustering solution that lets it scale linearly like the IdP can. The new one will address that as it’s a core goal.

The option in recent versions to put data into a cookie and have that recreate sessions across servers is worth looking at to get better linear scaling.

The other step is to reduce the reliance on its sessions. Most heavily loaded systems need to get out of the model of "protect it all" and only use the SP on the front-door, so that a more effective session and clustering model can take over. But you're probably already doing that.

That's far and away all I can say as a non-member support question, though nothing here is actionable from a member-support perspective either to be fair.

-- Scott