Shibboleth Developers

["Scott Cantor" <>]

> The servers are behind a loadbalancer, each of these servers handle
> about 20k unique users a day.

That's higher than ours, but it's in the same ballpark. We have 4 servers in
a cluster (last I knew) and we handle about 60,000 logins a day on that
application. We never restart shibd (only if it crashes, and I used to check
on it periodically to ensure that it wasn't crashing routinely). That
suggests to me that this is a platform bug.

> I've included two graphs (Load Average and Memory Usage) from one of
> these application servers. The shibd process is being restarted 2 times
> a day, at 12u15 and 03u15.
> Today, for example, the daemon was restarted at 12u15 and the memory
> usage has reached 2,3GB.

That's just not my experience at all. In fact, it was the key thing I was
thrilled about when we moved from 1.2 to 1.3 here, it completely solved our
memory issues (particularly once I patched the Xerces bloat).

> > Also, what's your cacheTimeout set to (assuming this is just standard
> > in-memory caching)? Does your transaction log show expired sessions
being
> > purged?
> 
> We use the default memory session cache. The cacheTimeout is reduced to
> 600 seconds.

We do too, but that would not be advisable unless you're just using
Shibboleth at the front door of an application that has its own session
mechanism. Is that the case here? Otherwise you're forcing relogin every 10
minutes. We actually use 5 minutes, I think.

The other thing I asked though...are you seeing the transaction log record
all the purging?

> Are you planning to make SRPMS for the new Xerces library?

When I release my updates, yes, but the spec file inside Xerces now is from
me, it works fine, so anybody can create one. If you want to rebuild 1.3f on
top of 2.8.0, you're also welcome to do it, it will work fine. I just don't
know that it will help.

I really don't know what to tell you. Something's leaking, but anything that
blatant should show up on Windows.

-- Scott