Shibboleth Developers

["Scott Cantor" <>]

> I have a couple of questions:
> - have you ever heard of similar problems? I'm using an ODBC SS and
>   everything has worked fine from april to mid-august. Nothing was
>   touched in the configuration. I remember of André Cruz reporting some
>   memory investigations.

Not enough people have switched to 2.x, so I don't have any reliability data
yet. I do know there was/is a leak in 2.0 that I fixed for 2.1 (I didn't get
a report on it, I just found it eyeballing something).

IIRC, Andre's memory leak issue was with 1.3 and turned out to be a bug in
the Linux RPC library, which isn't used by 2.x.

> - don't you think it's better not to shutdown completely shibd in that
>   case? Maybe the catch() on line 244 of SocketListener.cpp should be
>   "refined" a bit to just try another time the select() or not to exit
>   on memory problems (actually I don't know how this is handled in other
>   select()-using daemons).

I'll review it. Chances are, of course, that once memory's gone lots of
stuff might go wrong anyway, but I don't know whether that code path needs
work or not.

I would have to assume there's a leak involved, either the one I fixed, or
others. Could easily be ODBC-related too, since nobody really uses that
either.

The 1.3 shibd doesn't run out of memory on Windows under much higher load
than what you probably have. Granted, Windows is much better than Unix at
long-lived processes, but I'm talking about extremely high use and a very
low stable memory state.

-- Scott