[18602] in Athena Bugs
Re: sgi 8.4.17: getpwuid
daemon@ATHENA.MIT.EDU (Robert A Basch)
Sat Dec 9 16:11:51 2000
Message-Id: <200012092111.QAA123851@aupair.mit.edu>
To: Nickolai Zeldovich <kolya@MIT.EDU>
cc: bugs@MIT.EDU, rbasch@MIT.EDU
In-Reply-To: Your message of "Fri, 08 Dec 2000 20:21:25 EST."
<200012090121.UAA07084@pepsi-one.mit.edu>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Sat, 09 Dec 2000 16:11:47 -0500
From: Robert A Basch <rbasch@MIT.EDU>
> > I will report this to SGI, but I would not be optimistic about them
> > fixing it.
Apparently I was unduly pessimistic; the problem had been previously
reported, and is supposed to be fixed in the latest IRIX update (by
having the lookup retry once immediately after getting the EIO error).
> Is it the case that we just need to disable caching for the getpw*()
> lookups? It looks like the intentional 100ms sleep applies to almost
> any other lookup in IRIX as well, which makes, say, "netstat" run a
> bit slow, due to getserv*(), and probably makes various other things
> slower too.
I don't think caching would help netstat much, as I imagine there tends
to be only one failing getservbyport() call for a particular port number.
> Perhaps it would be reasonable to disable caching for just the passwd
> map, and allow caching of other maps? This should be possible by moving
> the timeout attribute settings from /etc/config/nsd.config to the passwd
> map line in /etc/nsswitch.conf, which seems to work on an 8.4.17 SGI.
Actually, I believe we could disable caching for only the passwd.byname
map, and allow it for passwd.byuid, as a way to improve the "ls -l"
problem (even in large directories, files are typically owned by a
small set of users, so it's different from the netstat case). I've been
thinking of trying this for Athena 8.5, and will look into it if we
decide not to bother upgrading to the latest IRIX update. Given our
experience with nsd, any such change should get the extensive testing
that only a full release cycle can offer.
Bob
