linda@cc.brunel.ac.uk (Linda Birmingham) (10/18/89)
I had quite a lot of responses to my nfs in device wait state query.
UKRTOC have just received a reply from RTOC to say that this is
SPR 5077 and there is a fix.
To summarise:
1) Sun have bugs reported as:
(This is a precis of the info I received from Mario Wolczko)
case 1017518 : nfsd's go into DW. Process is blocked waiting for the keep
count on the page it wants to go to zero...but it doesn't get decremented
and never gets to zero.
2 patches available.
case 1017893 : Server problem similar to client problem in bugId 1018954.
Process blocked waiting for an mbuf structure to be released back to NFS.
No patch available at this time.
2) This can happen if swap space gets used up and the pager and swapper are
swapped out.
3) A similar problem with disk intensive applications that shared directories
turned out to be deadlocks waiting for disk resources. This was a
bug in Pyramids OSx directory routines.
4) There was a bug in Suns implementation of the nfsd code that caused
the nfsd's to be swapped out under heavy NFS load.
Claimed to be fixed in SUNOS 4.0.3.
5) Try sending SIGHUP to the inetd. The inetd re-reads the configuration file
and may 'unlock' the nfsd's.
I hope this is of help.
Thanks to everyone that replied.
Linda.
--
Brunel University, Uxbridge, Middlesex, England.
janet: linda@uk.ac.brunel.cc | :-)
uucp:...ukc!cc.brunel!linda | hedrick@geneva.rutgers.edu (Charles Hedrick) (10/25/89)
I didn't see the original question, but since we have seen and fixed nfsd is D wait on both Pyramid and Suns, I thought I'd respond. Pyramid: We saw the problem on the Pyramid under a version early release of 4.1. I seem to recall a report here of a similar problem under 4.4. We moved to 4.4 hoping to fix it. We succeeded. However we are using 4.4c and a very recent set of PTF's: 1003. We suspect that any PTF's within the last few months would be good enough. In general I have to say that I'm very impressed with the reliability of 4.4 so far. SunOS: We saw NFS hangs on Suns under 4.0.1. I've seen reports of similar things under 4.0.3, though I suspect 4.0.3 has fixed at least some of the problems. Unfortunately there are several different causes of NFS hangs. Some result in servers getting nfsd in D state. Others result in clients hanging. There are now PTF's available from Sun. If you have all NFS and file system-related PTF's from Sun installed, I think you'll find that NFS works reliably. Unfortunately I can't give you specific bugid's or other information, since we worked with several people at Sun for months to accumulate a set of fixes that got everything. We're still using 4.0.1, since we haven't gotten 4.0.3 source yet.