On Thu, Jul 27 2017, Phil Kauffman wrote: > On 07/26/2017 11:37 PM, NeilBrown wrote: >> Weird. > I thought so as well. > >> So it seems like the context is different in some way. > It is in the sense that I am now using 1.3.4 and 2.1.1 vanilla on Ubuntu Xenial(provides 1.2.8) without any of debian's patches or niceties. > > >> Your nfs_test_notes.txt doesn't show /etc/export.d getting filled in >> ... maybe that it done automatically somehow... > I copied the files from the original NFS server into /etc/exports.d as part of the VM snapshot I revert back to. My notes say 'mkdir -p /etc/exports.d', but this step is not necessary. > > root@nfsserver:~# ls /etc/exports.d/ | wc -l > 972 > > I only added 3 extra files so that 'nfsclient' would be able to mount from the testing server (nfsserver). > > The ZFS datasets were also precreated: > > root@nfsserver:~# zfs list | wc -l > 6062 > > >> Can you try with unpatches 2.1.1? > I assume you mean without any patches. vanilla 2.1.1 > >> Also maybe provide an strace starting before any attempt to mount >> anything, and with an extra option "-s 1000".. > > This is all that happens *before* trying to mount from the client. > root@nfsserver:~# systemctl restart nfs-server; exportfs -f; m=$(pgrep rpc.mountd); strace -s 1000 -ttt -T -p $m 2>&1 | tee strace_rpc.mountd2.txt > strace: Process 2517 attached > 1501190436.659800 select(1024, [3 4 5 7 8 9 10 11 12 13 14 15 16 17 18], NULL, NULL, NULL > > Stracing rpc.nfsd: > root@nfsserver:~# exportfs -f; strace -s1000 -ttt -T /usr/sbin/rpc.nfsd 0 2>&1| tee strace_nfsd.txt; /usr/sbin/sm-notify > http://people.cs.uchicago.edu/~kauffman/nfs-kernel-server/take3/strace_rpc.nfsd.txt > > When mounting from the client I get a strace that is 1.3G (it seems '-s' was not respected for some reason): http://people.cs.uchicago.edu/~kauffman/nfs-kernel-server/take3/strace_rpc.mountd.txt > I'm sorry but I cannot get anything useful from these. It isn't crystal clear what was being tested in each case.... Could you try: 1/ build the current 'master' branch from the nfs-utils git tree, with no extra patches. 2/ Install that and perform the same test which produce the 1.3G strace file above. Provide the resulting strace of rpc.mountd. 3/ Tell me what happened when you mounted on the client. Did it work as expected? 4/ Apply the patch I provided on top of the 'master' branch but change hcreate_r(1000, &hdata); to hcreate_r(20000, &hdata); 5/ Install that and perform the same test as in step to, producing a new strace of rpc.mounted. Presumably it will be difference. 6/ Tell me what happened when you mounted on the client. Did it work as expected, or did it fail like I think it did last time. If one worked and the other didn't, I can compare the strace and try to find a difference. If both worked, then we should have useful information about how much speed up the patch provides (if any). If neither worked, then something weird is happening... NeilBrown
Attachment:
signature.asc
Description: PGP signature
