Conference lassie::ucx

5516.0. "NFS and RENAME errors" by QUARK::LIONEL (Free advice is worth every cent) Thu May 15 1997 11:53

[I received the following by e-mail from a customer.  Please follow up with
the customer directly - thanks. - Steve]

Hello Steve,

I contacted you before about Fortran and UCX questions, as you maybe rem=
ember.
You were so kind and answered me and it encourages me to send to this le=
tter.

The situation I describe below concerns TCP/IP, I think. So if you feel =
that it
is not your table, pass it to the UCX-expert, please.

I'm using UCX and I mount 2 NFS disks that are on 2 HP-UX systems.
There are several batch jobs that run alomost simultaneously and that ac=
cess
these NFS disks. Initially 3 DCL commands are performed by the jobs:
    1. f$search(<file>)
    2. DIRECTORY <file>
    3. RENAME <file> <new_file>
Sometimes when RENAME is executed the following messages are displayed:
  %RENAME-E-OPENIN, error opening <file> as input
  -RMS-F-REENT, file could not be renamed and recovery failed; file has =
been
lost
  -SYSTEM-W-NOSUCHFILE, no such file
This situation happens when I have 2 jobs that perform the commands
simultaneously". The time difference could be less then 00:00:00.50. One=
 job
wins and performs its commands and the other one produces the messages a=
bove.
This behavior I've tested several times with a help of a simple job:
    $ cnt=3D0
    $ on error then exit
    $ set verify
    $ set prefix "(!%T) "
    $nxt_rename:
    $ cnt1=3Dcnt+1
    $ directory/date 'p1':tmpfile.'cnt'
    $ rename/log 'p1':tmpfile.'cnt' .'cnt1'
    $ cnt=3Dcnt1
    $ if cnt .le. 99 then goto nxt_rename
    $ exit
The job is submitted twice with the same /AFTER value. P1's value is a l=
ogical
name that points to each HP-UX system. tmpfile is an empty UNIX file, cr=
eated
(touch) before jobs start.

In one batch log you find:
------------------------------------------------------------------------=
----
(08:35:01.23) $ directory/date nfsdisk1:tmpfile.0

Directory DNFS1:[000000]

TMPFILE.0;1          14-MAY-1997 08:28:16.00

Total of 1 file.
(08:35:01.31) $ rename/log nfsdisk1:tmpfile.0 .1
%RENAME-E-OPENIN, error opening DNFS1:[000000]TMPFILE.0;1 as input
-RMS-F-REENT, file could not be renamed and recovery failed; file has be=
en lost
-SYSTEM-W-NOSUCHFILE, no such file
------------------------------------------------------------------------=
---
and in the second log file:
------------------------------------------------------------------------=
---
(08:35:01.22) $ directory/date nfsdisk2:tmpfile.0

Directory DNFS2:[000000]

TMPFILE.0;1          14-MAY-1997 08:33:06.00

Total of 1 file.
(08:35:01.29) $ rename/log nfsdisk2:tmpfile.0 .1
%RENAME-I-RENAMED, DNFS2:[000000]TMPFILE.0;1 renamed to
DNFS2:[000000]TMPFILE.1;1
------------------------------------------------------------------------=
-----

The question is if there is any explanation for this error and an obviou=
s
conflict
during the NFS access. Since the NFS are placed on different UNIX-comput=
ers the
reason for the error should be looked for on VMS-site, shouldn't it.
Is it something in UCX configuration that causes the error or is it a bu=
g.
I could synchronize my jobs to avoid the conflict but I don't thing it i=
s the
way. I can't synchronize all NFS accesses!

After each failed RENAME I find a file on UNIX-site. The file has name l=
ike
.$NFS$9bbaed. It contains this lost file and it is hidden file (note a d=
ot).
Is there any NFS way to access the hidden files? What mechanism creates =
the
files?

I hope you or some of your colleagues can give answers on my questions.

Best Regards
Teofil Smolowicz
E-mail: [email protected]

Not a solution, but an occurence of the same. My customer uses the VMS NFS
client in combination a NT server running NFS-Maestro (whatever this may be)
software.

Renaming a file (always) results in:

$ RENAME x.x y.y
%RENAME-E-OPENIN, error opening DNFS1:[000000]X.X;1 as input
-RMS-F-REENT, file could not be renamed and recovery failed; file has been lost
-SYSTEM-W-NOSUCHFILE, no such file
$

The result is:

$ DIR

Directory DNFS1:[000000]

.$$NFS$$9B4B83;1

He also has a Digital UNIX client for this server and renaming file from
there works fine.

I have two questions:

- I guess this is some timing problem, would this be a client or server
  problem?
- How do I troubleshoot this, a nfs/debug=180 command only works on systems
  running the nfs server process?

Thanks in advance,

Oswald

Ok, I made a trace with tcpiptrace. Does anybody have some tips on how to
interpret this, tcpiptrace unfortunally has no knowledge of nfs....

For the interested... Trace is at utrtsc::netw_user:[knoppers]nfs.trace

Regards,

Oswald

>Ok, I made a trace with tcpiptrace. Does anybody have some tips on how to
>interpret this, tcpiptrace unfortunally has no knowledge of nfs....
    
    No tips.  Just get out your RFC 1094 and RFC 1057 and start slogging. 
    It gets faster with practice.  8-{
    
    But, cumpulsive personality that I am, I have had a look at your trace. 
    It's a server problem.  
    
    For some background on why the UCX client does what it does, have a
    look at #4985.* in this conference.
    
    seq # 9, 10
    client does readdir; response shows directory containing only a.a
    
    some lookups and getattrs
    
    seq # 18, 19
    client does lookup on .$NFS$9b4043 to make sure it does not already
    exist; response confirms it does not
    
    seq # 20, 21
    client renames a.a to .$NFS$9b4043; gets successful status returned
    
    seq # 22, 23
    client does lookup on .$NFS$9b4043; response claims there is no such
    file!
    
    seq # 24, 25
    client checks directory attributes; modification time has changed since
    last attribute check before the rename
    
    seq # 26, 27
    client does readdir; response shows directory now contains only
    .$nfs$9b4043  Note wrong case!
    
    client does more housekeeping, including caching attributes on this
    .$nfs$9b4043 file
    
    seq # 37, 38
    client tries renaming .$NFS$9b4043 to b.b; server is still sure there
    is no such file
    
    seq # 39, 40
    client tries to back out the first rename by renaming .$NFS$9b4043 back
    to a.a; server is still sure there is no such file
    
    At this point the client gives up trying to go forward or backward on
    the rename from a.a to b.b.
    
    Julie

>    But, cumpulsive personality that I am, I have had a look at your trace. 
>    It's a server problem.  

Thanks for the analysis, I've informed the customer.

Regards,

Oswald