T.R | Title | User | Personal Name | Date | Lines |
---|
1223.3 | SPR? | AIMTEC::WICKS_A | DEC Mail Works for ME sometimes | Wed Aug 12 1992 21:28 | 10 |
| Dan'l
note 3122 in the old conference, 799, 1299 amongst others discuss this
error. I don't recall it still being a known problem post-K603 so I
guess your large chemical company customer must have found a new
instance of it - SPR time
Regards,
Andrew.D.Wicks
|
1223.5 | This is an OLD problem... | AIMTEC::PORTER_T | Terry Porter, ALL-IN-1 Support, Atlanta CSC | Tue Sep 01 1992 20:51 | 41 |
| I have been working on this problem (FILCDERR and NBS files not deleted) for
the last 4 years (both as a memeber of IOSG and a member of the Atlanta CSC).
And now (roll of drums ...) I have a solution!!!!
Unfortunately it is a code level change so don't get too excited.
For some reason (which I am still persuing with VMS engineering) a directory
file can be locked when it should not be. The file access conflict is when
ALL-IN-1 (via RMS) is trying to delete the file from the directory. I suspect
(but have not yet proved) that the cause is related to a combination of factors
- MR creates files in OA$MTI_DATA accross DECnet
- ALL-IN-1 deletes files by opening them with the close on delete flag set
and then closing the file (in this case the open works but the close fails)
- The filenames are such that files are deleted and created at the end of
the directory file.
The workround for ALL-IN-1 is to avoid the problem by re-trying the failed close
(with a limit on the re-trys) with a short delay between re-trys.
I have only been anle to reproduce the problem using DEBUG to get the timing
right (stopping ALL-IN-1 between the open and the close, locking the
directory file, and stepping through the close) so I am not 100% sure that
re-trying will fix the problem on a 'live' system.
I am working closely with a customer on this and he is testing this workround
over the next week or so. If it works I will be sending it back to IOSG for
incorporation into a patch.
When/if this fix gets into a patch is a decision for IOSG, but considering the
low impact of the problem my personal recomendation would be to include the
fix in the V3.0 patch after the one currently being developed (presumably
V3.0-2) and not to patch it for earlier versions. If you have customers who
need an earlier patch than this then now is the time to make a bid to IOSG :-)
I'll add updates when the customer has finished the testing
Hoping this is really the solution ...
Terry
|
1223.6 | Good news, and Bad news | AIMTEC::PORTER_T | Terry Porter, ALL-IN-1 Support, Atlanta CSC | Wed Sep 16 1992 00:10 | 19 |
| The good news is that the fix I put together for the FILCDERR errors works.
The customer (who gets 10 of these a day normally) has been running with the
fix for over a week without a single occurence of the error.
The bad news is that Message Router hits the same problem and generates
%MRROUTER-E-FLOCK errors. The same fix should work here, but it needs to
be in the MR code which I do not have access to.
I have SPRed the whole problem and solution to IOSG with the recomendation
that it be fixed, and that they work with MR engineering (who are in the same
building) to get a similar fix into MR.
I have also QARed the base problem (the fact that the directory files are
locked in the first place) against VMS V5.
Now we wait to see what happens ...
Terry
|
1223.7 | | FORTY2::ASH | Grahame Ash @REO | Tue Sep 22 1992 18:59 | 5 |
| Having been faced with trying to fix this back in pre-Terry days, I'm
delighted that he's got a solution after all of the work involved.
Congratulations, Terry! Now you'll need a new career goal!!
grahame
|