Eric Thomas <ERIC@FRECP11>
Thu, 14 Jan 88 23:24:00 SET
|
The cause of the 'loop' has been eventually found, and fixes are on their way
to the LSTSRV-L sites (most of them are already loaded and online). I'm glad
to have that sorted out before having to go home... It will make sleep much
easier... :-) /Eric
PS: Note that the FIX15M3A will be sent ONLY to the relevant servers. It's not
an urgent fix since the (DIST1) loop can normally not be triggered without
using DIST2.
/../
(irrelevant lines deleted)
But finally... I think you gave me the key to the DIST2 'loop' problem. Ross
(resp Richard) sent me a console log (resp a DIST job) showing that
LISTSERV@PSUVM and LISTSERV@WVNVM kept distributing files to each other with
the origin being set to [log in to unmask] What is funny is that there was *NO*
'From LISTSERV@stuff: DIST2 blah' in the console log. Now if you go through
LSVCMD, you can't escape this message. Now I realize that they must have gone
through LSVFSEND, ie LISTSERV sent a file to the list userid via direct NJE
send. In this case you don't have a 'From' message since the distribution is
implicit, and you don't go via DIST2 but via regular DISTRIBUTE, and the 'via'
tag is reset since it's a new file, and the DIST 'FROM' origin is the NJE
origin.
All of this is due to some piece of old code in LSVFSEND that is commented as
'supporting the old protocol'. This refers to the pre-DISTRIBUTE way of
sending files to a list. For obvious reasons I had to support this 'old
protocol' for a few months while people installed the new version (which uses
DISTRIBUTE). I intended to remove this code later and to trap all files sent
by a LISTSERV machine to a list, since this normally should never happen. But
as you can see, I forgot to do it...
Now why was a file sent from the LISTSERV userid to a list in the first place?
I'm not sure about what I'm saying, especially as a number of links are down
and I can't check what I'd like to. But UIUCVMD is not on the DIST2 backbone,
and there is a LSTSRV-L peer there. When LISTSERV@POLYGRAF processed this
recipient, it had to use direct NJE send. The file landed on LSTSRV-L@UIUCVMD
with an NJE origin of LISTSERV@POLYGRAF, and this 'old protocol' was triggered
(a new distribution was started with a new VIA tag, etc).
So now how do we stand, given that proper explosion (without loss of DIST<2>
control info) cannot be achieved without sending a job to the LISTSERV
address?
1) Non-backbone sites should not be allowed to house peer lists. This
situation cannot happen if all peer sites are on the backbone. There is no
problem about non-peer lists.
2) The 'old protocol' code will be removed. However, we should keep in mind
that it should be possible for a LISTSERV userid to submit a file to a
NON-PEERED list via direct NJE send. It should NOT be possible to submit
the file in that fashion for a peer list.
I will be sending a FIX15M3A update with just LSVFSEND EXEC in it to your
servers. This should stop the loop (assuming it is still running). I will then
resend a copy of the note to LSTSRV-L when everything is solved.
Eric
|
|
|