I will be sending a FIX15O1 shipment early next week (I mean the one that
starts tomorrow). It will contain all the updates that were made since 1.5o
was released, most of which are minor changes. The one important change is a
fix to DIST2 to prevent the "VIA check on DISTRIBUTE job" message that some of
you may already have seen. The VIA check of the DIST2 command was originally
copied from that of DIST1 and expanded upon; considering the difference in
algorithms, it is too stringent in the case of DIST2, and could be triggered
in normal conditions where:
1. FILES (not mail) are being distributed to a list.
2. The list is peered.
3. The distribution of the recipients across the various peers is not optimal,
OR it is optimal but some complicated conditions on the network topology
and distribution of backbone servers in that area are met. In practice,
most of the time the first condition is the trigger.
The DIST2 VIA list contains "list entries" (an extension of DIST1), which
prevent a particular list from being included into the recipients list more
than once, and "server entries", which prevent a given DIST2 job from being
processed by more than one server. This latter test was an absolute
requirement of DIST1, because each server re-evaluated the "next hop"
according to its own vision of the network, and is redundant for DIST2. It was
kept for safety reasons, but can be triggered when new recipients cause the
same job to be sent back to a server which already processed it (an admittedly
undesirable situation, but it can happen). The fix I introduced removes the
servers in the path to the new nodes introduced by the expansion of the list
from the copy of the job which is forwarded along that path, without altering
the VIA lists of jobs sent along other paths. For example, if FINHUTC expands
a DIST2 job which came from CEARN and includes recipients for Finland and
Switzerland, CEARN will be removed from the copy of the VIA list going back to
CEARN, but not from the copy going to FINTUVM. In theory I could have
completely removed the test, but I felt safer this way.
By the way, now that the 64k line to Montpellier is working (albeit only in
the CEARN-MOP direction), it no longer takes two weeks for LISTSERV updates to
reach the Outer World. However, console logs from you will take the opposite
route, which is still a 1-week delay for anything > 1000 lines, so be sure to
edit them until they are <= 100 lines before sending them.
Eric
|