Attending: RobC, Ste, Luke, Patrick, Jens (chair+mins), PeteC, Duncan, Teng, Wenlong, Raja, Winnie, Sam, Matt

1. Update on Bristol

DPM's GridFTP gets stuck for unknown reasons.  Rob suggests ECDF had a similar problem ~2yrs ago, so might be
due to an old DPM?  The Bristol version is "fairly recent" with the constraint that they cannot use 1.9 or later.
At the most busy time, it survives about 15-20 mins, then gets stuck.  OTOH, there have not been any problems
in the past ~1.5 yrs.

Sam did have a look at it as well.  OTOH, maybe it's less important to understand the problem if the medium
to long term solution is to move away from DPM anyway, since Bristol cannot migrate to DOME DPM. 

Another example is where the deletions through GridFTP caused heavy load on the system; when deletions were
switched to HTTP, it worked fine.

Load seems to be lower now, and can switch kill script (cron) back on.  Of course it could kill a legitimate
transfer thus making it fail and having to retry.  No known way currently to distinguish failed (stuch) transfer
from busy one.

The plan is to migrate to a HTTP and/or xroot solution not using DPM, so need to check HDFS plugins for HTTP(?)
and xroot (the xroot server supports HTTP).  This configuration is used by OSG, but comes in a magic potion
brewed in an OSG cauldron, so may not directly apply to a non-OSG site.  Sam is curious about the setup as well,
having not configured HDFS before.

In the short term, can switch off GridFTP transfers and use DPM with HTTP.  Then, until the end of the (calendar)
year, Bristol will run xroot in test mode, e.g. with PhEDEx-Dev transfers, aiming to migrate early 2020 if
successful.


DOME: Only Glasgow and Durham thought to be missing (we haven't got the monitoring any more; it disappeared with
the BDII).  Durham expected to switch before the end of the year; Glasgow will also switch if possible.

Thus, all going well, remaining non-DOME sites should have migrated to DOME or away from DPM by early next year.


Documentation.  Makes no sense to waste time updating non-useful docs.  However, the vast majority of keydocs are
currently very red (not just read).

Worst case scenario: someone outside of GridPP's (or IRIS') immediate support structure comes to the web site
and finds misleading information.

Best case scenario: every piece of doc is reviewed by someone else, who then provides feedback or (even better) fixes.

Ste makes the point that most docs that are out of date, but it's sometimes better than none at all.  Even his recent
page on HTCondor (7 Nov) now has 3,000+ views, so they are being picked up (although this could be by search engines)


From Winnie Lacesso to Everyone:  10:12 AM
Except ganglia - we do have ganglia monitoring
From Matt Doidge to Everyone:  10:15 AM
sorry, need to step out of the meeting
From Nandakumar, Raja (STFC,RAL,PPD) to Everyone:  10:33 AM
Apologies - got to go now
From Matt Doidge to Everyone:  10:37 AM
I think making kids read grid documentation might be some kind of abuse
From Ste to Everyone:  10:38 AM
I pay them! I have to.