Date: Fri, 29 Mar 2024 10:25:02 +0000 (UTC)
Message-ID: <1195704032.21877.1711707907169@confluence>
Subject: Exported From Confluence
MIME-Version: 1.0
Content-Type: multipart/related;
boundary="----=_Part_21876_1213658733.1711707902158"
------=_Part_21876_1213658733.1711707902158
Content-Type: text/html; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
Content-Location: file:///C:/exported.html
LSST Database Hangout 2014-07-16
LSST Database Hangout 2014-07-16
Date
Jul 16, 2014
Attendees
Goals
- Get together, resolve all bottlenecks, make progress faster, exchange i=
deas
Discussion Items
master inadvertently broken (merge of DM-622)
- probably incorrect merge
- fixed in u/fjammes/dm-967 during the meeting, reviewed and merged<=
/li>
How to install system dependencies?
- Can't do through eups
- Maybe special package that checks and reports missing dependencies?
- Check with Kian-Tat Lim?
- (KTL) We don't have a solution for this in the stack yet. My pref=
erence is currently for a "stub" package that checks for system dependencie=
s. I don't want to have system dependency installation in
newin=
stall.sh
, which should only be for installing eups
.
Running on multi-node
- runServer, people not happy at large scale
- pmg used by LHC/Atlas
- ideally, just be flexible and use whatever is recommended for a given c=
luster
contrib/eupspkg/* packages that Fabric=
e Jammes created
- Should there be a better place for these things?
- Ask Kian-Tat Lim?
- (KTL)
LSST/external/*
once they have been reviewed.
xrootd integration tests
- Get the latest custom xrootd package, see DM-944
- Get qserv from u/daniel/kirkwood
- Jacek will set up test environment
- identified first issue: context not passed during the meeting
25 node cluster for Qserv test
- Likely available in Sept, maybe even in Aug, Fabrice will double check<=
/li>
Refmatch
- About week away from testing
- Includes name resolution
- Can't support outer joins that are spatial
- So queries like "find objects without matches" would not work
- Chat with Mario Juric about it,=
maybe in Phoenix?
- That is because now we concatenate results from two joins: one against =
subchunk data, one from overlap
- Possible fix: maintain in memory subchunk, and subchunk+overlap (now we=
have subchunk and overlap). But This doubles memory requirement. Need to r=
un through spreadsheet and check cost increase
- Or... load to memory subchunk+overlap, filter out overlap through =
a non-materialized view
boost dependency checker script (by And=
y Salnikov)
- code non-trivial, naming conventions in boost crazy, likely to change&n=
bsp;in the future, code needed in more than one place
- needed for partitioner (see DM-666)
- want to put the boost dependency checker into sconsUtil
- but sconsUtil depend on eups
- partitioning package is simple, does not want all these dependencies
- sentiment: maybe better to duplicate code for checking boost versions&n=
bsp;than use sconsUtil?
- get advice from Kian-Tat Lim
- (KTL) I suggested to Daniel that maybe we should punt on all this and j=
ust require the user to specify where Boost is (equivalent of "
--with=
-boost=3D
"). We can of course look in BOOST_DIR
i=
f eups
has set that.
------=_Part_21876_1213658733.1711707902158--