Working Group Minutes/EWG 2013-01-14


IRC nick Real name
apmon_ Kai Krueger
pnorman Paul Norman
RichardF Richard Fairhurst
TomH Tom Hughes
zere Matt Amos


  • Review of the year:
    • Why was EWG not as effective as we'd hoped at the beginning of 2012?
      • Perhaps its goals are not concrete enough?
      • Examples of where EWG was effective?
  • Usage of development server (errol):
    • For a while, errol was being monopolised by OWL processing, making it difficult to run anything else. This is over and OWL has been moved to a dedicated machine, so it should be free now to do other processing.
    • pnorman wants to look into using it for some imagery proxying and a snapshot-server instance.
    • apmon_ to look into importing and keeping up-to-date an osm2pgsql database for read-only use and analysis by other dev users.


18:02:23 <zere> minutes of the previous meeting: please let me know of any inaccuracies or problems
18:03:00 <zere> following on from last week, i thought we might try and gather some "goals" for the year
18:03:55 <zere> specifically, the type of "direction-setting" goals. perhaps not specific projects, but areas of improvment to make in the code / new code.
18:04:52 <zere> for example, i still think it's important to have the bugs/notes branch as a goal
18:06:59 <zere> #topic Gathering goals for 2013
18:07:15 <zere> and we can use the #idea meetbot command, like so:
18:08:02 <zere> #idea system for adding bugs or notes via the web interface (bugs/notes branch)
18:14:41 <zere> TomH, apmon_: log up to this point
18:15:16 <apmon_> zere: thanks
18:16:51 <apmon_> perhaps at some point we should also have a meta discussion
18:17:27 <apmon_> i.e. what actually worked in ewg and how to apply that to other parts that didn't achieve their objectives
18:17:50 <zere> i tried to have that last meeting, but it didn't really seem like anyone was interested
18:18:10 <zere> maybe we need to have a meta-meta-discussion - is EWG worth continuing with?
18:19:10 <apmon_> Yes, that is the extreme version of that discussion
18:20:02 * pnorman checks in
18:21:03 <zere> tmcw: log so far
18:21:38 <apmon_> It feels like the efforts to recruit new mappers have not really worked out so far
18:21:50 <zere> s/mappers/developers/?
18:22:22 <apmon_> sorry yes
18:22:23 <zere> recruiting new mappers is a whole other kettle of fish
18:22:38 <apmon_> So it kind of leaves a communication platform of a group of devs that hang out on irc anyway.
18:22:44 <apmon_> Which might never the less be worth it.
18:24:54 <zere> there were several reasons for EWG originally: 1) to promote and help new development, 2) to provide a discussion ground for building consensus on divisive topics (that would otherwise be flame-wars on the lists), 3) to provide coordination with the rest of OSMF and access OSMF resources where needed.
18:25:42 <pnorman> hmm, reminds me I had some questions on 3) if we have the time later
18:26:13 <zere> depending on how you read the TTTs, we've done pretty poorly at (1). attendance is low, so it seems (2) isn't being met either. and where we're not meeting (1) and (2), (3) seems utterly pointless
18:27:10 <zere> i'd love to say that we're all fine because everyone is JFDIing and such, but that's not true either.
18:28:58 <zere> is it a fault with the meeting format? - should we have meetings less often / longer or shorter meetings?
18:29:24 <zere> is it just that no-one sees the point in participating in EWG?
18:34:21 <apmon_> Do we have some good examples of where EWG worked?
18:34:43 <RichardF> I suspect (as is often the case with OSMF WGs) that EWG doesn't really have a concrete goal, and it's hard to get volunteers motivated by abstracts.
18:34:52 <apmon_> that might help analyze what people hope to get from EWG
18:38:51 <zere> RichardF: how do you think the goals could be made more concrete?
18:40:18 <zere> apmon_: sure... i had hoped to get more developers interested and developing on OSM-related software, and providing a platform for them to collaborate constructively. how about you?
18:40:22 <RichardF> zere: I wish I knew. ultimately OSM developers are hackers at heart, and hackers like hacking. most of us aren't organisational consultants who like organisational consulting
18:41:38 <zere> on the basis of that, we'd seem to be better off just leaving it. when the organisational consultants come, they can tell us where we went wrong ;-)
18:48:02 <zere> ok, moving on. pnorman: what were your questions?
18:48:30 <pnorman> I've got a few projects going on that when I started OWL was running on the dev server, making it unusable for pretty much anything
18:49:43 <pnorman> Now that its not, someone at mappy hour suggested moving some of them there, but I'm not really sure whats suitable for running on it
18:51:26 <zere> i'm not the best person to answer that - ppawel would be better, but he's not here.
18:51:40 <zere> oh, wait
18:51:45 <apmon_> That actually might in general be a good topic for EWG in general. How well does the dev server setup work? And would some additional hardware help in that respect to support osm hackers better?
18:52:13 <apmon_> As that is something where OSMF (with its resources) might actually be needed
18:52:18 <zere> completely the wrong question! i thought you were asking about whether OWL was suitable for building on top of... but you were asking about the dev server...
18:52:26 <zere> pnorman: sure. what do you need?
18:52:50 <pnorman> zere: well, the two main ones are some imagery proxying and snapshot-server testing
18:53:14 <zere> imagery proxying?
18:53:26 <pnorman> wms -> tms with caching
18:54:19 <zere> according to munin, errol is mostly idle:
18:54:33 <zere> i'd say try it and see :-)
18:55:58 <zere> the policy has always been that people can run things there (within reason), and if they prove popular then will be considered for their own server.
18:56:30 <pnorman> I know i've always put off consideration of errol in the past because of OWL
18:56:59 <zere> yeah, sorry about that - it really mashed the machine for a while...
18:57:23 <zere> but its off now, so errol should be back to normal again
18:57:41 <apmon_> imho, the fact that there are no pre-existing databases (and it doesn't have the disk capacity to run a full scale db) has been somewhat limiting.
18:58:11 <pnorman> Doesn't it have a massive RAID5 array?
18:58:24 <zere> sadly not
18:58:47 <pnorman> needs updating then?
18:59:36 <zere> oh, sorry. i thought you meant massive in terms of number of disks. yes, it has a slow 2TB*5 raid5 array
19:01:58 <pnorman> Ya, I'll stick my canvec/NHD snapshot-server tests there. if it takes off then I'd need to talk to someone about resources but that's a bit off
19:02:11 <zere> pnorman: cool :-)
19:02:46 <zere> apmon_: hmmm... did we not think the worldwide osm2pgsql database would work on slow disks?
19:03:10 <pnorman> Hourly updates maybe?
19:03:33 <apmon_> It might be interesting to see if the flat-nodes stuff might have improved the situation with respect to I/O requirements for updates
19:04:03 <zere> apmon_: if i create you a tablespace somewhere on that large disk, do you want to have a go?
19:04:04 <apmon_> but would e.g. $200 for an SSD for erol be out of the question?
19:04:24 <apmon_> zere: Yes, I can give it a shot to see if it kills the server
19:05:12 <zere> it would appear to have 3 bays free. and i can always stick in my 250GB SSD to see whether that works...
19:05:41 <zere> does anyone have anything else to discuss?
19:06:10 <TomH> errol is supposed to be a general purpose dev server - we're not really in the business of filling it with special hardware for people
19:06:32 <zere> sure, but SSDs are hardly special purpose...
19:06:43 <apmon_> Is having an OSM database on an OSM devserver not pretty general?
19:07:08 <zere> it might enable more general purpose dev to have an updated, read-only rendering db on there
19:07:17 <apmon_> Does it make sense to try and prepare for the hackweekend in some way? To make the most efficient use of it?
19:07:23 <zere> or apidb database - either or both ;-)
19:07:40 <pnorman> zere: apidb would require much more hardware :)
19:08:07 <pnorman> an osm2pgsql db with hstore would allow stuff like to be on errol
19:08:22 <zere> to run quickly, sure. but enough to noodle around with maybe.
19:08:45 <pnorman> do you mean with the planet loaded, or just a general apidb?
19:09:05 <zere> yes - i think the idea is to have the whole planet on there
19:09:33 <zere> for general apidbs, there are already a bunch of dev rails envs, which i guess would serve the other purpose
19:09:49 <pnorman> it took my server a month to import the planet half a year ago and that was on 4x 7200 RPM in RAID0, and it took about 1TB
19:10:16 <zere> was it able to keep up to date with minutely diffs?
19:11:14 <pnorman> easily, I only had to stop it because a disk died. I just question if there's many people who would find a use for a planet-sized apidb
19:12:14 <zere> i don't know - i've always loaded extracts locally...
19:12:28 <zere> apmon_: you were talking about an osm2pgsql database, or an apidb database?
19:13:17 <apmon_> my guess would be that an osm2pgsql db would find more users as it has postgis geometries
19:13:28 <pnorman> There are lots of uses for an osm2pgsql db, a reasonable number for pgsnapshot, but apidb is very specialized
19:13:35 <apmon_> and with the slim tables and hstore, it pretty much has all information in it too
19:14:32 <zere> fair enough - i think it's worth a try. we can always look at it in a few months time and see if it's monopolising errol and whether it's worth it...
19:14:52 <zere> cheers, guys - have a good week & see you next week.