WPWG-20100525-transcript

From GO Wiki
Jump to: navigation, search
- Tuesday May 25 2010, 08:31 -
sjcarbon: morning'
ben: howdy
j-lo: hiya
sjcarbon: hey!

- 08:37 -
ben: seth, probably you and I should chat for a few minutes after
ben: about systems issues
sjcarbon: okay--no problem
 - gwg has joined the room
sjcarbon: Well, I wanted to talk about the three graphs in the demo page, and maybe some problems with doing in through QuickGO.
sjcarbon: but we don't really have the audience that we had last week...
sjcarbon: Too bad I guess.
 - gwg has disconnected (Quit: Leaving.)
sjcarbon: For the people who really liked the QG graph (any here?), could I ask you a question about it?

- 08:42 -
sjcarbon: Or for the smaller audience: do either of you prefer the QG graph?
 - gwg has joined the room
ben: 3 looks cleaner than 2
ben: but of course, less informative
j-lo: sorry - yes I liked the QG graph
ben: the reality is that this is too small an example anyway
sjcarbon: ben: i discovered that having labels in graphviz really messes the layout algorithm
ben: you need a big ol nasty graph
sjcarbon: yeah, I was hoping people could bring their favorites for show and tell.
sjcarbon: here...I have one on the board...lemme see...
j-lo: are there links we should be looking at seth?

- 08:47 -
sjcarbon: on the wiki, but if you have a convoluted area of the graph that you like, we can use that as a discussion  point. 
sjcarbon: how about:
sjcarbon: http://www.ebi.ac.uk/QuickGO/GTerm?id=GO:0034223
sjcarbon: http://amigo.berkeleybop.org/cgi-bin/amigo/amigo?mode=term_details4&term=GO:0034223
 - pascale has joined the room
j-lo: arggh - my client turning the :0 into smily faces so links don't work!
j-lo: gottit now

- 08:52 -
sjcarbon: 
sjcarbon: unfortunately, my cache of the quickgo isn't working there (will debug later), so just follow the quickgo link for the comparison.
j-lo: see, these are the cases where you would want the parent and child view only
j-lo: all graphs are useless
sjcarbon: personally, i'd say that the graphviz (non-quickgo one) is less useless here. what are the cases where the graph is complicated, but not useless?
ben: lets just say they have equal value
sjcarbon: besides some other reasons (like we'd have different graph formats in amigo because quickgo's is not general purpose), it seems to be better at showing neighborhood/rank, rather than structure.
sjcarbon: i'm trying to get a feel for where the border sits.
sjcarbon: and whether or not the main thing that people like about the quickgo one is the rank feel of it.

- 08:57 -
j-lo: I think it's quite compact
j-lo: and it has a very visible key
j-lo: and the boxes help with interpretation somehow
ben: i agree that it's compact, that's a plus
sjcarbon: i also find it easier to read,
sjcarbon: but only when comparing ranks.
ben: I find the "key" to be almost useless
gwg: is it right that when I click on the QuickGO image link, I get a message saying "http://amigo.berkeleybop.org/cgi-bin/amigo/visualize?inline=false&mode=quickgo&beta=0&term=GO:0034223"
ben: he just said that's a bug, you have to go directly to quickGO for now
sjcarbon: yes--bug from earlier, use the quickgo link.
gwg: sorry, that must have been when I dropped out of the chat
sjcarbon: when the ontology jumps multiple "levels", it gets very confusing.
sjcarbon: (to me.)
sjcarbon: for example:

- 09:03 -
sjcarbon: http://amigo.berkeleybop.org/cgi-bin/amigo/amigo?mode=term_details4&term=GO:0021787
sjcarbon: (all tabs work in this one)
ben: quick go is much better for this one, simply because it is more compact
sjcarbon: i sense a compactness fixation. i guess that is the information i wanted.
sjcarbon: i find the squiggly lines of quickgo maddening though.
sjcarbon: i'll see if i can tease a more compact graphviz graph out, if not, i'll go with the quickgo.
j-lo: I think the OE graph view is pretty nice too
j-lo: v similar to QGO
sjcarbon: that means though, that amigo will present different graphs in different places--the quickgo system isn't something we can use in the general case.
sjcarbon: (it actually is quickgo.)
sjcarbon: (we lifted the code a while back)
sjcarbon: do y'all think people can deal with that?
j-lo: yep

- 09:08 -
j-lo: (btw - it's the OE Graph Viewer)
sjcarbon: for fun, on the wiki page there is also also a fully tabbed layout for term details.
sjcarbon: also, there is a link to Sven's work--it is now folded in to experimental and should be getting regular updates.
sjcarbon: I'm not sure how much outside comment hes taking yet though.
sjcarbon: Pascale might know?
gwg: the fully tabbed version looks good
ben: brb
sjcarbon: I found it a bit maddening, but that might just be me this week.
gwg: it looks maddening with js turned off 

- 09:13 -
sjcarbon: I'll keep that but of code dusted off so we can visit it all more seriously when we have a chance.
gwg: I think all those graph views are enough to send anyone mad tho...
sjcarbon: Gave me a headache...
gwg: thank you for doing all the experimenting!
sjcarbon: I see lines and boxes when I close my eyes...
sjcarbon: No problem. This should be wrapped up this week, then onwards.
sjcarbon: Anybody have anything else?
ben: back
sjcarbon: I think we're done then. I'll send out links to a "final" layout after I track down the caching bug and check for graphviz "compactness".
 - pascale has disconnected (Ping timeout)
 - pascale has joined the room
 - pascale has left the room

- 09:20 -
ben: seth - the tricks we tried with the load balancer didn't work last night
sjcarbon: argh.
ben: goweb2 still exploaded
sjcarbon: should we disable that search?
sjcarbon: or will the kill scripts be effective?
ben: that might be best - or at least cripple it
gwg: perhaps there should be some documentation about what dbxrefs exist in the GO DB
ben: well, we tweaked the kill script, but not sure it's working still
ben: gwg: no, the software should gracefully handle any input
ben: not spin mysql for 2 minutes if someone uses a perfectly valid id
ben: that we don't happen to cross index
gwg: what do you suggest? gracefully killing their query?
ben: the kill script was designed for REALLY long running queries - like 30 minutes.
ben: like not running queries that take 2 minutes
ben: maybe the problem is the %blah%
ben: wild cards?
sjcarbon: i'll see if i can get something temporary in to just get those, and die with a message to the lucene on experimental for now.
ben: then at least we could use the indexes
sjcarbon: i'll check both.
ben: yeah, any bogo patch will work for now.  I currently think there is something wrong at the hardware level as well
ben: because it's always the SAME machine that blows up.  goweb2 never goweb2
ben: sry never goweb1
gwg: can you try sending all those queries to goweb1 in that case?

- 09:25 -
ben: we are still messing with it here... but Stuart is really backlogged because he had pnumonia
sjcarbon: okay. i'll work on that until around lunch, and then let you know what i've come up with.
ben: gwg: well the stupid LOAD BALANCER is supposed to send queries to less loaded machine!
gwg: but if one of the machines seems to handle them better...
gwg: anyway, no point in arguing about it
ben: the machines are supposed to be identical...
ben: another issue is the sessions.
ben: this forces us to make the IP address "sticky" so that a user always gets the same machine (which has it's session stored on it)
ben: we suspect this stickyness might be part of the trouble, but I don't think amigo will even work without session ids currently
sjcarbon: it should work...but it will be doing more work (no cached data).
ben: what does it store?  The filters, for sure.  what else

- 09:31 -
sjcarbon: i think there's chewed search results in there as well sometime. as i recall, if it can't find the cache, it just soldiers on.
sjcarbon: i don't thnk any of the production new stuff uses session/cache in any non-agnostic way.
ben: hmmm... I recall that it redirects to the home page if you give it a bad session id... but that expereince from a few years ago
gwg: it used to, but I don't know if it still does
sjcarbon: lemme see...
ben: I just tested by backspacing a few chars out of the session and it seems to work fine
sjcarbon: that should be the behavior.
ben: ok, I will suggest that we remove the stickyness and see how it goes
sjcarbon: okay, i'll look at the queries, and see how easy catching them are. i'll contact you again around lunch.
ben: kk
ben: later