#concourse

/

      • alexsuraci_
        xoebus: so it looks like being in a different user namespace prohibits you from mounting cgroups. pretty odd
      • even though the process is running as true uid 0 (maps to uid 0 outside)
      • xoebus
        new ui is awesome
      • would privileged help?
      • alexsuraci_
        no
      • it's already running with all capabilities, as root
      • (and thanks)
      • for now i think i'll lock the version of garden-linux down to the previous one
      • by having testflight say passed: [shipit] for its input :P
      • ...and making it so the new ui can render cycles
      • xoebus
        did you push?
      • time to kick off all the builds
      • alexsuraci_
        nope
      • xoebus
        k
      • have you used the go tls support?
      • (for http)
      • alexsuraci_
        for what now?
      • xoebus
        https
      • or are we saying that people should terminate at the elb?
      • alexsuraci_
        for now, yea. or some other service (cloudflare)
      • xoebus
        mhm
      • alexsuraci_
        (also don't try to view the pipeline atm)
      • unless you like stack overflows
      • fix is working its way through
      • xoebus
        k
      • alexsuraci_
        xoebus: fix is in, and back on garden 0.55, gonna try triggering cf-resource
      • worked
      • bump errored because of what looks like docker registry bs
      • couldn't find image immediately after creating it. works now
      • cppforlife
        alexsuraci_: de-dup and sort alpha resources
      • alexsuraci_
        ?
      • also sorting is hard, that's why it's so slow
      • it gives up around ~10 iterations atm
      • need to optimize
      • what do you mean by alpha resources
      • sort alphabetically?
      • cppforlife
        yea
      • sorting nodes should be fast n easy
      • alexsuraci_
        eh maybe as a lower-order influence
      • not if you want it to look good
      • i.e. not crossing lines too much and not being too far from your source/sink
      • cppforlife
        actually i take it back
      • place them at the same level
      • that's hard and slow
      • but will be so much better
      • alexsuraci_
        place what at the same level?
      • cppforlife
        concourse resource for each job should be at the same level
      • i.e. straight line
      • through the jobs
      • alexsuraci_
        oh, yea that's another thing i want to do
      • that's hard because the order of the in/out line is the same on both sides
      • for nwo it only sorts the nodes, i want to extend it to finding an optimal position of those too
      • and then it'll be straight
      • cppforlife
        sorting alpha will help btw
      • alexsuraci_
        here's the current brute-force "algorithm": https://github.com/concourse/atc/blob/master/we...
      • yea sorting them alphabetically might work pretty well come to think of it
      • at least for columns containing only resources
      • because they'd be consistent with each other
      • but if i apply it to both jobs and resources it'd probably look like a twist-tie
      • cppforlife
        looking at ci.concourse.ci sorting alpha fixes lots of crossing
      • alexsuraci_
        so one cheaper algorithm could be: sort resource columns alphabetically, and then sort jobs with brute-force to fix any remaining crossing lines/extra distance
      • cppforlife
        just to clarify are you talking about crossing resource->job connections?
      • alexsuraci_
        yea
      • sorting jobs alphabetically won't help since they're uniquely named
      • cppforlife
        just sort resources and connect them in that order
      • to the job
      • alexsuraci_
        question is more of if there are two jobs in the same column that require different resources
      • should do a best effort to align them
      • could end up criss-crossing
      • end result might not justify the cost though
      • cppforlife
        let them cross
      • make finding resources real quick
      • alexsuraci_
        sure, just saying that's not the silver bullet
      • could end up with the resources being neatly sorted but still lines crossing everywhere
      • from resources <-> jobs
      • sorting them alphabetically solves some of it
      • cppforlife
        i think tweaking width of connection might ease up crossing
      • thickness*
      • i like this though; def an improvement
      • alexsuraci_
        thx
      • yea i'm glad to kill hidden: true
      • cppforlife
        de-dup btw should clean up some of the confusing things too
      • alexsuraci_
        don't think i can really dedup anything
      • in this case i can because those 4 concourses go to the same place, but that's not what the semitransparent boxes are supposed to represent
      • i think deduping is what i had before but it implied the wrong thing in some other pipeline i tried
      • cppforlife
        not sure what you mean
      • alexsuraci_
        those 4 concourse boxes just represent the outputs of 4 different jobs, it's not meant to be a fan-in target
      • if some other job comes in and only requires the output of 'concourse passed: [blackbox]' for example it'd only come from one of those boxes
      • just happens that they all go to one job that needs all 4
      • it's not clear that it's an 'and' though
      • cppforlife
        you have a fan-in example somwhere?
      • alexsuraci_
      • 4x concourse -> rc
      • cppforlife
        oh so that one is fan-in
      • all concourses need to be same ver right?
      • alexsuraci_
        yea
      • only indication of that is that it's all going to one input, but that could also mean 'when any of those passes'
      • so it's not clear that it's an 'and'
      • same goes for if there were 1 concourse and 4 lines going into it
      • man dropbox's ui is obnoxious
      • k here's with alphabetical sorting on everything: https://www.dropbox.com/s/vh9ra7cg08ijbue/Scree...
      • and here it is with brute-force sorting for jobs: https://dl-web.dropbox.com/get/concourse-progre...
      • cppforlife
        2nd is forbidden
      • alexsuraci_
        woops
      • cppforlife
        btw you gotta change that black color for jobs -- so confusing compared to resources
      • alexsuraci_
      • main win for the latter is having the 4 fan-ins bunched together
      • cppforlife
        would be nice to be able to change connection angles such that connection made from a job->resource do not overrun each other (e.g. testflight -> garden-r, version)
      • alexsuraci_
        yea
      • could have it sort by alphabetical and name length :P
      • push longer ones further down
      • cppforlife
        have you tried to make jobs as long as all resources coming into them
      • e.g. connections would end up to be straight in most cases, unless multiple jobs involed
      • i dont think sorting jobs is helpful since jobs are not threaded through
      • alexsuraci_
        increase the spacing between the connection points you mean?
      • so it matches the resource node spacing?
      • cppforlife
        right
      • no need to make jobs tiny in height
      • alexsuraci_
        sorting jobs only helps to detangle that single column
      • it's not really 'sorting'
      • just finding the optimal positioning
      • so reduce visual noise
      • cppforlife
        may be even flip text on a job 90*
      • alexsuraci_
        btw i switched from wiggling to pulsing
      • before i did though they were wiggling funnily
      • looked like they were tapping their feet
      • also experimented with making the nodes all the same width. it looks pretty nice but makes it harder to pick out a job/resource visually amongst the rest
      • https://www.dropbox.com/s/bz3vyr1m99botz4/Scree... (picked an arbitrary number, so some overflow)
      • actually guess it's not so hard; the text is easy enough to pick out
      • cppforlife
        i meant the height of the job
      • making width the same is a bit visually dull
      • e.g. testflight job will span concourse to lite-stemcell resource in height
      • alexsuraci_
        yea, was just doing something different
      • with job heights matching total height of their resource nodes
      • cppforlife
        gotta bigger res?
      • alexsuraci_
        i can zoom in on a bit but otherwise no
      • can also just dump the js somewhere if you want to set a breakpoint and swap it out
      • cppforlife
        nm
      • ill just go ahead and kill my eyesight
      • alexsuraci_
        move closer to the screen
      • it's a vector, it scales
      • cppforlife
        vector my ass
      • alexsuraci_
        if you think about it your physical screen is a vector
      • whoa man
      • dont think that change helped much really
      • lots of wavy stuff, and breaks down as soon as you have multiple jobs in the column
      • *still lots of wavy lines