https://dagster.io/ logo
Powered by
#announcements
  • b

    Ben Gotow

    12/04/2018, 8:50 PM
    sounds good
  • b

    Ben Gotow

    12/04/2018, 8:51 PM
    hey sidenote, I’m hacking the play button to just call 
    startPipelineExecution
    and give me a runId, and then storing the runId and the executionPlan in the localstorage session (so the execution plan is cached at-run-time and not live)
  • b

    Ben Gotow

    12/04/2018, 8:51 PM
    but when I call 
    startPipelineExecution
    the graphql websocket never emits a 
    type:stop
    command so the call to client.mutate hangs - I assume this is where the subscription stuff needs to be put in right?
  • b

    Ben Gotow

    12/04/2018, 8:52 PM
    in that case I’ll just mock the response data and let you put the real thing in there
  • b

    Ben Gotow

    12/04/2018, 8:52 PM
    weird part is that in the 
    /graphql
    environment, it seems to resolve that mutation to a single response just fine
  • s

    schrockn

    12/04/2018, 9:09 PM
    Bah ok this might be the executor problem that @freiksenet foresaw
  • s

    schrockn

    12/04/2018, 9:24 PM
    @Ben Gotow another usability thing to look at at some point
  • s

    schrockn

    12/04/2018, 9:25 PM
    when the edge is highlighted I think the input and the output it connects should also be highlighted. It is still ambiguous
    👍 1
    💯 2
  • b

    Ben Gotow

    12/04/2018, 11:14 PM
    hey folks - what do y’all think of this for logs and execution plan display?
  • b

    Ben Gotow

    12/04/2018, 11:14 PM
  • b

    Ben Gotow

    12/04/2018, 11:14 PM
    the idea would be that you’d click on the failed node to see the last logs associated with that node (eg stack trace) and the section at the bottom just tails the logs
  • s

    schrockn

    12/04/2018, 11:14 PM
    i was imagining the execution node grids to be much smaller
  • s

    schrockn

    12/04/2018, 11:14 PM
    more like the size of the log messages
  • s

    schrockn

    12/04/2018, 11:15 PM
    the information will need to be dense
  • s

    schrockn

    12/04/2018, 11:15 PM
    for non-trivial dags
  • b

    Ben Gotow

    12/04/2018, 11:15 PM
    yeah makes sense - I imagine having hundreds of them isn’t unlikely
  • b

    Ben Gotow

    12/04/2018, 11:15 PM
    is there other data that should go in the nodes?
  • s

    schrockn

    12/04/2018, 11:16 PM
    yes but not sure yet 🙂
  • s

    schrockn

    12/04/2018, 11:16 PM
    i could imagine displaying some sort of thing per-output
  • s

    schrockn

    12/04/2018, 11:17 PM
    a click target where we could end up loading some sort of summary info about them
  • s

    schrockn

    12/04/2018, 11:17 PM
    for expectations, we will likely end up returning metadat about the computation
  • s

    schrockn

    12/04/2018, 11:17 PM
    (e.g. we expected variance less than 2 but it was value X)
  • s

    schrockn

    12/04/2018, 11:18 PM
    i suspect we’ll think of stuff as we use it
  • s

    schrockn

    12/04/2018, 11:18 PM
    one thing that would be cool is the if you click on one elemenet of the execution plan
  • s

    schrockn

    12/04/2018, 11:18 PM
    it filters the logs down to that elementl
  • s

    schrockn

    12/04/2018, 11:18 PM
    *element
  • b

    Ben Gotow

    12/04/2018, 11:19 PM
    yeah I think that’d be really nice
  • b

    Ben Gotow

    12/04/2018, 11:19 PM
    shrunk it down a bit
  • b

    Ben Gotow

    12/04/2018, 11:19 PM
    😮 1
  • b

    Ben Gotow

    12/04/2018, 11:20 PM
    i think we may want to consider making this three columns side-by-side instead of dividing this second panel vertically
1...838485...185Latest