Fork me on GitHub
#onyx
<
2017-10-11
>
eriktjacobsen03:10:45

Are there any special un-listed steps for onyx-dashboard? I'm running 0.10.0, I grabbed the 0.10.0 dashboard jar from the releases, it starts fine, connects to zookeeper. There is a job running, but when I go to the dashboard, the "select tenancy" dropdown is empty. We only have one tenancy, "1".

eriktjacobsen03:10:25

handful of js errors, Uncaught ReferenceError: FSM is not defined

eriktjacobsen03:10:11

this stands out: WebSocket connection to 'ws://[redacted]:3000/chsk?client-id=8bd[redacted]2653b' failed: HTTP Authentication failed; no valid credentials available

lucasbradstreet04:10:24

@eriktjacobsen That’s odd, sounds like a bug, given the reference error and the auth failure, though we don’t actually use any auth on the dashboard. I don’t really have time to look into it right now, but if you figure anything out let me know.

ben.mumford15:10:40

i have a task that processes a batch of segments. some will fail with a possibly transient error, some will fail with a non-transient error and others may pass. how can i log and retry the segments that are failing due to a transient error, just log those that are failing for non-transient reasons and allow all others in the batch to proceed?

michaeldrogalis16:10:20

@ben.mumford Is it a task that’s actually running in batch mode?

michaeldrogalis16:10:54

Onyx doesn’t give granularity at the task level to retry only some pieces of data and pass others - its generally assuming that it’s one independent piece of data

michaeldrogalis16:10:35

Batch functions help stave off async activity that needs to block before it comes back, but splitting apart after the result is tough unless you funnel them to a new task that breaks them apart and then does the rerouting

zirmite18:10:15

they ^ are using clojure in a few different layers of their data pipeline

zirmite18:10:24

thought it might be of interest here

zirmite18:10:34

(i’ve never used stitch)

michaeldrogalis21:10:58

@ben.mumford Yeah, thought about it for a bit. Best option is to introduce another task immediately after that splits the big segment into its smaller components, and route on each piece separately.