https://prefect.io logo
d

Deceivious

07/19/2023, 6:47 PM
Feature request: Maybe an expandable section under work pool, that lists work queues along with "Flow run running" / "max allowed flow run in the queue" / "Total scheduled"
In addition, something is wrong with current implementation of page refresh on dashboard. I have large number of flow runs. When I add a tag, it takes ages for the new filter to be reflected on the UI. Its always faster to hard refresh the page than wait for the new data to be populated on the dashboard.
c

Craig Harshbarger

07/19/2023, 8:51 PM
@Deceivious thanks for flagging. If this is still a consistent issue please file a github issue. If you could let me know approximately how long "ages" is that would be helpful.
1
d

Deceivious

07/20/2023, 7:57 AM
Cleared the network log before selecting the tag
vs half a second when hard refresh
I think its more about the transition not being seamless. After I select a tag, the old data is still displayed - not sure if wiping the old data old and showing a loading animation would be better. I feel like nothing is being done in the background as there is no visual indicator.
c

Craig Harshbarger

07/20/2023, 3:28 PM
hmm in my testing the change is almost instant. Am I reading that screenshot correctly that its taking 26 seconds to show you updated data? A loom video would be very useful here if that is an option.
d

Deceivious

07/20/2023, 3:29 PM
Yes. How many flows are we talking about pre and post filter?
In ur case
c

Craig Harshbarger

07/20/2023, 3:35 PM
admittedly not a lot. Hundreds of flow runs, tens of thousands of task runs, and hundreds of thousands of events. Toggling a tag that doesn't exist on any flow runs (so going from all the data to none) and things respond quickly like I'd expect. Does it behave any differently if you change the time span? Like is it also slow changing it from 24h to 8h or 1w?
d

Deceivious

07/20/2023, 3:38 PM
Will test it out
thank you 1
c

Craig Harshbarger

07/20/2023, 3:49 PM
I just tried this in a different workspace that has thousands of flow runs, hundreds of thousands of task runs, and millions of events. And its pretty snappy. What range are your counts?
d

Deceivious

07/20/2023, 3:50 PM
110k flows
c

Craig Harshbarger

07/20/2023, 4:01 PM
It would be helpful to get an idea of the counts the dashboard is showing. For the flow runs and task runs. Also, this is OSS and not cloud correct? Possibly there's something there. I don't have an oss instance with a large dataset but I'll see if I can get access to one
d

Deceivious

07/20/2023, 4:02 PM
making a loom video - seeing how much I can actually share
c

Craig Harshbarger

07/20/2023, 4:07 PM
understood, and thank you!
d

Deceivious

07/20/2023, 4:11 PM
Recording 2023-07-20 181000.mp4
The first attempts are click and wait. 2nd attempt is click and ctrl+shift+R for hard refresh.
c

Craig Harshbarger

07/20/2023, 4:13 PM
very interesting, that isn't a lot of data IMO and should be very responsive. And the fact that it loads quick on a page reload is interesting as well. Okay I'll dig into this. Thanks for answering all my questions!
d

Deceivious

07/20/2023, 4:17 PM
🫡
c

Craig Harshbarger

07/20/2023, 7:44 PM
Is this running over the network with a proxy or anything like that? Or are you accessing the ui on the machine its running on directly?
d

Deceivious

07/20/2023, 7:47 PM
yes there is a proxy
But that doesnt explain the fast loading on hard refresh
c

Craig Harshbarger

07/20/2023, 8:03 PM
Just trying to get a full picture. Would be curious if you have the same issue on the machine without hitting the network though. Just to eliminate pieces
So I am able to reproduce this though its not quite as slow for me as it is for you. I believe its because requests are stalling. Which basically means they are stuck in a queue. My working theory as to why its fast on a page load is because not as many requests get made all at once. But when changing the tag or time span it kicks off a bunch of requests all at once and many of them stall. My best guess as to why mine is faster is my prefect is local so less network time on each request. When you change the tag do you see this in the network tab on a lot of requests? That gray bar and the stalled time
d

Deceivious

07/23/2023, 7:27 AM
I assume it is because my flow count is higher. so more partition/pagination to pull off?
c

Craig Harshbarger

07/24/2023, 4:20 PM
The more flow runs you have the longer each api call related to flow runs takes to some extent. It feels fast on a page load because when the page first loads we incrementally load data and make checks to to see if flow runs exists at all and such. But when you change filters the page tries to update everything at once and requests stall out if you have a lot of data (slower api responses) and more requests stall due to the request limit of http1. What you could do to alleviate this is implement http2/3 and HTTPS for your prefect server. This would remove the bottle neck that http1 imposes which limits the browser to making 6 concurrent requests at a time. In cloud 1 cannot reproduce this issue (since no requests stall). We opened an issue to add some docks around this which you might want to follow. https://github.com/PrefectHQ/prefect/issues/10280
d

Deceivious

07/24/2023, 4:21 PM
I've got like 30 flows running and around 30 other subflows running at any given time.
How would I change it to http 2?
c

Craig Harshbarger

07/24/2023, 4:43 PM
I don't have the exact details on that but that github issue is there so we can add some instructions on how to do so to our docs. You could also start a new slack thread about that and you might find someone else who has some insight on getting that set up.
d

Deceivious

07/24/2023, 6:47 PM
Thanks .🫡
🫡 1
HI @Craig Harshbarger , We are running the prefect server UI behind azure web services so for browser's context it is using HTTPS. The prefect server is hosted on HTTP [non secrure] but the azure web service handles the ssl part. I dont know much about these concept but the assumption in gtihub ticket is that we are using HTTP in browsers context hence the issue right? But since we are using HTTPS [again atleast in browsers context] this should not be an issue?
c

Craig Harshbarger

07/25/2023, 2:01 PM
Not 100% sure. Can you look at the timing for a request that is taking a while and send me a screenshot? Curious if your requests are stalling like mine were
d

Deceivious

07/25/2023, 9:45 PM
image.png
This is the SS of the request that took the longest.
c

Craig Harshbarger

07/26/2023, 3:03 PM
so 10s on the server is pretty slow too. But it also stalled for 12s. Which means it was stuck browser side before it even sent the request.
I'm surprised to see 10s for "Waiting for server response" too though. That makes me wonder if things are getting jammed up server side as well