Feel free to share your thoughts here
#Serverless New UI Feedback
dusk gate
maiden tundra
Previously you could see the states + small logs of the workers and observe the request at the same time. I kind of liked that. But the current one feels more organized and clean.
sonic wyvern
Previously you could see the states + small logs of the workers and observe the ...
good call out, will see what we can do
maiden tundra
Uhmm... Did you also remove the "Stream" button on requests to see the async worker yields? 🤔
sharp spear
looks neat, but more seeems more complex too yeah, overall the pages looks very good
haha just tested and this came up in the logs
hmm interesting new way to deploy from github directly excited to develop with this.
Free build time?
final latch
Feel free to share your thoughts here
Cancel button for requests is missing...
maiden tundra
haha just tested and this came up in the logs
I am also getting a lot of 404 errors for https://hapi.runpod.net while inspecting the browser dev console and working with the endpoint. Sometimes, it freezes the UI or doesn't apply changes in the new release, even when the UI says it applied it successfully. Does this also happen to you, or is it just my browser?
sharp spear
Idk haven't tried any change but the ui works fine I guess
Just checking around, the sending requests
sharp spear
nvm'
sonic wyvern
`Cancel` button for requests is missing...
already planned to have it back
sharp spear
not sure but seems like the network storage config in serverless endpoints reset ( in the form) everytime i want to change something
icy crane
can we pls have the template id in the dashboard again? It can be someplace deeper but currently there’s no way to see which endpoint is relying on which template if two templates have same docker image tag
oak cedar
I liked the old UI
Old one had all things at one place, requests count, workers status, logs etc..,
icy crane
one other thing I found. if you have a big container start command, you can't scroll down to set override(overflow: scroll probably works as fix)
it works if you have a small start command(probably 2-3 lines).
not really related to serverless but I assume the redesign as a whole is related
shy cape
one other thing I found. if you have a big container start command, you can't sc...
what browser are you using? if you dont mind
icy crane
what browser are you using? if you dont mind
chrome
its on the pod template override btw, haven't checked if it also happens on the edit pod for an existing pod
shy cape
and OS? (i ask because this is proving difficult to replicate on my end)
icy crane
macOs
yea looks like a mac only thing, I remoted into my windows pc and I don't see it there
shy cape
i can't seem to replicate this on windows or macOS on chrome or safari
icy crane
i can't seem to replicate this on windows or macOS on chrome or safari
weird, I still see it
sharp spear
Mac OS has cool Background and screen recording tools lel
icy crane
Mac OS has cool Background and screen recording tools lel
oh nah its screenstudio
the vid recording thing you see on all indiehacker landing pages lol
dusk gate
I’m also unable to replicate this issue. Could it be caused by a browser extension or something? You might want to try using a different browser or switching to incognito mode
fringe nimbus
Am I the only one who gets this errors on each request after update?
icy crane
I’m also unable to replicate this issue. Could it be caused by a browser extensi...
hmm I don't see this on incognito, I guess its a me-only thing then. I can still save by selecting the template name input and pressing enter, a bit annoying but workable
dusk gate
Am I the only one who gets this errors on each request after update?
We’re aware of this bug, which occurs when you check the status of a request but are not the team owner. For now, you can use Postman or another tool to pull the status. We’re working on a fix and expect to release it soon, possibly later today.
late zinc
The screen glitches too much while using the streaming logs...
sharp spear
btw new ui have bug here, this should be scrollable, the two buttons there will dissapear when you expand environment variables
shy cape
what browser/os?
sharp spear
mine? windows 10, MS edge
shy cape
mine? windows 10, MS edge
what's it look like now?
lone lagoon
I think it looks great but before I could see the requests in real time (e.g. 1 in progress, 2 queued etc), did you remove that ? It was SUPER useful.
shy cape
that hasn't been removed no
shy cape
right where it used to be, it's labeled "jobs"
shy cape
yep it says jobs, with the count of completed jobs
shy cape
you can hover over the smaller sections for exact numbers. though admittedly this display method works better for workers rather than job counts, and some changes in the near future should address that
lone lagoon
what exact number? The exact number of workers? It still doesn't show the number of Jobs in queue, in progress, etc
ok I hope that feature is brought back, I don't see any reason to not display that info, and it's quite useful
expecially for testing
lunar reef
It is much cleaner, which is apparently the intent, but makes it much more difficult to access critical data when troubleshooting problems. Evidence of worker misbehavior gets buried in various tabs and there is no longer a single overview of the state of the endpoint. Information that used to be available with one or two clicks now requires three or four or more to get to, and then it must be navigated away from to review other relevant information. Observations that used to be present on one view concurrently now require swapping between multiple views and sometimes numerous clicks.
shy cape
It is much cleaner, which is apparently the intent, but makes it much more diffi...
can you provide some examples of flows that were possible before but now are not? what's a specific thing that is now more tedious in your eyes?
lunar reef
Absolutely!
Previously, I could review requests on the bottom and queue state while seeing the worker state information on the top, and two clicks got me into specific worker logs.
Now I can see queue state on Metrics tab, but getting to the specific worker logs requires going to the workers tab, which hides the queue state and requests info.
The states would also previously update behind the log view overlay and allow me to compare current queue state, run state, and log tailings. Now that is not an option.
For reference, I'm prototyping against several providers and I keep gravitating back to Runpod because of the cost savings, but I'm encountering more and more problems recently that are forcing me to try to diagnose and evaluate more deeply. This is now more diffcult. But the concern parts are for a different thread.
shy cape
Absolutely!
Previously, I could review requests on the bottom and queue state w...
I see, re: requests + workers. You can see which worker is running your request in the requests table (and click it to navigate to that worker). is the ask that you are able to tail logs as well as seeing request state?
re: log view & states, i assume you are referring to the worker logs in the modal view, would this be solved by solving the former issue?
also, thanks for taking the time to give detailed feedback like this, its how we improve!
lunar reef
A: Yes. Having real-time information for transient data has frequently proven to be invaluable, especially with certain GPU worker types.
B: Yes. Regardless of the method, as long as workers are prone to misbehavior, being able to look at everything at the same time is important.
And you're quite welcome.
(Sorry about the delay, dealing with local things at the moment)
shy cape
alright, will take this into account as we continue to iterate on this design, thanks again
lunar reef
Thank you! I would love to bring our business here. I just need to get a reliable prototype over time since the idea is that problems scale along with use.
Oh, also for vLLM endpoints, having the URLs of the OpenAI generation API visible somewhere again would be extremely helpful for when I need somebody who is not me to make a code change here.
shy cape
Oh, also for vLLM endpoints, having the URLs of the OpenAI generation API visibl...
yep this is something I'm actively ideating on, surfacing openai endpoints
lunar reef
Wonderful. Trust me, I absolutely understand that "Just because it worked before doesn't mean it's easy or trivial to reimplement on the new system".
maiden tundra
Or like... having the OpenAI endpoint documented or mentioned somewhere. There's still only documentation about how to use the official VLLM image, but no information whatsoever for devs that RunPod can accept and return raw payloads when using /openai/* paths on any image.
sonic wyvern
we plan to make further changes in coming weeks, we have heard similar feedback and plan to make it simpler to get to the critical stuff
tulip rain
Hello! I'm primarily a serverless customer and I just wanted to share that I'm loving the new UI. I agree with some of the points suggested by others in this chat concerning troubleshooting but for me, I'm enjoying the redesign. Extra kudos to whoever designed the worker tab, I like the hexagonal eye-candy and being able to see worker locations in a table is a nice plus. The only thing I'd like to see added is a way to see how many requests are in queue within an endpoint's details instead of only in the serverless overview. Great redesign overall :)
sonic wyvern
Hello! I'm primarily a serverless customer and I just wanted to share that I'm l...
changes coming in upcoming weeks to provide an overview tab to give you birds eye view
icy crane
one bug I've been having with datacenter selection, basically if you changed your datacenter selection and wait like 10 seconds to do smth before saving endpoint the changes are reversed. guessing its because of the API sending new data updating the select component. it happens around the end of the video
also would be helpful if there is a button to select all datacenters instead of doing it manually
lone lagoon
Please show a real timestamp here. Like this is useless.
silver willow
In this window, the 3000 one isn't clickable because it thinks it isn't ready. In the old UI, you could click it anyway to give it a go, but now you can't. Thing is, it is ready so now I have to click one of the other ones and then edit the URL.
sonic wyvern
In this window, the 3000 one isn't clickable because it thinks it isn't ready. I...
its being fixed
sonic wyvern
Please show a real timestamp here. Like this is useless.
got it
shy cape
In this window, the 3000 one isn't clickable because it thinks it isn't ready. I...
the button should be clickable, this particular issue was fixed yesterday, could you verify that you cannot click on unready ports? it seems like that button is clickable from the screenshot you provide
silver willow
the button should be clickable, this particular issue was fixed yesterday, could...
Hey this works now - thanks!
sharp spear
#🎤｜general message
silver willow
I really feel like the log viewer got worse - there's plenty of room to show the full log line but the width is artificially limited to like less than half the available space. Could a more sensible style be possible here?
shy cape
what screen size? that seems like an old version of that modal
sonic wyvern
that looks like pod logs, not serverless
maiden tundra
The "Upgrade Total Workers" button popup in serverless does not show the required balance to upgrade anymore. It says only, "A minimum balance of $ is required." Any reason why?
sonic wyvern
The "Upgrade Total Workers" button popup in serverless does not show the require...
ill check
pure heron
ill check
Any news on that? I would like to upgrade but I don't know how much money I need
sonic wyvern
pm me email, ill do that for you, we will have fix out early this week