Empty machine information in Pod
I’m using the REST API to retrieve information about running pods and noticed that the API returns an empty machine field in the responses to both find pod and list pods requests.
Is this expected behavior? If so, is there another way to obtain machine details?
Also, is there any way to fetch current pod utilization — similar to what’s shown in the UI?
At the very least, I’m looking to retrieve GPU memory usage.
Thanks in advance!
10 Replies
Unknown User•3mo ago
Message Not Public
Sign In & Join Server To View
@Jason Yeah, my bad, thnx
Machine info doesn't contain maintenanceStart field in the real response. Doest it mean that maintenanceStart is not set?
And if set - what is the format of the string?
And one more thing, a can't find where public ip is shown and it's empty in the API response while I've selected public ip = true and it's specified as true in the machine details. Can you assist with that please?)
Unknown User•3mo ago
Message Not Public
Sign In & Join Server To View
I noticed this in logs yesterday, it seems as though fairly often the internal service where we get that info from times out
this is the only info that comes from that service, so we don’t give up, we just send the rest of the response without it and log the error
I will see if raising timeout on that api call helps
@nathaniel The API responds within a second, so it doesn’t seem to be blocked by any call that fails with a timeout — or you have some very strict timeout rules in place 🙂
this is the frequency of that error over the past 2 days, looks like it got much worse overnight
@nathaniel interesting thing: portMapping == null until the pod is ready and becomes empty map/object after it's ready
yeah thats expected actually
the mapping is not determined until that point and the service which retrieves the information won't respond with anything until then
(well, not the empty map part)
thats cause it's timing out
But is means that responses are different, while I would expect it would be null in both cases, it's strange that on timeout the value becomes empty map. Or it's a fallback on error while before the start service responds with null, not error?
@nathaniel Hi, are there any estimates of fixing the issue with the port mapping / ip? Or it's better to make a ticket?
awkwardly, this is my last day working at runpod. I pushed an attempt to fix by increasing timeout on the api call, but it seems not to have worked
i made a ticket for our eng team
hopefully gets sorted out