RTXPRO6000 Instance Creation Requests Timing Out (>30 second)

Bug Reports
1

Instance creation requests for `1RTXPRO6000.30V` GPU instances are not returning responses within 30 seconds, causing HTTP timeouts. This is a new issue that has not been observed with other GPU instance types (H100, A100, L40S, etc.) or with RTXPRO6000 instances previously.

## Expected Behavior

Instance creation requests should return responses within 30 seconds, consistent with other GPU instance types and previous RTXPRO6000 behavior.

## Actual Behavior

RTXPRO6000 instance creation requests exceed 30 seconds before returning a response, causing client-side HTTP timeouts.

## Impact

- RTXPRO6000 instances cannot be provisioned

- Users cannot access RTXPRO6000 GPU resources

- This affects workflows requiring this specific GPU type

## Technical Details

- **API Endpoint**: `POST https://api.datacrunch.io/v1/instances`

- **Affected Instance Type**: `1RTXPRO6000.30V`

- **Timeout**: Requests exceed 30-second response time

- **Error Type**: HTTP ReadTimeout (client-side timeout due to slow server response)

## Timeline

This issue began occurring today and was not present in previous RTXPRO6000 instance creation attempts.

2

Update: this is happening with B200 instances too as well now.

3

Using the exact same API request, able to provision other instances at other locations (e.g., H200 at FIN-02). Appears to be yet another FIN-03 API issue…

4

Hi,

Sorry for the late reply, we had to restart the service yesterday around 17:00 (UTC) and the total downtime was roughly 30 minutes. Everything should be back to normal now. Sorry for the inconvenience!

5

OK, let me turn this bug report into a feature request:

Delay Instance Availability Responses Until Services Are Fully Ready

When services are undergoing restarts or are otherwise not yet ready to handle API calls, the https://api.datacrunch.io/v1/instance-availability endpoint can return misleading availability information. This causes downstream issues — for example, clients attempt to provision instances based on this availability, only to encounter timeouts because the backend is not actually ready to accept POST /v1/instances requests.

Proposed Behavior

The instance-availability endpoint should not return availability for any instance type until all relevant services are fully ready to handle API provisioning requests. This would prevent unnecessary failed attempts and improve the robustness of automated client workflows.

6

Thank you for the detailed feedback. The issue has been forwarded to the team, we’ll discuss and come up with a plan to make our daily user’s experience more pleasant.