Compute Issue In databricks

Question

Compute Issue In databricks

Anubhav Aggarwal (TATA CONSULTANCY SERVICES LTD) 105 Microsoft External Staff

I have tried to start create and start the compute several times but it never starts even after several hours. Do you know what could be the issue here?

SAI JAGADEESH KUDIPUDI 3,640 Reputation points Microsoft External Staff Moderator

2026-07-10T01:30:20.9366667+00:00
Hi @Anubhav Aggarwal (TATA CONSULTANCY SERVICES LTD) ,
It sounds like your Azure Databricks compute/cluster keeps getting stuck and never reaches a “started/running” state, even after a long time. When this happens, the most common causes tend to be service/startup health, resource/quota constraints, or configuration/driver-type/runtime constraints. Here are some practical things you can try (and what to check next):

What to check first

Check for any Databricks service issues (diagnostics)Before digging into your cluster settings, it’s worth running/looking at Databricks service diagnostics to confirm there isn’t an ongoing platform issue affecting your workspace. These diagnostics check for service issues that might impact your resource startup/execution.

If you’re using GPU compute: verify region availability + quota

If the compute you’re trying to start is GPU-enabled, there are additional gotchas:

Ensure the GPU node type is available in your workspace region (GPU accelerated instances can be region-specific; if they’re not available, they may not start).

Ensure you have enough subscription quota for the node type family in the region. In Azure portal, check Quotas → Compute and confirm the quota for the family corresponding to the node type you’re starting.

If the GPU types are unselectable during creation, it can indicate either:

you’re not using a compatible GPU-enabled Databricks Runtime for ML, or

the node type isn’t available in your region.

If it’s not GPU-specific: consider whether startup is blocked by resource/performance constraints

The documentation we have here focuses more on notebook/job execution issues than cluster provisioning specifically, but the key idea is the same: when “start” operations stall or clusters become unresponsive, logs/diagnostics are the fastest way to identify whether the driver is failing due to resource pressure (for example, memory/GC-related behavior during operations). If you can capture cluster event logs for that compute attempt, those usually point you to the root cause.

Kindly confirm whether the issue still persists. If so, please share the details requested in the private message so that we can investigate further and assist you more effectively.

Hope this helps. If you have any follow-up questions, please let me know. I would be happy to help.
SAI JAGADEESH KUDIPUDI 3,640 Reputation points Microsoft External Staff Moderator

2026-07-29T12:14:42.46+00:00

Hi @Anubhav Aggarwal (TATA CONSULTANCY SERVICES LTD) ,
I hope you had a chance to review the information shared earlier, and I hope this information has been helpful! If you still have questions, please let us know what is needed in the comments so the question can be answered.

Answer accepted by question author

Alex Burlachenko 25,110 MVP Volunteer Moderator

hi Anubhav Aggarwal (TATA CONSULTANCY SERVICES LTD) & thx for sharing urs issue here at Q&A portal,

compute that stays in Starting for hours isn't normal. The first place to look is the cluster Event Log, since it usually shows whether the failure is due to capacity, VM allocation, permissions, init scripts, or library installation. If the compute never transitions past Starting, check whether there's an error in the cluster events or in the Azure Activity Log around the same time. If this is happening with a brand new compute as well as existing ones, it could be a workspace-level issue, quota/capacity problem, or a transient platform issue rather than a cluster configuration problem.

https://learn.microsoft.com/azure/databricks/compute/clusters-manage#view-compute-events

If the Event Log doesn't show a clear cause and the compute remains in Starting for hours, I'd open a support case with the workspace ID, cluster ID, region, timestamps, and the cluster Event Log. That will let Microsoft/Databricks check the backend provisioning logs.

rgds,

Alex

&

If my answer was helpful pls mark it and additional thx if u follow me at Q&A portal

and at my blog https://ctrlaltdel.blog/

0 comments

1 additional answer

Your answer

SAI JAGADEESH KUDIPUDI 3,640 Reputation points Microsoft External Staff Moderator

2026-07-29T12:14:42.46+00:00

Hi @Anubhav Aggarwal (TATA CONSULTANCY SERVICES LTD) ,
I hope you had a chance to review the information shared earlier, and I hope this information has been helpful! If you still have questions, please let us know what is needed in the comments so the question can be answered.

Answer 1

Hello Anubhav Aggarwal (TATA CONSULTANCY SERVICES LTD),

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you are having Compute Issue in Databricks remaining unavailable for hours.

Since there is no error, and GPU availability, quota, memory pressure, or a regional incident must not be assumed without supporting evidence. In addition to @SAI JAGADEESH KUDIPUDI and @Alex Burlachenko ’s response, what you can do is to:

Stop recreating the compute and open Azure Databricks > Compute > affected compute > Event log.
Capture the first failed startup event, including the cluster ID, Databricks error code, Azure error message, request or correlation ID, node type, region, and UTC timestamp.
Open the Databricks workspace’s managed resource group > Activity log and correlate the failure with operations from Microsoft.Compute, Microsoft.Network, Microsoft.Resources, or Microsoft.Databricks. Azure Activity Log records VM provisioning and Azure Resource Manager deployment failures. - https://learn.microsoft.com/en-us/azure/databricks/compute/troubleshooting/cluster-error-codes, and https://learn.microsoft.com/azure/databricks/_extras/documents/reference-use-case-lh-federation-for-azure.pdf
If no driver VM was created, correct the reported quota, capacity, policy, authorization, subscription, or VM-allocation failure.
If the driver VM was created but compute did not reach RUNNING, correct the reported bootstrap, DNS, firewall, NSG, UDR, NAT, Private Link, subnet, initialization-script, library, instance-pool, or custom-container failure.
Validate the correction using small CPU-based compute with a supported LTS runtime and without an instance pool, spot instances, GPU, init scripts, additional libraries, custom containers, or custom Spark properties.

Therefore:

For a quota error, verify both the total regional vCPU quota and the exact VM-family quota. For a capacity error, select another supported VM size because available quota does not guarantee physical capacity. - https://learn.microsoft.com/azure/databricks/notebooks/notebooks-manage
For a network or bootstrap error, verify outbound connectivity from both Databricks subnets and permit the required Databricks, Storage, and Event Hubs paths through DNS, NSGs, UDRs, firewalls, and NAT. Microsoft recommends Azure service tags rather than hard-coded address ranges. - https://learn.microsoft.com/en-us/azure/service-health/impacted-resources-outage
For subnet exhaustion, ensure that both dedicated Databricks subnets have sufficient addresses. Each node uses one address in each subnet, while Azure reserves five addresses in each subnet. - https://learn.microsoft.com/azure/azure-portal/quickstart-portal-dashboard-azure-cli

I hope this is helpful. Please! Do not hesitate to let me know if you have any other questions, steps or clarifications.

Please do not close the thread by upvoting and accepting the answer if any part of it is helpful.

Anubhav Aggarwal (TATA CONSULTANCY SERVICES LTD) 105 Reputation points Microsoft External Staff

2026-07-30T19:46:14.28+00:00

Thank you @Sina Salam @Alex Burlachenko @SAI JAGADEESH KUDIPUDI , this was resolved and issue was because of NSG. Thank you for your help.