Intermittent connectivity issues between Power BI and Databricks

[Malthe]

We're experiencing connectivity issues between Power BI (refreshing a dataset) and Databricks.

On the Databricks side, we're using a SQL Warehouse (serverless), which reports queries timing out:

Query has been timed out due to inactivity.

On the Power BI side, we see the connection closing unexpectedly: 

Data source error: DataSource.Error: ODBC: ERROR [HY000] [Microsoft][ThriftExtension] (14) Unexpected response from server during a HTTP connection: connect() failed: A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.

These are relatively small queries, normally responding in two minutes or less.

What could explain this situation? It's normally working fine and we have made no changes.

Related issues:

• https://community.databricks.com/t5/administration-architecture/query-has-been-timed-out-due-to-inac...

[Malthe]

For reference, the issue began at 13/12/2025, 02:43 UTC, and ended (last failure) at 19:43 UTC, 17 hours later.

We did nothing to alleviate the issue and the SQL Warehouse seemed to operate normally except when the issue began and concluded when there seems to have been less activity:

Microsoft has not reported any issues on their Power BI status page. Previously, we've been able to resolve a similar issue by switching from a serverless to classic warehouse, suggesting that indeed the issue lies with Databricks.

[emma_s · Accepted solution]

Hi, 
There are a few things that could cause the issues you've outlined:

1. Temporary resource scarcity (compute, network bandwidth, or concurrency limits on the warehouse) can lead to queries taking longer than usual or connections being dropped unexpectedly—even for small queries.
2. The errors reference Thrift/ODBC, suggesting the issue may sometimes be within the gateway (Power BI Gateway) or driver stack, especially if running locally or in a virtualized environment. If Power BI or its Gateway has been restarted, updated, or is over-utilized, driver sessions may drop unexpectedly.
3. Occasional Azure or AWS cloud maintenance or network events can disrupt serverless Databricks endpoint connectivity. These are typically short-lived but may cause the symptoms you see.

Short-Term Troubleshooting

• Review Warehouse Event Logs: Look for corresponding error or warning events around query timeout and Power BI access attempts.
• Monitor Network Health: Use network monitoring tools or test manual ODBC connections from the Power BI host to Databricks to rule out endpoint or internet issues.

Longer-Term Mitigation Steps

• Increase Client Timeout Settings: In Power BI (and Databricks), raise query and inactivity timeout thresholds to tolerate brief network or resume delays.
• Schedule Warehouse Warming: If feasible, start the SQL Warehouse ahead of scheduled Power BI refreshes to guarantee readiness.
• Update Drivers/Gateways: Ensure you have the latest ODBC driver version and that Power BI Gateway is up to date and healthy.
• Review Regional Status: Check with your cloud provider for any announcements or incidents in the target region.

[Malthe]

This is the capacity in Power BI:

In the timeframe where we had connectivity issues, the capacity was plenty, but also we can see how the interaction was:

1. Queries were executed fine by Databricks, taking the normal amount of time (from query history);
2. Somehow, by the time that the results were supposed to be downloaded to Power BI and processed, this didn't happen.

Now, is that an issue in Power BI or Databricks–? That's hard to answer. But one fact remains: we basically don't see this kind of issue and we're currently refreshing data every hour and have been for years. So what could explain that suddenly, for 17 hours the process stops working?

[emma_s]

From everything you've said, it's most likely some kind of Cloud issue with regional availability or network connectivity between Power BI and Databricks. If you want to pursue further I would suggest you log a support ticket with your cloud provider.