Forum Discussion

XCT's avatar
XCT
Resolver I
2 months ago
Solved

Snowflake ADBC connector error: Error: ADBC: [Snowflake] arrow/ipc:undefined

Hi everyone,

 

For 2 days, I've receive this error message with a daily refresh in Power BI Service for a semantic model:

 

DataSource.Error: ADBC: [Snowflake] arrow/ipc: could not read message body: read tcp 10.0.4.xx:xxxxx->3.5.xxx.xxx:xxx: wsarecv: An existing connection was forcibly closed by the remote host.. Microsoft.Data.Mashup.ErrorCode = 10941. Microsoft.Data.Mashup.ValueError.DataSourceKind = Snowflake. DataSourcePath = xxx.eu-central-1.snowflakecomputing.com;FR_WH_PRD. . The exception was raised by the IDataReader interface. Please review the error message and provider documentation for further information and corrective action. Table: F_ELEC_CONSUMPTION. Partition: F_ELEC_CONSUMPTION. 0xC14700F0: {"RootActivityId":"XXX"}

 

Context: the semantic model is connected directly to snowflake database using the PowerQuery Connector without any implementation (because it's an "old" dataset). It retrieves some Snowflake views. One of them contains about 20,000,000 rows (since the beginning of the project). I connect the semantic model with a cloud connection with a service account with Key Pair.

With the error message, I undestand that the issue is a timeout on Power BI Service side to resolve queries caused by the ADBC driver. On the Snowflake side, no problem with this log (no timeout with the queries sent).

I've tried to force the connector from nothing (ADBC by default?) to Implementation = "2.0" (with Key Pair) => same issue.

I've tried to force the Implementation = "1.0" (I think the ODBC driver) with SSO (personnal account). Curioulsy, it works (with the same duration of the last successful refresh).

With SSO, I think you can go back to an ODBC driver. But with a Key Pair identification, it's obligatory to have the ADBC driver...

Other Datasets (with less data) connected with Key Pair work fine in Import or DirectQuery mode, no matter.

I read somewhere that Microsoft will switch the ODBC connectors for Snowflake, BigQuery,...to ADBC one during this summer. But, currently, all aren't fixed.
Did you face the same issue?
What are the workarounds?
 
Many thx for your answers!

 

  • hi XCT I found this one and looks like this is a confirmed bug in the Snowflake ADBC driver, not a Power BI timeout or network config issue. Take a look and have a read.
    SNOW-3449105: arrow/ipc: could not read message body · Issue #1781 · snowflakedb/gosnowflake
    Good news is the fix was merged and released with gosnowflake v2.1.0 (released ~June 9, 2026). It adds a Reset() method to ArrowStreamBatch that lets the ADBC driver re-download a chunk after a mid-stream TCP failure without re-executing the entire query. 
    However, this fix still needs to be picked up by Microsoft to incorporate into Power BI Service/Desktop. Their target is around July release cutoff.
    For workaround, You already found this which is ODBC fallback (Implementation 1.0). Note that ODBC uses a completely different data transfer mechanism that handles large result sets more reliably. 

    Definitely worth filing a support ticket referencing SNOW-3449105 (link above) to add more pressure so that the patched driver lands in Power BI Service quicker 🙂

10 Replies

  • hi XCT I found this one and looks like this is a confirmed bug in the Snowflake ADBC driver, not a Power BI timeout or network config issue. Take a look and have a read.
    SNOW-3449105: arrow/ipc: could not read message body · Issue #1781 · snowflakedb/gosnowflake
    Good news is the fix was merged and released with gosnowflake v2.1.0 (released ~June 9, 2026). It adds a Reset() method to ArrowStreamBatch that lets the ADBC driver re-download a chunk after a mid-stream TCP failure without re-executing the entire query. 
    However, this fix still needs to be picked up by Microsoft to incorporate into Power BI Service/Desktop. Their target is around July release cutoff.
    For workaround, You already found this which is ODBC fallback (Implementation 1.0). Note that ODBC uses a completely different data transfer mechanism that handles large result sets more reliably. 

    Definitely worth filing a support ticket referencing SNOW-3449105 (link above) to add more pressure so that the patched driver lands in Power BI Service quicker 🙂

    • XCT's avatar
      XCT
      Resolver I

       

      I'm glad to hear that this issue is known and not related to a network configuration problem.

       

      It looks like this is an issue with the "new ADBC" driver. Replacing ODBC with ADBC may introduce some problems.

       

      Based on your input, I have opened a support ticket referencing the link you mentioned. Let's wait and see how it progresses.

       

    • XCT's avatar
      XCT
      Resolver I

      Hi rizalard0684,

       

      Following my support ticket, Microsoft has confirmed that the fix will be deployed in the July 2026 release of Power BI Desktop and Power BI Service. We must be patient!

      Thx for your support!

      • NSnoap's avatar
        NSnoap
        Frequent Visitor

        Do we know when the July 2026 release for PBI Service is supposed to go out? It's 7/24 right now, and I just started getting this error for the first time on one of our most important semantic models that needs to refresh daily.

    • NSnoap's avatar
      NSnoap
      Frequent Visitor

      So, quick update on this. I started getting this error on a large model of mine (using just a select * query against a large Snowflake table) this morning. Never had it before today.

       

      Was the new ADBC driver with the fix already pushed to PBI Service? Because if I route the refresh through our on-prem PBI gateway it completes successfully without triggering this new error. And I assume that the ADBC driver being used in that scenario is a bit older than whatever PBI Service uses when I route the refresh for that model through a Shared Cloud Connection instead of our gateway.

       

      I ask because I'm worried that an update to the ADBC driver in PBI Service today is what caused this issue in the first place, whereas it was indicated in this thread that the July 2026 PBI Service update for the ADBC driver was supposed to be the fix.

  • Hi XCT

     

    Were there any changes to configuration on the snowflake side? 

    THe error message "An existing connection was forcibly closed by the remote host" tells me that Pwoer BI connected successfully, and then snowflake killed the connection. 

    Was any sort of IP whitelisting or anything of that nature configured? 

    Do you have other connections to snowflake that are still working? 

    • XCT's avatar
      XCT
      Resolver I

      Hi tayloramy 

      The only change we've made on Snowflake side is for the identification, we've changed all our technical accounts from basic authentification method to KeyPair.

  • v-achippa's avatar
    v-achippa
    Community Support

    Hi XCT,

     

    Thank you for reaching out to Microsoft Fabric Community.

     

    Thank you tayloramy and rizalard0684 for the prompt response.

     

    As we haven’t heard back from you, we wanted to kindly follow up to check if the solution provided by the user's for the issue worked? or let us know if you need any further assistance.

     

    Thanks and regards,

    Anjan Kumar Chippa