A proxy or NAT gateway has an idle connection timeout shorter than the Snowflake session duration. A long-running query caused the TCP connection to appear idle during execution, triggering proxy timeout. AWS NAT Gateway idle timeout (350 seconds by default) killing connections during result streaming. Corporate firewall closing connections idle for more than a configured threshold. The JDBC keep-alive is not configured, allowing the TCP connection to go stale

Enable TCP keep-alive at the OS level to prevent idle timeouts from killing active connections.. Set a shorter Snowflake statement timeout to avoid long-running queries that appear idle: ALTER SESSION SET STATEMENT_TIMEOUT_IN_SECONDS = 600.. Increase the proxy or NAT idle timeout to be longer than the maximum expected query duration.. Use connection pooling with keepAlive settings enabled.. Add &networkTimeout= to the JDBC URL to configure socket-level timeout handling.

Medium severityconnectionSnowflake →

Power BI Refresh Error:
JDBC 8

What does this error mean?

The JDBC connection was dropped mid-session with a 'broken pipe' error. This typically happens when an intermediate network appliance (proxy, load balancer, NAT gateway) closes idle or long-lived connections before the Snowflake session ends.

Common causes

1A proxy or NAT gateway has an idle connection timeout shorter than the Snowflake session duration
2A long-running query caused the TCP connection to appear idle during execution, triggering proxy timeout
3AWS NAT Gateway idle timeout (350 seconds by default) killing connections during result streaming
4Corporate firewall closing connections idle for more than a configured threshold
5The JDBC keep-alive is not configured, allowing the TCP connection to go stale

How to fix it

1Enable TCP keep-alive at the OS level to prevent idle timeouts from killing active connections.
2Set a shorter Snowflake statement timeout to avoid long-running queries that appear idle: ALTER SESSION SET STATEMENT_TIMEOUT_IN_SECONDS = 600.
3Increase the proxy or NAT idle timeout to be longer than the maximum expected query duration.
4Use connection pooling with keepAlive settings enabled.
5Add &networkTimeout=<ms> to the JDBC URL to configure socket-level timeout handling.

Frequently asked questions

How do I enable TCP keep-alive for Snowflake JDBC?

Set the JVM system property -Djava.net.preferIPv4Stack=true and configure OS-level TCP keep-alive (net.ipv4.tcp_keepalive_time on Linux). The JDBC driver itself does not have a built-in keep-alive setting.

Does Snowflake have a session idle timeout that could cause this?

Yes — the default session idle timeout is controlled by SESSION_POLICY. However, broken pipe errors are almost always caused by intermediate network appliances timing out before Snowflake does.

Official documentation: https://docs.snowflake.com/en/user-guide/client-connectivity-troubleshooting/error-messages