[java-bigquery] getQueryResults does not retry a SocketTimeoutException (read timeout) #14579

Issue Details

BigQueryBaseService.BIGQUERY_EXCEPTION_HANDLER retries ConnectException, UnknownHostException and SocketException, but not SocketTimeoutException. So when one getQueryResults call inside Job.waitFor takes longer than the HTTP read timeout, the query fails at once with BigQueryException: Read timed out, after one attempt. A read timeout on this idempotent poll is as transient as a SocketException, and the job itself is still running.

Job.waitFor calls getQueryResults with no timeoutMs, so BigQuery holds each call for up to 10 s. java-bigquery sets the HTTP read timeout to 60,000 ms by default. A poll that gets no answer within 60 s, most likely a connection that died, fails the whole query, although the job is still running and the next poll would succeed. A longer read timeout does not help a dead connection. Only a retry does.

Edited 2026-10-04: an earlier version said the default read timeout is 20,000 ms. It is 60,000 ms.

Today a second defect hides this one: gax turns the SocketTimeoutException into a thread interrupt, so the caller sees BigQueryException: java.lang.InterruptedException instead (#12860, PR #14578). With that gax fix applied, the same stall fails with Read timed out after one attempt. That remaining failure is this issue.

Expected: getQueryResults (and other idempotent reads) retries a SocketTimeoutException under the same retry settings as a SocketException.

Environment

• OS Type and Version: macOS 26.6.2 (arm64)
• Java Version and JDK Vendor: OpenJDK 25 (2025-09-16 LTS)
• Deployment: a JVM server that calls BigQuery through com.google.cloud:google-cloud-bigquery, with Application Default Credentials.

Dependencies

• Client library: com.google.cloud:google-cloud-bigquery 2.54.1 (not using libraries-bom). The handler is the same on main today.
• Gax: gax and gax-httpjson 2.84.0
• Auth: google-auth-library-oauth2-http 1.37.1 to 1.51.0
• Google-Http-Java-Client: 1.47.1 to 2.2.0

Reproducer

1. Build a BigQuery client with a short HTTP read timeout, for example 2 s, through HttpTransportOptions.
2. Run a query, and call job.waitFor() while BigQuery holds getQueryResults (any query that takes more than 2 s).
3. With a gax that has the fix(gax): do not treat a SocketTimeoutException as a thread interrupt #14578 fix, waitFor throws BigQueryException: Read timed out (cause java.net.SocketTimeoutException) after one attempt. With current gax, it throws BigQueryException: java.lang.InterruptedException.

The same wait with the default timeout succeeds. The handler in question is in google-cloud-bigquery/src/main/java/com/google/cloud/bigquery/BigQueryBaseService.java:

.abortOn(RuntimeException.class)
.retryOn(java.net.ConnectException.class)
.retryOn(java.net.UnknownHostException.class)
.retryOn(java.net.SocketException.class)