fix SolrStream long-running timeouts

solr/solrj-streaming/src/java/org/apache/solr/client/solrj/io/SolrClientCache.java

@@ -29,20 +29,47 @@
 import org.apache.solr.client.solrj.impl.HttpSolrClient;
 import org.apache.solr.client.solrj.impl.SolrHttpConstants;
 import org.apache.solr.common.AlreadyClosedException;
+import org.apache.solr.common.util.EnvUtils;
 import org.apache.solr.common.util.IOUtils;
 import org.apache.solr.common.util.URLUtil;
 
 /** The SolrClientCache caches SolrClients, so they can be reused by different TupleStreams. */
 public class SolrClientCache implements Closeable {
 
-  // Set the floor for timeouts to 60 seconds.
-  // Timeouts can be increased by setting the system properties defined below.
+  /**
+   * Sentinel value meaning "no practical timeout" for streaming clients.
+   *
+   * <p>Tuple streams (export, analytics, etc.) can run for a very long time. We default both the
+   * total request timeout and idle timeout to this value so that long-lived result streams are not
+   * artificially killed. Lower-level TCP/OS timeouts still apply.
+   */
+  static final long NO_TIMEOUT = Long.MAX_VALUE;
+
+  // Connection timeout floor (only used when we create the underlying transport ourselves).
   private static final int MIN_TIMEOUT = 60000;
   private static final int minConnTimeout =
       Math.max(
           Integer.getInteger(SolrHttpConstants.PROP_CONNECTION_TIMEOUT, MIN_TIMEOUT), MIN_TIMEOUT);
-  private static final int minSocketTimeout =
-      Math.max(Integer.getInteger(SolrHttpConstants.PROP_SO_TIMEOUT, MIN_TIMEOUT), MIN_TIMEOUT);
+
+  /**
+   * Configurable timeouts for clients created by this cache.
+   *
+   * <p>These default to "no timeout" because this cache is used almost exclusively by streaming
+   * expressions that legitimately produce long-running responses. Use the system properties (or Env
+   * var equivalents via EnvUtils) to tighten them if desired for your environment.
+   *
+   * <ul>
+   *   <li>solr.solrclientcache.request.timeout.ms - total time allowed for a request/response
+   *       (including streaming the entire result set).
+   *   <li>solr.solrclientcache.idle.timeout.ms - max idle time between data on the connection
+   *       (affects how long we wait for the first response bytes and gaps during streaming).
+   * </ul>
+   */
+  private static final long streamingRequestTimeoutMs =
+      EnvUtils.getPropertyAsLong("solr.solrclientcache.request.timeout.ms", NO_TIMEOUT);
+
+  private static final long streamingIdleTimeoutMs =
+      EnvUtils.getPropertyAsLong("solr.solrclientcache.idle.timeout.ms", NO_TIMEOUT);
 
   protected String basicAuthCredentials = null; // Only support with the httpJettySolrClient
 
@@ -155,13 +182,26 @@ public synchronized SolrClient getHttpSolrClient(String baseUrl) {
     } else {
       builder.withConnectionTimeout(minConnTimeout, TimeUnit.MILLISECONDS);
     }
-    builder.withIdleTimeout(
-        Math.max(minSocketTimeout, builder.getIdleTimeoutMillis()), TimeUnit.MILLISECONDS);
+
+    // Streaming / TupleStream use cases need very long (or unlimited) timeouts for both total
+    // request duration and idle time between chunks. The header-wait path in Http*SolrClient also
+    // uses the client's idle timeout, so a high value here enables slow-starting long queries.
+    builder.withIdleTimeout(streamingIdleTimeoutMs, TimeUnit.MILLISECONDS);
+    builder.withRequestTimeout(streamingRequestTimeoutMs, TimeUnit.MILLISECONDS);
+
     builder.withOptionalBasicAuthCredentials(basicAuthCredentials);
 
     return builder;
   }
 
+  /**
+   * Visible for testing. Returns the builder that would be used so tests can inspect the configured
+   * timeouts without using reflection or forbidden APIs.
+   */
+  HttpSolrClient.BuilderBase<?, ?> newHttpSolrClientBuilderForTest(String url) {
+    return newHttpSolrClientBuilder(url, null);
+  }
+
   @Override
   public synchronized void close() {
     if (isClosed.compareAndSet(false, true)) {

solr/solrj-streaming/src/test/org/apache/solr/client/solrj/io/SolrClientCacheTest.java

@@ -110,4 +110,20 @@ public void testZkACLsUsedWithDifferentChroot() {
           CloudSolrClient.CloudSolrClientConnection.parse(zkClient().getZkServerAddress()));
     }
   }
+
+  @Test
+  public void testStreamingClientsDefaultToNoPracticalTimeout() {
+    try (SolrClientCache cache = new SolrClientCache()) {
+      // Use the test helper so we can inspect configured values without reflection.
+      var builder = cache.newHttpSolrClientBuilderForTest("http://localhost:8983/solr");
+      Assert.assertEquals(
+          "request timeout for streaming cache clients should default to no practical limit",
+          SolrClientCache.NO_TIMEOUT,
+          builder.getRequestTimeoutMillis());
+      Assert.assertEquals(
+          "idle timeout for streaming cache clients should default to no practical limit",
+          SolrClientCache.NO_TIMEOUT,
+          builder.getIdleTimeoutMillis());
+    }
+  }
 }