As we are on tcp and only local connection we can afford to be a bit more aggressive on failing a backend.

5 files changed, 12 insertions, 15 deletions

diff --git a/container-search/src/main/java/com/yahoo/prelude/Pong.java b/container-search/src/main/java/com/yahoo/prelude/Pong.java
index 55ae2c77fdb..017fceb2942 100644
--- a/container-search/src/main/java/com/yahoo/prelude/Pong.java
+++ b/container-search/src/main/java/com/yahoo/prelude/Pong.java
@@ -9,6 +9,7 @@ import java.util.Optional;
 import com.yahoo.fs4.PongPacket;
 import com.yahoo.search.result.ErrorMessage;
 import com.yahoo.search.statistics.ElapsedTime;
+import com.yahoo.container.protect.Error;
 
 /**
  * An answer from Ping.
diff --git a/container-search/src/main/java/com/yahoo/prelude/cluster/ClusterSearcher.java b/container-search/src/main/java/com/yahoo/prelude/cluster/ClusterSearcher.java
index 8341cf6884c..2709755da5d 100644
--- a/container-search/src/main/java/com/yahoo/prelude/cluster/ClusterSearcher.java
+++ b/container-search/src/main/java/com/yahoo/prelude/cluster/ClusterSearcher.java
@@ -82,8 +82,6 @@ public class ClusterSearcher extends Searcher {
     // Mapping from rank profile names to document types containing them
     private final Map<String, Set<String>> rankProfiles = new HashMap<>();
 
-    private final boolean failoverToRemote;
-
     private final FS4ResourcePool fs4ResourcePool;
 
     private final long maxQueryTimeout; // in milliseconds
@@ -122,7 +120,6 @@ public class ClusterSearcher extends Searcher {
         clusterModelName = clusterConfig.clusterName();
         QrSearchersConfig.Searchcluster searchClusterConfig = getSearchClusterConfigFromClusterName(qrsConfig, clusterModelName);
         documentTypes = new LinkedHashSet<>();
-        failoverToRemote = clusterConfig.failoverToRemote();
 
         String eventName = clusterModelName + ".cache_hit_ratio";
         cacheHitRatio = new Value(eventName, manager, new Value.Parameters().setNameExtension(false)
@@ -234,7 +231,6 @@ public class ClusterSearcher extends Searcher {
 
     /** Do not use, for internal testing purposes only. **/
     ClusterSearcher(Set<String> documentTypes) {
-        this.failoverToRemote = false;
         this.documentTypes = documentTypes;
         monitor = new ClusterMonitor(this, new QrMonitorConfig(new QrMonitorConfig.Builder()), Optional.of(new VipStatus()));
         cacheHitRatio = new Value("com.yahoo.prelude.cluster.ClusterSearcher.ClusterSearcher().dummy",
@@ -257,10 +253,6 @@ public class ClusterSearcher extends Searcher {
         return new CacheControl(config.cacheSize(), config.cacheTimeout());
     }
 
-    public String getClusterModelName() {
-        return clusterModelName;
-    }
-
     ClusterMonitor getMonitor() {
         return monitor;
     }
diff --git a/container-search/src/main/java/com/yahoo/prelude/cluster/MonitorConfiguration.java b/container-search/src/main/java/com/yahoo/prelude/cluster/MonitorConfiguration.java
index f3360d1b8d2..cd354b8451a 100644
--- a/container-search/src/main/java/com/yahoo/prelude/cluster/MonitorConfiguration.java
+++ b/container-search/src/main/java/com/yahoo/prelude/cluster/MonitorConfiguration.java
@@ -12,7 +12,7 @@ public class MonitorConfiguration {
     /**
      * The interval in ms between consecutive checks of the monitored nodes
      */
-    private final long checkInterval = 1000;
+    private final long checkInterval = 100;
 
     /**
      * The number of milliseconds to attempt to complete a request before giving
diff --git a/container-search/src/main/java/com/yahoo/prelude/cluster/NodeMonitor.java b/container-search/src/main/java/com/yahoo/prelude/cluster/NodeMonitor.java
index 5ccaff0f198..298eda9b201 100644
--- a/container-search/src/main/java/com/yahoo/prelude/cluster/NodeMonitor.java
+++ b/container-search/src/main/java/com/yahoo/prelude/cluster/NodeMonitor.java
@@ -66,15 +66,16 @@ public class NodeMonitor {
     public void failed(ErrorMessage error) {
         long respondedAt = System.currentTimeMillis();
 
-        if (error.getCode() == BACKEND_COMMUNICATION_ERROR.code 
-            || error.getCode() == NO_ANSWER_WHEN_PINGING_NODE.code)
-        {
+        if (error.getCode() == NO_ANSWER_WHEN_PINGING_NODE.code) {
             // Only count not being able to talk to backend at all
             // as errors we care about
             if ((respondedAt - succeededAt) > 10000) {
                 this.searchNodesOnline = false;
                 setWorking(false, "Not working for 10 s: " + error.toString());
             }
+        } else if (error.getCode() == BACKEND_COMMUNICATION_ERROR.code) {
+            this.searchNodesOnline = false;
+            setWorking(false, "Backend communication error: " + error.toString());
         } else {
             succeededAt = respondedAt;
         }
diff --git a/container-search/src/main/java/com/yahoo/search/cluster/ClusterSearcher.java b/container-search/src/main/java/com/yahoo/search/cluster/ClusterSearcher.java
index 4be8d571b0f..f901194d7e6 100644
--- a/container-search/src/main/java/com/yahoo/search/cluster/ClusterSearcher.java
+++ b/container-search/src/main/java/com/yahoo/search/cluster/ClusterSearcher.java
@@ -3,11 +3,9 @@ package com.yahoo.search.cluster;
 
 import com.yahoo.component.ComponentId;
 import com.yahoo.container.protect.Error;
-import com.yahoo.fs4.mplex.Backend;
 import com.yahoo.log.LogLevel;
 import com.yahoo.prelude.Ping;
 import com.yahoo.prelude.Pong;
-import com.yahoo.prelude.fastsearch.FastSearcher;
 import com.yahoo.yolean.Exceptions;
 import com.yahoo.search.Query;
 import com.yahoo.search.Result;
@@ -16,7 +14,12 @@ import com.yahoo.search.result.ErrorMessage;
 import com.yahoo.search.searchchain.Execution;
 
 import java.util.List;
-import java.util.concurrent.*;
+import java.util.concurrent.Callable;
+import java.util.concurrent.ExecutionException;
+import java.util.concurrent.Executor;
+import java.util.concurrent.FutureTask;
+import java.util.concurrent.TimeUnit;
+import java.util.concurrent.TimeoutException;
 
 /**
  * Implements clustering (failover and load balancing) over a set of client