Use Airlift Duration in MonitorConfiguration

Author: andythsu

docs/operation.md

@@ -49,11 +49,11 @@ modules:
 ```
 - The router operates based on the stats it receives from the clusters, such as 
 the number of queued and running queries. These values are retrieved at regular 
-intervals. This interval can be configured by setting `taskDelaySeconds` under
-`monitor` section in the config file. The default interval is 60
+intervals. This interval can be configured by setting `taskDelay` under
+`monitor` section in the config file. The default interval is 1 minute
 ```yaml
 monitor:
-  taskDelaySeconds: 10
+  taskDelay: 1m
 ```
 
 ## Monitoring <a name="monitoring"></a>

docs/routers.md

@@ -50,7 +50,7 @@ Use the following steps to create a new router:
 - The router listens to the list of `ClusterStats` via the`updateBackEndStats`
   method.
 - This method is called on regular intervals defined in the config 
-  parameter `monitor=>taskDelaySeconds`.
+  parameter `monitor=>taskDelay`.
 - Each element in the list corresponds to each backend cluster.
 - Only the stats from the healthy cluster are reported, unhealthy clusters are
   not included in the list. If you have three cluster backends and one is
@@ -71,9 +71,6 @@ backendState:
 clusterStatsConfiguration:
   monitorType: UI_API
 
-monitor:
-  taskDelaySeconds: 10
-
 modules:
   - io.trino.gateway.ha.module.QueryCountBasedRouterProvider
 ```

gateway-ha/config.yaml

@@ -18,4 +18,4 @@ clusterStatsConfiguration:
 
 # This can be adjusted based on the coordinator state
 monitor:
-  taskDelaySeconds: 10
+  taskDelay: 1m

gateway-ha/src/main/java/io/trino/gateway/ha/clustermonitor/ActiveClusterMonitor.java

@@ -15,6 +15,7 @@
 
 import com.google.inject.Inject;
 import io.airlift.log.Logger;
+import io.airlift.units.Duration;
 import io.trino.gateway.ha.config.MonitorConfiguration;
 import io.trino.gateway.ha.config.ProxyBackendConfiguration;
 import io.trino.gateway.ha.router.GatewayBackendManager;
@@ -27,21 +28,19 @@
 import java.util.concurrent.Executors;
 import java.util.concurrent.Future;
 import java.util.concurrent.ScheduledExecutorService;
-import java.util.concurrent.TimeUnit;
 
 import static java.util.Objects.requireNonNull;
 
 public class ActiveClusterMonitor
 {
-    public static final int MONITOR_TASK_DELAY_SECONDS = 60;
     public static final int DEFAULT_THREAD_POOL_SIZE = 20;
     private static final Logger log = Logger.get(ActiveClusterMonitor.class);
 
     private volatile boolean isInitialized;
     private final List<TrinoClusterStatsObserver> clusterStatsObservers;
     private final GatewayBackendManager gatewayBackendManager;
 
-    private final int taskDelaySeconds;
+    private final Duration taskDelay;
     private final ClusterStatsMonitor clusterStatsMonitor;
     private final ExecutorService executorService = Executors.newFixedThreadPool(DEFAULT_THREAD_POOL_SIZE);
     private final ScheduledExecutorService scheduledExecutor = Executors.newSingleThreadScheduledExecutor();
@@ -56,13 +55,13 @@ public ActiveClusterMonitor(
         this.clusterStatsMonitor = requireNonNull(clusterStatsMonitor, "clusterStatsMonitor is null");
         this.clusterStatsObservers = requireNonNull(clusterStatsObservers, "clusterStatsObservers is null");
         this.gatewayBackendManager = requireNonNull(gatewayBackendManager, "gatewayBackendManager is null");
-        this.taskDelaySeconds = monitorConfiguration.getTaskDelaySeconds();
+        this.taskDelay = monitorConfiguration.getTaskDelay();
     }
 
     @PostConstruct
     public void start()
     {
-        log.info("Running cluster monitor with connection task delay of %d seconds", taskDelaySeconds);
+        log.info("Running cluster monitor with connection task delay of %s", taskDelay);
         scheduledExecutor.scheduleAtFixedRate(() -> {
             try {
                 log.info("Getting stats for all active clusters");
@@ -89,7 +88,7 @@ public void start()
             catch (Exception e) {
                 log.error(e, "Error performing backend monitor tasks");
             }
-        }, 0, taskDelaySeconds, TimeUnit.SECONDS);
+        }, 0, (long) taskDelay.getValue(), taskDelay.getUnit());
     }
 
     @PreDestroy

gateway-ha/src/main/java/io/trino/gateway/ha/config/MonitorConfiguration.java

@@ -15,15 +15,15 @@
 
 import com.google.common.collect.ImmutableMap;
 import io.airlift.units.Duration;
-import io.trino.gateway.ha.clustermonitor.ActiveClusterMonitor;
 
 import java.util.Map;
 
+import static java.util.concurrent.TimeUnit.MINUTES;
 import static java.util.concurrent.TimeUnit.SECONDS;
 
 public class MonitorConfiguration
 {
-    private int taskDelaySeconds = ActiveClusterMonitor.MONITOR_TASK_DELAY_SECONDS;
+    private Duration taskDelay = new Duration(1, MINUTES);
 
     private int retries;
 
@@ -44,14 +44,14 @@ public class MonitorConfiguration
 
     public MonitorConfiguration() {}
 
-    public int getTaskDelaySeconds()
+    public Duration getTaskDelay()
     {
-        return this.taskDelaySeconds;
+        return this.taskDelay;
     }
 
-    public void setTaskDelaySeconds(int taskDelaySeconds)
+    public void setTaskDelay(Duration taskDelay)
     {
-        this.taskDelaySeconds = taskDelaySeconds;
+        this.taskDelay = taskDelay;
     }
 
     public int getRetries()

gateway-ha/src/test/resources/test-config-template.yml

@@ -13,7 +13,7 @@ clusterStatsConfiguration:
   monitorType: INFO_API
 
 monitor:
-  taskDelaySeconds: 1
+  taskDelay: 1s
 
 extraWhitelistPaths:
   - '/v1/custom.*'

gateway-ha/src/test/resources/test-config-with-routing-rules-api.yml

@@ -12,7 +12,7 @@ clusterStatsConfiguration:
   monitorType: INFO_API
 
 monitor:
-  taskDelaySeconds: 1
+  taskDelay: 1s
 
 extraWhitelistPaths:
   - '/v1/custom.*'

gateway-ha/src/test/resources/test-config-with-routing-template.yml

@@ -12,7 +12,7 @@ clusterStatsConfiguration:
   monitorType: INFO_API
 
 monitor:
-  taskDelaySeconds: 1
+  taskDelay: 1s
 
 extraWhitelistPaths:
   - '/v1/custom.*'

gateway-ha/src/test/resources/test-config-without-x-forwarded-template.yml

@@ -12,7 +12,7 @@ clusterStatsConfiguration:
   monitorType: INFO_API
 
 monitor:
-  taskDelaySeconds: 1
+  taskDelay: 1s
 
 extraWhitelistPaths:
   - '/v1/custom.*'