Don't throw error when updateStatus has not run yet

When Manager.StatusThread.updateStatus has not run yet, then the set of current tablet server information will be empty.
dlmarion · Oct 27, 2023 · 40b41d1 · 40b41d1
1 parent 759b5d5
commit 40b41d1
Show file tree

Hide file tree

Showing 3 changed files with 27 additions and 9 deletions.
diff --git a/core/src/main/java/org/apache/accumulo/core/manager/balancer/AssignmentParamsImpl.java b/core/src/main/java/org/apache/accumulo/core/manager/balancer/AssignmentParamsImpl.java
@@ -34,8 +34,13 @@
 import org.apache.accumulo.core.spi.balancer.TabletBalancer;
 import org.apache.accumulo.core.spi.balancer.data.TServerStatus;
 import org.apache.accumulo.core.spi.balancer.data.TabletServerId;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
 
 public class AssignmentParamsImpl implements TabletBalancer.AssignmentParameters {
+
+  private static final Logger LOG = LoggerFactory.getLogger(AssignmentParamsImpl.class);
+
   private final SortedMap<TabletServerId,TServerStatus> currentStatus;
   private final Map<TabletId,TabletServerId> unassigned;
   private final Map<TabletId,TabletServerId> assignmentsOut;
@@ -50,16 +55,23 @@ public static AssignmentParamsImpl fromThrift(
       Map<KeyExtent,TServerInstance> unassigned, Map<KeyExtent,TServerInstance> assignmentsOut) {
 
     SortedMap<TabletServerId,TServerStatus> currentStatusNew = new TreeMap<>();
-    currentStatus.forEach((tsi, status) -> currentStatusNew.put(new TabletServerIdImpl(tsi),
-        TServerStatusImpl.fromThrift(status)));
-
     Map<String,Set<TabletServerId>> tserverGroups = new HashMap<>();
-    currentTServerGrouping.forEach((k, v) -> {
+    currentTServerGrouping.forEach((group, serversInGroup) -> {
       Set<TabletServerId> servers = new HashSet<>();
-      v.forEach(tsi -> servers.add(TabletServerIdImpl.fromThrift(tsi)));
-      tserverGroups.put(k, servers);
+      serversInGroup.forEach(tsi -> {
+        TabletServerIdImpl id = TabletServerIdImpl.fromThrift(tsi);
+        if (currentStatus.containsKey(tsi)) {
+          currentStatusNew.put(id, TServerStatusImpl.fromThrift(currentStatus.get(tsi)));
+          servers.add(id);
+        } else {
+          LOG.debug("Dropping tserver {} from group as it's not in set of all servers", id, group);
+        }
+      });
+      tserverGroups.put(group, servers);
     });
 
+    LOG.debug("TServer groups for balancer assignment: {}", tserverGroups);
+
     Map<TabletId,TabletServerId> unassignedNew = new HashMap<>();
     unassigned.forEach(
         (ke, tsi) -> unassignedNew.put(new TabletIdImpl(ke), TabletServerIdImpl.fromThrift(tsi)));

diff --git a/core/src/main/java/org/apache/accumulo/core/spi/balancer/SimpleLoadBalancer.java b/core/src/main/java/org/apache/accumulo/core/spi/balancer/SimpleLoadBalancer.java
@@ -356,6 +356,10 @@ private static TableId busiest(Map<String,TableStatistics> tables) {
 
   @Override
   public void getAssignments(AssignmentParameters params) {
+    if (params.currentStatus().isEmpty()) {
+      log.debug("No known TabletServers, skipping tablet assignment for now.");
+      return;
+    }
     params.unassignedTablets().forEach((tabletId, tserverId) -> params.addAssignment(tabletId,
         getAssignment(params.currentStatus(), tserverId)));
   }

diff --git a/core/src/main/java/org/apache/accumulo/core/spi/balancer/TableLoadBalancer.java b/core/src/main/java/org/apache/accumulo/core/spi/balancer/TableLoadBalancer.java
@@ -149,10 +149,12 @@ private SortedMap<TabletServerId,TServerStatus> getCurrentSetForTable(
     tserversInGroup.forEach(tsid -> {
       TServerStatus tss = allTServers.get(tsid);
       if (tss == null) {
-        throw new IllegalStateException("TabletServer " + tsid + " in " + groupNameInUse
-            + " TabletServer group, but not in set of all TabletServers");
+        log.warn(
+            "Excluding TabletServer {}  from group {} because TabletServerStatus is null, likely that Manager.StatusThread.updateStatus has not discovered it yet.",
+            tsid, groupNameInUse);
+      } else {
+        group.put(tsid, tss);
       }
-      group.put(tsid, tss);
     });
     return group;
   }