add config support delete shard from fe onlyfix

Signed-off-by: drake_wang <[email protected]>

fe/fe-core/src/main/java/com/starrocks/common/Config.java

@@ -2538,6 +2538,16 @@ public class Config extends ConfigBase {
     @ConfField(mutable = true)
     public static boolean meta_sync_force_delete_shard_meta = false;
 
+    /**
+     * Fast deletion large volume of shard meta from FE to lower GC overhead in shared-data mode cluster.
+     * Usually shard number will increase quickly if frequent adding/dropping partitions happened.
+     *
+     * If it is enabled, shard meta will be deleted from FE only.
+     * Users might need to delete tablet data from storage manually.
+     */
+    @ConfField(mutable = true)
+    public static boolean meta_sync_delete_shard_meta_in_fe_only = false;
+
     // ***********************************************************
     // * BEGIN: Cloud native meta server related configurations
     // ***********************************************************

fe/fe-core/src/main/java/com/starrocks/lake/StarMgrMetaSyncer.java

@@ -153,7 +153,6 @@ public static void dropTabletAndDeleteShard(List<Long> shardIds, StarOSAgent sta
                 }
             } catch (DdlException e) {
                 LOG.warn("failed to delete shard from starMgr");
-                continue;
             }
         }
     }
@@ -197,17 +196,33 @@ private void deleteUnusedShardAndShardGroup() {
         long nowMs = System.currentTimeMillis();
         List<Long> emptyShardGroup = new ArrayList<>();
         for (long groupId : diffList) {
-            if (Config.shard_group_clean_threshold_sec * 1000L + Long.parseLong(groupToCreateTimeMap.get(groupId)) < nowMs) {
-                try {
-                    List<Long> shardIds = starOSAgent.listShard(groupId);
-                    if (shardIds.isEmpty()) {
-                        emptyShardGroup.add(groupId);
-                    } else {
-                        dropTabletAndDeleteShard(shardIds, starOSAgent);
+            if (Config.shard_group_clean_threshold_sec * 1000L + Long.parseLong(groupToCreateTimeMap.get(groupId)) >= nowMs) {
+                continue;
+            }
+            try {
+                List<Long> shardIds = starOSAgent.listShard(groupId);
+                if (shardIds.isEmpty()) {
+                    emptyShardGroup.add(groupId);
+                    continue;
+                }
+
+                // delete shard from star manager only, not considering tablet data on be/cn
+                if (Config.meta_sync_delete_shard_meta_in_fe_only) {
+                    LOG.debug("delete shards from starMgr only, shard group: {}", groupId);
+                    try {
+                        starOSAgent.deleteShards(new HashSet<>(shardIds));
+                    } catch (DdlException e) {
+                        LOG.warn("failed to delete shards from starMgr");
                     }
-                } catch (Exception e) {
                     continue;
                 }
+                // drop meta and data
+                long start = System.currentTimeMillis();
+                dropTabletAndDeleteShard(shardIds, starOSAgent);
+                LOG.debug("delete shards from starMgr and FE, shard group: {}, cost: {} ms", groupId,
+                        (System.currentTimeMillis() - start));
+            } catch (Exception e) {
+                LOG.warn("delete shards from starMgr and FE failed, shard group: {}, {}", groupId, e.getMessage());
             }
         }
 

fe/fe-core/src/test/java/com/starrocks/lake/StarMgrMetaSyncerTest.java

@@ -629,6 +629,85 @@ Future<DeleteTabletResponse> deleteTablet(DeleteTabletRequest request) throws Ex
         Config.meta_sync_force_delete_shard_meta = false;
     }
 
+    @Test
+    public void testDeleteSyncerMetaOnly() {
+        boolean originalConfig = Config.meta_sync_delete_shard_meta_in_fe_only;
+        Config.shard_group_clean_threshold_sec = 0;
+        long groupIdToClear = shardGroupId + 1;
+        // build shardGroupInfos
+        List<Long> allShardIds = Stream.of(1000L, 1001L, 1002L, 1003L).collect(Collectors.toList());
+        int numOfShards = allShardIds.size();
+        List<ShardGroupInfo> shardGroupInfos = new ArrayList<>();
+        ShardGroupInfo info = ShardGroupInfo.newBuilder()
+                .setGroupId(groupIdToClear)
+                .putProperties("createTime", String.valueOf(System.currentTimeMillis() - 86400 * 1000))
+                .addAllShardIds(allShardIds)
+                .build();
+        shardGroupInfos.add(info);
+
+        new MockUp<StarOSAgent>() {
+            @Mock
+            public void deleteShardGroup(List<Long> groupIds) throws
+                    StarClientException {
+                for (long groupId : groupIds) {
+                    shardGroupInfos.removeIf(item -> item.getGroupId() == groupId);
+                }
+            }
+            @Mock
+            public List<ShardGroupInfo> listShardGroup() {
+                return shardGroupInfos;
+            }
+
+            @Mock
+            public List<Long> listShard(long groupId) throws DdlException {
+                if (groupId == groupIdToClear) {
+                    return allShardIds;
+                } else {
+                    return Lists.newArrayList();
+                }
+            }
+
+            @Mock
+            public void deleteShards(Set<Long> shardIds) throws DdlException {
+                allShardIds.removeAll(shardIds);
+            }
+        };
+
+        new MockUp<BrpcProxy>() {
+            @Mock
+            public LakeService getLakeService(String host, int port) throws RpcException {
+                return new PseudoBackend.PseudoLakeService();
+            }
+        };
+
+        // mock dropTabletAndDeleteShard failed to delete meta data
+        {
+            new MockUp<PseudoBackend.PseudoLakeService>() {
+                @Mock
+                Future<DeleteTabletResponse> deleteTablet(DeleteTabletRequest request) {
+                    DeleteTabletResponse resp = new DeleteTabletResponse();
+                    resp.status = new StatusPB();
+                    resp.status.statusCode = TStatusCode.INTERNAL_ERROR.getValue();
+                    resp.failedTablets = new ArrayList<>(request.tabletIds);
+                    return CompletableFuture.completedFuture(resp);
+                }
+            };
+
+            Config.meta_sync_delete_shard_meta_in_fe_only = false;
+            Deencapsulation.invoke(starMgrMetaSyncer, "deleteUnusedShardAndShardGroup");
+            Assert.assertEquals(numOfShards, allShardIds.size());
+        }
+
+        // mock success to delete meta after set meta_sync_delete_shard_meta_in_fe_only
+        {
+            Config.meta_sync_delete_shard_meta_in_fe_only = true;
+            Deencapsulation.invoke(starMgrMetaSyncer, "deleteUnusedShardAndShardGroup");
+            Assert.assertEquals(0, allShardIds.size());
+        }
+
+        Config.meta_sync_delete_shard_meta_in_fe_only = originalConfig;
+    }
+
     @Test
     public void testSyncTableMetaInternal() throws Exception {
         long dbId = 100;