[Feature] Unify load profile through analyze profile & support sessio…

…n/table profile collect granularity

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

docs/en/administration/management/FE_configuration.md

@@ -4549,9 +4549,9 @@ ADMIN SET FRONTEND CONFIG ("key" = "value");
 -->
 
 <!--
-##### stream_load_profile_collect_second
+##### load_profile_collect_threshold_second
 
-- Default: 10
+- Default: 0
 - Type: Long
 - Unit: Seconds
 - Is mutable: Yes

docs/zh/administration/management/FE_configuration.md

@@ -4555,9 +4555,9 @@ Compaction Score 代表了一个表分区是否值得进行 Compaction 的评分
 -->
 
 <!--
-##### stream_load_profile_collect_second
+##### stream_load_profile_collect_threshold_second
 
-- 默认值：10
+- 默认值：0
 - 类型：Long
 - 单位：Seconds
 - 是否动态：是

fe/fe-core/src/main/java/com/starrocks/alter/AlterJobExecutor.java

@@ -373,6 +373,9 @@ public Void visitModifyTablePropertiesClause(ModifyTablePropertiesClause clause,
             } else if (properties.containsKey(PropertyAnalyzer.PROPERTIES_MUTABLE_BUCKET_NUM)) {
                 schemaChangeHandler.updateTableMeta(db, tableName.getTbl(), properties,
                         TTabletMetaType.MUTABLE_BUCKET_NUM);
+            } else if (properties.containsKey(PropertyAnalyzer.PROPERTIES_ENABLE_LOAD_PROFILE)) {
+                schemaChangeHandler.updateTableMeta(db, tableName.getTbl(), properties,
+                        TTabletMetaType.ENABLE_LOAD_PROFILE);
             } else if (properties.containsKey(PropertyAnalyzer.PROPERTIES_BINLOG_ENABLE) ||
                     properties.containsKey(PropertyAnalyzer.PROPERTIES_BINLOG_TTL) ||
                     properties.containsKey(PropertyAnalyzer.PROPERTIES_BINLOG_MAX_SIZE)) {

fe/fe-core/src/main/java/com/starrocks/alter/SchemaChangeHandler.java

@@ -2100,6 +2100,11 @@ public void updateTableMeta(Database db, String tableName, Map<String, String> p
             if (mutableBucketNum == olapTable.getMutableBucketNum()) {
                 return;
             }
+        } else if (metaType == TTabletMetaType.ENABLE_LOAD_PROFILE) {
+            boolean enableLoadProfile = Boolean.parseBoolean(properties.get(PropertyAnalyzer.PROPERTIES_ENABLE_LOAD_PROFILE));
+            if (enableLoadProfile == olapTable.enableLoadProfile()) {
+                return;
+            }
         } else if (metaType == TTabletMetaType.PRIMARY_INDEX_CACHE_EXPIRE_SEC) {
             int primaryIndexCacheExpireSec = Integer.parseInt(properties.get(
                     PropertyAnalyzer.PROPERTIES_PRIMARY_INDEX_CACHE_EXPIRE_SEC));

fe/fe-core/src/main/java/com/starrocks/catalog/OlapTable.java

@@ -285,6 +285,8 @@ public enum OlapTableState {
     protected Map<Long, Long> doubleWritePartitions = new HashMap<>();
 
     // Both the following two flags are used by StarMgrMetaSyncer
+    private long lastCollectProfileTime = 0;
+
     // The flag is used to indicate whether the table shard group has changed.
     public AtomicBoolean isShardGroupChanged = new AtomicBoolean(false);
     // The flag is used to indicate whether the table is doing automatic bucketing.
@@ -2552,6 +2554,23 @@ public void setMutableBucketNum(long bucketNum) {
         tableProperty.buildMutableBucketNum();
     }
 
+    public Boolean enableLoadProfile() {
+        if (tableProperty != null) {
+            return tableProperty.enableLoadProfile();
+        }
+        return false;
+    }
+
+    public void setEnableLoadProfile(boolean enableLoadProfile) {
+        if (tableProperty == null) {
+            tableProperty = new TableProperty(new HashMap<>());
+        }
+        tableProperty
+                .modifyTableProperties(PropertyAnalyzer.PROPERTIES_ENABLE_LOAD_PROFILE,
+                        Boolean.valueOf(enableLoadProfile).toString());
+        tableProperty.buildEnableLoadProfile();
+    }
+
     public TWriteQuorumType writeQuorum() {
         if (tableProperty != null) {
             return tableProperty.writeQuorum();
@@ -3253,6 +3272,12 @@ public Map<String, String> getProperties() {
             properties.put(PropertyAnalyzer.PROPERTIES_MUTABLE_BUCKET_NUM, mutableBucketNum.toString());
         }
 
+        // enable load profile
+        Boolean enableLoadProfile = enableLoadProfile();
+        if (enableLoadProfile) {
+            properties.put(PropertyAnalyzer.PROPERTIES_ENABLE_LOAD_PROFILE, "true");
+        }
+
         // locations
         Multimap<String, String> locationsMap = getLocation();
         if (locationsMap != null) {
@@ -3529,4 +3554,12 @@ public void unlockCreatePartition(String partitionName) {
         }
     }
 
+    public long getLastCollectProfileTime() {
+        return lastCollectProfileTime;
+    }
+
+    public void updateLastCollectProfileTime() {
+        this.lastCollectProfileTime = System.currentTimeMillis();
+    }
+
 }

fe/fe-core/src/main/java/com/starrocks/catalog/TableProperty.java

@@ -262,6 +262,8 @@ public static String valueList() {
     // the default mutable bucket number
     private long mutableBucketNum = 0;
 
+    private boolean enableLoadProfile = false;
+
     // 1. This table has been deleted. if hasDelete is false, the BE segment must don't have deleteConditions.
     //    If hasDelete is true, the BE segment maybe have deleteConditions because compaction.
     // 2. Before checkpoint, we relay delete job journal log to persist.
@@ -358,6 +360,9 @@ public TableProperty buildProperty(short opCode) {
             case OperationType.OP_MODIFY_MUTABLE_BUCKET_NUM:
                 buildMutableBucketNum();
                 break;
+            case OperationType.OP_MODIFY_ENABLE_LOAD_PROFILE:
+                buildEnableLoadProfile();
+                break;
             case OperationType.OP_MODIFY_BINLOG_CONFIG:
                 buildBinlogConfig();
                 break;
@@ -650,6 +655,13 @@ public TableProperty buildMutableBucketNum() {
         return this;
     }
 
+    public TableProperty buildEnableLoadProfile() {
+        if (properties.get(PropertyAnalyzer.PROPERTIES_ENABLE_LOAD_PROFILE) != null) {
+            enableLoadProfile = Boolean.parseBoolean(properties.get(PropertyAnalyzer.PROPERTIES_ENABLE_LOAD_PROFILE));
+        }
+        return this;
+    }
+
     public TableProperty buildEnablePersistentIndex() {
         enablePersistentIndex = Boolean.parseBoolean(
                 properties.getOrDefault(PropertyAnalyzer.PROPERTIES_ENABLE_PERSISTENT_INDEX, "false"));
@@ -906,6 +918,10 @@ public long getMutableBucketNum() {
         return mutableBucketNum;
     }
 
+    public boolean enableLoadProfile() {
+        return enableLoadProfile;
+    }
+
     public String getStorageVolume() {
         return storageVolume;
     }
@@ -1022,6 +1038,7 @@ public void gsonPostProcess() throws IOException {
         buildPartitionLiveNumber();
         buildReplicatedStorage();
         buildBucketSize();
+        buildEnableLoadProfile();
         buildBinlogConfig();
         buildBinlogAvailableVersion();
         buildDataCachePartitionDuration();

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

@@ -2882,11 +2882,17 @@ public class Config extends ConfigBase {
     public static int routine_load_scheduler_interval_millisecond = 10000;
 
     /**
-     * Only when the stream load time exceeds this value,
+     * Only when the stream/routine load time exceeds this value,
      * the profile will be put into the profileManager
      */
+    @ConfField(mutable = true, aliases = {"stream_load_profile_collect_second"})
+    public static long stream_load_profile_collect_threshold_second = 0;
+
+    /**
+     * The interval of collecting load profile through table granularity
+     */
     @ConfField(mutable = true)
-    public static long stream_load_profile_collect_second = 10; //10s
+    public static long load_profile_collect_interval_second = 0;
 
     /**
      * If set to <= 0, means that no limitation.

fe/fe-core/src/main/java/com/starrocks/common/util/ProfileManager.java

@@ -78,6 +78,10 @@ public class ProfileManager implements MemoryTrackable {
     public static final String DEFAULT_DB = "Default Db";
     public static final String VARIABLES = "Variables";
     public static final String PROFILE_COLLECT_TIME = "Collect Profile Time";
+    public static final String LOAD_TYPE = "Load Type";
+
+    public static final String LOAD_TYPE_STREAM_LOAD = "STREAM_LOAD";
+    public static final String LOAD_TYPE_ROUTINE_LOAD = "ROUTINE_LOAD";
 
     public static final ArrayList<String> PROFILE_HEADERS = new ArrayList<>(
             Arrays.asList(QUERY_ID, USER, DEFAULT_DB, SQL_STATEMENT, QUERY_TYPE,

fe/fe-core/src/main/java/com/starrocks/common/util/PropertyAnalyzer.java

@@ -171,6 +171,8 @@ public class PropertyAnalyzer {
 
     public static final String PROPERTIES_MUTABLE_BUCKET_NUM = "mutable_bucket_num";
 
+    public static final String PROPERTIES_ENABLE_LOAD_PROFILE = "enable_load_profile";
+
     public static final String PROPERTIES_PRIMARY_INDEX_CACHE_EXPIRE_SEC = "primary_index_cache_expire_sec";
 
     public static final String PROPERTIES_TABLET_TYPE = "tablet_type";
@@ -445,6 +447,14 @@ public static long analyzeMutableBucketNum(Map<String, String> properties) {
         }
     }
 
+    public static boolean analyzeEnableLoadProfile(Map<String, String> properties) {
+        boolean enableLoadProfile = false;
+        if (properties != null && properties.containsKey(PROPERTIES_ENABLE_LOAD_PROFILE)) {
+            enableLoadProfile = Boolean.parseBoolean(properties.get(PROPERTIES_ENABLE_LOAD_PROFILE));
+        }
+        return enableLoadProfile;
+    }
+
     public static int analyzeAutoRefreshPartitionsLimit(Map<String, String> properties, MaterializedView mv) {
         if (mv.getRefreshScheme().getType() == MaterializedView.RefreshType.MANUAL) {
             throw new SemanticException(

fe/fe-core/src/main/java/com/starrocks/journal/JournalEntity.java

@@ -841,6 +841,7 @@ public void readFields(DataInput in) throws IOException {
             case OperationType.OP_MODIFY_REPLICATED_STORAGE:
             case OperationType.OP_MODIFY_BUCKET_SIZE:
             case OperationType.OP_MODIFY_MUTABLE_BUCKET_NUM:
+            case OperationType.OP_MODIFY_ENABLE_LOAD_PROFILE:
             case OperationType.OP_MODIFY_BINLOG_CONFIG:
             case OperationType.OP_MODIFY_BINLOG_AVAILABLE_VERSION:
             case OperationType.OP_MODIFY_ENABLE_PERSISTENT_INDEX:

fe/fe-core/src/main/java/com/starrocks/load/streamload/StreamLoadTask.java

@@ -46,10 +46,8 @@
 import com.starrocks.persist.gson.GsonPostProcessable;
 import com.starrocks.persist.gson.GsonPreProcessable;
 import com.starrocks.persist.gson.GsonUtils;
-import com.starrocks.qe.ConnectContext;
 import com.starrocks.qe.DefaultCoordinator;
 import com.starrocks.qe.QeProcessorImpl;
-import com.starrocks.qe.SessionVariable;
 import com.starrocks.qe.scheduler.Coordinator;
 import com.starrocks.rpc.ThriftConnectionPool;
 import com.starrocks.rpc.ThriftRPCRequestExecutor;
@@ -840,7 +838,7 @@ private void unprotectedWaitCoordFinish() throws UserException {
                 }
 
                 if (coord.isEnableLoadProfile()) {
-                    collectProfile();
+                    collectProfile(false);
                 }
 
                 this.trackingUrl = coord.getTrackingUrl();
@@ -1019,7 +1017,7 @@ public void afterCommitted(TransactionState txnState, boolean txnOperated) throw
 
         // sync stream load collect profile, here we collect profile only when be has reported
         if (isSyncStreamLoad() && coord != null && coord.isProfileAlreadyReported()) {
-            collectProfile();
+            collectProfile(false);
         }
 
         writeLock();
@@ -1037,31 +1035,24 @@ public void afterCommitted(TransactionState txnState, boolean txnOperated) throw
         }
     }
 
-    public void collectProfile() {
-        long currentTimestamp = System.currentTimeMillis();
-        long totalTimeMs = currentTimestamp - createTimeMs;
-
-        // For the usage scenarios of flink cdc or routine load,
-        // the frequency of stream load maybe very high, resulting in many profiles,
-        // but we may only care about the long-duration stream load profile.
-        if (totalTimeMs < Config.stream_load_profile_collect_second * 1000) {
-            LOG.info(String.format("Load %s, totalTimeMs %d < Config.stream_load_profile_collect_second %d)",
-                    label, totalTimeMs, Config.stream_load_profile_collect_second));
-            return;
-        }
-
+    public RuntimeProfile buildTopLevelProfile(boolean isAborted) {
         RuntimeProfile profile = new RuntimeProfile("Load");
         RuntimeProfile summaryProfile = new RuntimeProfile("Summary");
         summaryProfile.addInfoString(ProfileManager.QUERY_ID, DebugUtil.printId(loadId));
         summaryProfile.addInfoString(ProfileManager.START_TIME,
                 TimeUtils.longToTimeString(createTimeMs));
 
-        summaryProfile.addInfoString(ProfileManager.END_TIME, TimeUtils.longToTimeString(System.currentTimeMillis()));
+        long currentTimestamp = System.currentTimeMillis();
+        long totalTimeMs = currentTimestamp - createTimeMs;
+        summaryProfile.addInfoString(ProfileManager.END_TIME, TimeUtils.longToTimeString(currentTimestamp));
         summaryProfile.addInfoString(ProfileManager.TOTAL_TIME, DebugUtil.getPrettyStringMs(totalTimeMs));
 
         summaryProfile.addInfoString(ProfileManager.QUERY_TYPE, "Load");
+        summaryProfile.addInfoString(ProfileManager.LOAD_TYPE, getStringByType());
+        summaryProfile.addInfoString(ProfileManager.QUERY_STATE, isAborted ? "Aborted" : "Finished");
         summaryProfile.addInfoString("StarRocks Version",
                 String.format("%s-%s", Version.STARROCKS_VERSION, Version.STARROCKS_COMMIT_HASH));
+        summaryProfile.addInfoString(ProfileManager.SQL_STATEMENT, getStmt());
         summaryProfile.addInfoString(ProfileManager.DEFAULT_DB, dbName);
 
         Map<String, String> loadCounters = coord.getLoadCounters();
@@ -1071,19 +1062,20 @@ public void collectProfile() {
             summaryProfile.addInfoString("NumRowsAbnormal", loadCounters.get(LoadEtlTask.DPP_ABNORMAL_ALL));
             summaryProfile.addInfoString("numRowsUnselected", loadCounters.get(LoadJob.UNSELECTED_ROWS));
         }
-        ConnectContext session = ConnectContext.get();
-        if (session != null) {
-            SessionVariable variables = session.getSessionVariable();
-            if (variables != null) {
-                summaryProfile.addInfoString("NonDefaultSessionVariables", variables.getNonDefaultVariablesJson());
-            }
-        }
 
         profile.addChild(summaryProfile);
+
+        return profile;
+    }
+
+
+    public void collectProfile(boolean isAborted) {
+        RuntimeProfile profile = buildTopLevelProfile(isAborted);
+
         if (coord.getQueryProfile() != null) {
             if (!isSyncStreamLoad()) {
                 coord.collectProfileSync();
-                profile.addChild(coord.buildQueryProfile(session == null || session.needMergeProfile()));
+                profile.addChild(coord.buildQueryProfile(true));
             } else {
                 profile.addChild(coord.getQueryProfile());
             }
@@ -1140,6 +1132,10 @@ public void afterAborted(TransactionState txnState, boolean txnOperated, String
             return;
         }
 
+        if (isSyncStreamLoad() && coord.isProfileAlreadyReported()) {
+            collectProfile(true);
+        }
+
         writeLock();
         try {
             if (isFinalState()) {
@@ -1351,6 +1347,10 @@ public boolean isRoutineLoadTask() {
         return type == Type.ROUTINE_LOAD;
     }
 
+    public String getStmt() {
+        return "";
+    }
+
     // for sync stream load
     public void setCoordinator(Coordinator coord) {
         this.coord = coord;
@@ -1359,9 +1359,9 @@ public void setCoordinator(Coordinator coord) {
     public String getStringByType() {
         switch (this.type) {
             case ROUTINE_LOAD:
-                return "ROUTINE_LOAD";
+                return ProfileManager.LOAD_TYPE_ROUTINE_LOAD;
             case STREAM_LOAD:
-                return "STREAM_LOAD";
+                return ProfileManager.LOAD_TYPE_STREAM_LOAD;
             case PARALLEL_STREAM_LOAD:
                 return "PARALLEL_STREAM_LOAD";
             default:

fe/fe-core/src/main/java/com/starrocks/persist/EditLog.java

@@ -823,6 +823,7 @@ public void loadJournal(GlobalStateMgr globalStateMgr, JournalEntity journal)
                 case OperationType.OP_MODIFY_REPLICATED_STORAGE:
                 case OperationType.OP_MODIFY_BUCKET_SIZE:
                 case OperationType.OP_MODIFY_MUTABLE_BUCKET_NUM:
+                case OperationType.OP_MODIFY_ENABLE_LOAD_PROFILE:
                 case OperationType.OP_MODIFY_BINLOG_AVAILABLE_VERSION:
                 case OperationType.OP_MODIFY_BINLOG_CONFIG:
                 case OperationType.OP_MODIFY_ENABLE_PERSISTENT_INDEX:
@@ -1729,6 +1730,10 @@ public void logModifyMutableBucketNum(ModifyTablePropertyOperationLog info) {
         logEdit(OperationType.OP_MODIFY_MUTABLE_BUCKET_NUM, info);
     }
 
+    public void logModifyEnableLoadProfile(ModifyTablePropertyOperationLog info) {
+        logEdit(OperationType.OP_MODIFY_ENABLE_LOAD_PROFILE, info);
+    }
+
     public void logReplaceTempPartition(ReplacePartitionOperationLog info) {
         logEdit(OperationType.OP_REPLACE_TEMP_PARTITION, info);
     }

fe/fe-core/src/main/java/com/starrocks/persist/OperationType.java

@@ -691,6 +691,9 @@ public class OperationType {
     @IgnorableOnReplayFailed
     public static final short OP_MODIFY_MUTABLE_BUCKET_NUM = 11141;
 
+    @IgnorableOnReplayFailed
+    public static final short OP_MODIFY_ENABLE_LOAD_PROFILE = 11142;
+
     // external table analyze
 
     @IgnorableOnReplayFailed