[Enhancement] limit max predicate columns when using full analyze pre…

…dicate column instead of sample all columns on auto collect statistics (StarRocks#56162) Signed-off-by: stephen <[email protected]>
silverbullet233 · Feb 24, 2025 · a19f5c1 · a19f5c1
1 parent 0a75f1e
commit a19f5c1
Show file tree

Hide file tree

Showing 3 changed files with 23 additions and 5 deletions.
diff --git a/fe/fe-core/src/main/java/com/starrocks/common/Config.java b/fe/fe-core/src/main/java/com/starrocks/common/Config.java
@@ -2145,6 +2145,9 @@ public class Config extends ConfigBase {
     @ConfField(mutable = true, comment = "If full analyze predicate columns instead of sample all columns")
     public static boolean statistic_auto_collect_use_full_predicate_column_for_sample = true;
 
+    @ConfField(mutable = true, comment = "max columns size of full analyze predicate columns instead of sample all columns")
+    public static int statistic_auto_collect_max_predicate_column_size_on_sample_strategy = 16;
+
     /**
      * Max row count in statistics collect per query
      */

diff --git a/fe/fe-core/src/main/java/com/starrocks/statistic/StatisticsCollectJobFactory.java b/fe/fe-core/src/main/java/com/starrocks/statistic/StatisticsCollectJobFactory.java
@@ -464,7 +464,7 @@ private static void createJob(List<StatisticsCollectJob> allTableJobMap, NativeA
         // Use predicate columns if suitable
         TableName tableName = new TableName(db.getOriginName(), table.getName());
         int numColumns = table.getColumns().size();
-        List<String> predicateColumnNames = null;
+        List<String> predicateColNames = null;
         boolean existsPredicateColumns = false;
         boolean enablePredicateColumnStrategy = false;
         if (basicStatsMeta != null && !basicStatsMeta.isInitJobMeta() && useBasicStats &&
@@ -473,7 +473,7 @@ private static void createJob(List<StatisticsCollectJob> allTableJobMap, NativeA
             List<ColumnUsage> predicateColumns = PredicateColumnsMgr.getInstance().queryPredicateColumns(tableName);
             if (CollectionUtils.isNotEmpty(predicateColumns) && predicateColumns.size() < numColumns) {
                 OlapTable olap = (OlapTable) table;
-                predicateColumnNames = predicateColumns.stream().map(x -> x.getOlapColumnName(olap)).toList();
+                predicateColNames = predicateColumns.stream().map(x -> x.getOlapColumnName(olap)).toList();
                 existsPredicateColumns = true;
                 if (numColumns > Config.statistic_auto_collect_predicate_columns_threshold) {
                     enablePredicateColumnStrategy = true;
@@ -533,7 +533,8 @@ private static void createJob(List<StatisticsCollectJob> allTableJobMap, NativeA
             if (job.getAnalyzeType() != StatsConstants.AnalyzeType.HISTOGRAM &&
                     healthy < Config.statistic_auto_collect_sample_threshold &&
                     sumDataSize > Config.statistic_auto_collect_small_table_size) {
-                if (!(Config.statistic_auto_collect_use_full_predicate_column_for_sample && existsPredicateColumns)) {
+                if (!(Config.statistic_auto_collect_use_full_predicate_column_for_sample && existsPredicateColumns &&
+                        predicateColNames.size() < Config.statistic_auto_collect_max_predicate_column_size_on_sample_strategy)) {
                     LOG.debug("statistics job choose sample on real-time update table: {}" +
                                     ", last collect time: {}, current healthy: {}, full collect healthy limit: {}, " +
                                     ", update data size: {}MB, full collect healthy data size limit: <{}MB",
@@ -550,8 +551,8 @@ private static void createJob(List<StatisticsCollectJob> allTableJobMap, NativeA
             }
         }
 
-        if (enablePredicateColumnStrategy && CollectionUtils.isNotEmpty(predicateColumnNames)) {
-            columnNames = predicateColumnNames;
+        if (enablePredicateColumnStrategy && CollectionUtils.isNotEmpty(predicateColNames)) {
+            columnNames = predicateColNames;
             columnTypes = columnNames.stream().map(col -> table.getColumn(col).getType()).collect(Collectors.toList());
         }
 

diff --git a/fe/fe-core/src/test/java/com/starrocks/statistic/columns/ColumnUsageTest.java b/fe/fe-core/src/test/java/com/starrocks/statistic/columns/ColumnUsageTest.java
@@ -242,5 +242,19 @@ public void testAutoAnalyzePredicateColumns() throws Exception {
             Assertions.assertEquals(List.of("v1", "v2", "v3"), job0.getColumnNames());
             Config.statistic_auto_collect_predicate_columns_threshold = defaultValue;
         }
+
+        {
+            long defaultSmallTableSize = Config.statistic_auto_collect_small_table_size;
+            Config.statistic_auto_collect_small_table_size = -1;
+            int defaultPredicateColumnSize = Config.statistic_auto_collect_max_predicate_column_size_on_sample_strategy;
+            Config.statistic_auto_collect_max_predicate_column_size_on_sample_strategy = -1;
+            List<StatisticsCollectJob> collectJobs = StatisticsCollectJobFactory.buildStatisticsCollectJob(analyzeJob);
+            Assertions.assertEquals(1, collectJobs.size());
+            StatisticsCollectJob job0 = collectJobs.get(0);
+            Assertions.assertEquals(StatsConstants.AnalyzeType.SAMPLE, job0.getType());
+            Assertions.assertEquals(List.of("v1", "v2", "v3"), job0.getColumnNames());
+            Config.statistic_auto_collect_small_table_size = defaultSmallTableSize;
+            Config.statistic_auto_collect_max_predicate_column_size_on_sample_strategy = defaultPredicateColumnSize;
+        }
     }
 }