[Enhancement] add metrics for pk table error state (backport #52590)

be/src/storage/lake/meta_file.cpp

@@ -26,6 +26,7 @@
 #include "util/coding.h"
 #include "util/defer_op.h"
 #include "util/raw_container.h"
+#include "util/starrocks_metrics.h"
 #include "util/trace.h"
 
 namespace starrocks::lake {
@@ -407,6 +408,7 @@ Status MetaFileBuilder::update_num_del_stat(const std::map<uint32_t, size_t>& se
             std::string err_msg =
                     fmt::format("unexpected segment id: {} tablet id: {}", each.first, _tablet_meta->id());
             LOG(ERROR) << err_msg;
+            StarRocksMetrics::instance()->primary_key_table_error_state_total.increment(1);
             if (!config::experimental_lake_ignore_pk_consistency_check) {
                 set_recover_flag(RecoverFlag::RECOVER_WITHOUT_PUBLISH);
                 return Status::InternalError(err_msg);

be/src/storage/lake/update_manager.cpp

@@ -145,6 +145,7 @@ StatusOr<IndexEntry*> UpdateManager::prepare_primary_index(
     _index_cache.update_object_size(index_entry, index.memory_usage());
     if (!st.ok()) {
         if (st.is_already_exist()) {
+            StarRocksMetrics::instance()->primary_key_table_error_state_total.increment(1);
             builder->set_recover_flag(RecoverFlag::RECOVER_WITH_PUBLISH);
         }
         _index_cache.remove(index_entry);
@@ -305,6 +306,7 @@ Status UpdateManager::publish_primary_key_tablet(const TxnLogPB_OpWrite& op_writ
                         "v:$6",
                         tablet->id(), rssid, cur_old, cur_add, cur_new, old_del_vec->version(), metadata->version());
                 LOG(ERROR) << error_msg;
+                StarRocksMetrics::instance()->primary_key_table_error_state_total.increment(1);
                 if (!config::experimental_lake_ignore_pk_consistency_check) {
                     builder->set_recover_flag(RecoverFlag::RECOVER_WITH_PUBLISH);
                     return Status::InternalError(error_msg);

be/src/storage/tablet_updates.cpp

@@ -3662,6 +3662,7 @@ void TabletUpdates::_print_rowsets(std::vector<uint32_t>& rowsets, std::string*
 }
 
 void TabletUpdates::_set_error(const string& msg) {
+    StarRocksMetrics::instance()->primary_key_table_error_state_total.increment(1);
     _error_msg = msg;
     _error = true;
     _apply_version_changed.notify_all();

be/src/util/starrocks_metrics.cpp

@@ -96,6 +96,7 @@ StarRocksMetrics::StarRocksMetrics() : _metrics(_s_registry_name) {
     REGISTER_STARROCKS_METRIC(delta_column_group_get_hit_cache);
     REGISTER_STARROCKS_METRIC(delta_column_group_get_non_pk_total);
     REGISTER_STARROCKS_METRIC(delta_column_group_get_non_pk_hit_cache);
+    REGISTER_STARROCKS_METRIC(primary_key_table_error_state_total);
 
     // push request
     _metrics.register_metric("push_requests_total", MetricLabels().add("status", "SUCCESS"),

be/src/util/starrocks_metrics.h

@@ -264,6 +264,7 @@ class StarRocksMetrics {
     METRIC_DEFINE_INT_COUNTER(delta_column_group_get_hit_cache, MetricUnit::REQUESTS);
     METRIC_DEFINE_INT_COUNTER(delta_column_group_get_non_pk_total, MetricUnit::REQUESTS);
     METRIC_DEFINE_INT_COUNTER(delta_column_group_get_non_pk_hit_cache, MetricUnit::REQUESTS);
+    METRIC_DEFINE_INT_COUNTER(primary_key_table_error_state_total, MetricUnit::REQUESTS);
 
     // Gauges
     METRIC_DEFINE_INT_GAUGE(memory_pool_bytes_total, MetricUnit::BYTES);

be/test/storage/lake/meta_file_test.cpp

@@ -32,6 +32,7 @@
 #include "storage/lake/update_manager.h"
 #include "testutil/assert.h"
 #include "testutil/id_generator.h"
+#include "util/starrocks_metrics.h"
 #include "util/uid_util.h"
 
 namespace starrocks::lake {
@@ -572,4 +573,30 @@ TEST_F(MetaFileTest, test_trim_partial_compaction_last_input_rowset) {
     EXPECT_EQ(last_input_rowset_metadata.segments_size(), 2);
 }
 
+TEST_F(MetaFileTest, test_error_state) {
+    // generate metadata
+    const int64_t tablet_id = 10001;
+    auto tablet = std::make_shared<Tablet>(_tablet_manager.get(), tablet_id);
+    auto metadata = std::make_shared<TabletMetadata>();
+    metadata->set_id(tablet_id);
+    metadata->set_version(10);
+    metadata->set_next_rowset_id(110);
+
+    // add rowset with segment
+    RowsetMetadataPB rowset_metadata;
+    rowset_metadata.set_id(110);
+    rowset_metadata.add_segments("aaa.dat");
+    rowset_metadata.add_segments("bbb.dat");
+    metadata->add_rowsets()->CopyFrom(rowset_metadata);
+    std::map<uint32_t, size_t> segment_id_to_add_dels;
+    for (int i = 0; i < 10; i++) {
+        segment_id_to_add_dels[i] = 100;
+    }
+    // generate error state
+    MetaFileBuilder builder(*tablet, metadata);
+    Status st = builder.update_num_del_stat(segment_id_to_add_dels);
+    EXPECT_FALSE(st.ok());
+    EXPECT_TRUE(StarRocksMetrics::instance()->primary_key_table_error_state_total.value() > 0);
+}
+
 } // namespace starrocks::lake