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

Signed-off-by: luohaha <[email protected]> (cherry picked from commit 7c45354) # Conflicts: # be/test/storage/lake/meta_file_test.cpp
StarRocks · Nov 5, 2024 · c3b695e · c3b695e
1 parent 91d710d
commit c3b695e
Show file tree

Hide file tree

Showing 6 changed files with 371 additions and 0 deletions.
diff --git a/be/src/storage/lake/meta_file.cpp b/be/src/storage/lake/meta_file.cpp
@@ -25,6 +25,7 @@
 #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 {
@@ -159,6 +160,7 @@ 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);

diff --git a/be/src/storage/lake/update_manager.cpp b/be/src/storage/lake/update_manager.cpp
@@ -84,6 +84,7 @@ StatusOr<IndexEntry*> UpdateManager::prepare_primary_index(const TabletMetadata&
     _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);
@@ -223,6 +224,7 @@ 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);

diff --git a/be/src/storage/tablet_updates.cpp b/be/src/storage/tablet_updates.cpp
@@ -3393,6 +3393,7 @@ 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();

diff --git a/be/src/util/starrocks_metrics.cpp b/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"),

diff --git a/be/src/util/starrocks_metrics.h b/be/src/util/starrocks_metrics.h
@@ -255,6 +255,7 @@ 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);