From ac5f9694136b029099daf8b79e720f2dcf88c7e6 Mon Sep 17 00:00:00 2001
From: yangqise7en <877793735@qq.com>
Date: Tue, 26 Mar 2024 07:16:15 +0000
Subject: [PATCH] [CP]clean tablet_ls_pair/ckm_sum_array when failed & add
 obtest

---
 deps/oblib/src/lib/utility/ob_tracepoint.h    |  1 +
 .../freeze/ob_checksum_validator.cpp          | 17 +++++++++++++-
 src/share/compaction/ob_table_ckm_items.cpp   | 22 ++++++++++++++++---
 3 files changed, 36 insertions(+), 4 deletions(-)

diff --git a/deps/oblib/src/lib/utility/ob_tracepoint.h b/deps/oblib/src/lib/utility/ob_tracepoint.h
index a482a510ec..3ec9654f4a 100644
--- a/deps/oblib/src/lib/utility/ob_tracepoint.h
+++ b/deps/oblib/src/lib/utility/ob_tracepoint.h
@@ -711,6 +711,7 @@ class EventTable
       EN_CO_MREGE_DAG_SCHEDULE_REST = 739,
       EN_COMPACTION_SCHEDULE_MEDIUM_MERGE_AFTER_MINI = 740,
       EN_COMPACTION_MEDIUM_INIT_LARGE_PARALLEL_RANGE = 741,
+      EN_GET_TABLET_LS_PAIR_IN_RS = 742,
 
       // please add new trace point after 750
       EN_SESSION_LEAK_COUNT_THRESHOLD = 751,
diff --git a/src/rootserver/freeze/ob_checksum_validator.cpp b/src/rootserver/freeze/ob_checksum_validator.cpp
index 05a8be8a38..52b7a80fcc 100755
--- a/src/rootserver/freeze/ob_checksum_validator.cpp
+++ b/src/rootserver/freeze/ob_checksum_validator.cpp
@@ -219,6 +219,20 @@ int ObChecksumValidator::get_tablet_ls_pairs(
     } else if (OB_FAIL(tablet_ls_pair_cache_.get_tablet_ls_pairs(table_id_, tablet_ids, cur_tablet_ls_pair_array_))) {
       LOG_WARN("failed to tablet ls pair", KR(ret), K(tablet_ids));
     } else {
+#ifdef ERRSIM
+        static int64_t enter_cnt = 0;
+        if (OB_SUCC(ret) && simple_schema.is_global_index_table()) {
+          ret = OB_E(EventTable::EN_GET_TABLET_LS_PAIR_IN_RS) OB_SUCCESS;
+          if (OB_FAIL(ret)) {
+            if (enter_cnt++ == 0) {
+              ret = OB_ITEM_NOT_MATCH;
+              STORAGE_LOG(INFO, "ERRSIM EN_GET_TABLET_LS_PAIR_IN_RS", K(ret), K(simple_schema), K_(cur_tablet_ls_pair_array));
+            } else {
+              ret = OB_SUCCESS;
+            }
+          }
+        }
+#endif
       LOG_TRACE("success to get tablet ls pairs", KR(ret), K_(cur_tablet_ls_pair_array));
     }
   }
@@ -286,8 +300,9 @@ int ObChecksumValidator::validate_checksum(
     } else {
       last_table_ckm_items_.clear();
     }
-    cur_tablet_ls_pair_array_.reuse();
   }
+  cur_tablet_ls_pair_array_.reuse(); // need reuse array when get_tablet_ls_pairs failed
+
   if (FAILEDx(table_compaction_map_.set_refactored(table_id_, table_compaction_info_, true /*overwrite*/))) {
     LOG_WARN("fail to set refactored", KR(ret), K_(table_id), K_(table_compaction_info));
   } else {
diff --git a/src/share/compaction/ob_table_ckm_items.cpp b/src/share/compaction/ob_table_ckm_items.cpp
index 5ea3afbac7..a40363ed29 100644
--- a/src/share/compaction/ob_table_ckm_items.cpp
+++ b/src/share/compaction/ob_table_ckm_items.cpp
@@ -242,14 +242,18 @@ int ObTableCkmItems::build_column_ckm_sum_array(
     ret = OB_ERR_UNEXPECTED;
     LOG_WARN("checksum items or tablet pairs are empty", KR(ret), K_(ckm_items), K_(tablet_pairs));
   } else if (!ckm_sum_array_.empty()) {
-    // inited before
+    LOG_INFO("use cached ckm array", KR(ret), K_(row_count), K_(ckm_sum_array), K(compaction_scn));
   } else {
+    row_count_ = 0;
+    ckm_sum_array_.reuse();
+
     const int64_t column_checksums_cnt = ckm_items_.at(0).column_meta_.column_checksums_.count();
     uint64_t pre_tablet_id = OB_INVALID_ID;
     if (OB_FAIL(ckm_sum_array_.reserve(column_checksums_cnt))) {
       LOG_WARN("failed to reserve tablet column checksum array", KR(ret));
     }
     // items are order by tablet_id
+    int64_t pair_idx = 0;
     for (int64_t i = 0; OB_SUCC(ret) && (i < items_cnt); ++i) {
       const ObTabletReplicaChecksumItem &cur_item = ckm_items_.at(i);
       LOG_TRACE("build_column_ckm_sum_array", KR(ret), K(i), K(cur_item), K(compaction_scn), K(row_cnt));
@@ -271,7 +275,15 @@ int ObTableCkmItems::build_column_ckm_sum_array(
           } // end of for
           row_count_ += cur_item.row_count_;
         }
-        pre_tablet_id = cur_item.tablet_id_.id();
+
+        if (OB_FAIL(ret)) {
+        } else if (OB_UNLIKELY(cur_item.tablet_id_.id() != pre_tablet_id // meet new tablet
+            && (pair_idx >= tablet_pairs_.count() || tablet_pairs_.at(pair_idx++).get_tablet_id() != cur_item.tablet_id_))) {
+          ret = OB_ERR_UNEXPECTED;
+          LOG_WARN("tablet pair and ckm items are mismatch", KR(ret), K(i), K(cur_item), K(tablet_pairs_), K(pair_idx));
+        } else {
+          pre_tablet_id = cur_item.tablet_id_.id();
+        }
       } else {
         ret = OB_ITEM_NOT_MATCH;
         LOG_WARN("compaction scn mismtach", KR(ret), K(cur_item), K(compaction_scn));
@@ -280,6 +292,9 @@ int ObTableCkmItems::build_column_ckm_sum_array(
   }
   if (OB_SUCC(ret)) {
     row_cnt = row_count_;
+  } else {
+    row_count_ = 0;
+    ckm_sum_array_.reuse();
   }
   return ret;
 }
@@ -523,11 +538,12 @@ void ObTableCkmItems::reset()
 {
   is_inited_ = false;
   table_id_ = 0;
+  row_count_ = 0;
+  table_schema_ = NULL;
   tablet_pairs_.reset();
   ckm_items_.reset();
   sort_col_id_array_.reset();
   ckm_sum_array_.reset();
-  row_count_ = 0;
 }
 
 } // namespace compaction