SERVER-72677 Surface index validation errors arising from WT::verify

(cherry picked from commit 40c93f028e36f78c06756f4bfd358d240bdd9b34)

6 files changed, 57 insertions, 21 deletions

diff --git a/jstests/noPassthrough/background_validation_checkpoint_existence.js b/jstests/noPassthrough/background_validation_checkpoint_existence.js
index 8b5a1cd8ea0..438278b9aea 100644
--- a/jstests/noPassthrough/background_validation_checkpoint_existence.js
+++ b/jstests/noPassthrough/background_validation_checkpoint_existence.js
@@ -32,30 +32,30 @@ for (let i = 0; i < 5; i++) {
 
 // The collection has not been checkpointed yet, so there is nothing to validate.
 let res = assert.commandWorked(db.runCommand({validate: collName, background: true}));
-assert.eq(true, res.valid);
-assert.eq(false, res.hasOwnProperty("nrecords"));
-assert.eq(false, res.hasOwnProperty("nIndexes"));
+assert.eq(true, res.valid, res);
+assert.eq(false, res.hasOwnProperty("nrecords"), res);
+assert.eq(false, res.hasOwnProperty("nIndexes"), res);
 
 forceCheckpoint();
 
 res = assert.commandWorked(db.runCommand({validate: collName, background: true}));
-assert.eq(true, res.valid);
-assert.eq(true, res.hasOwnProperty("nrecords"));
-assert.eq(true, res.hasOwnProperty("nIndexes"));
+assert.eq(true, res.valid, res);
+assert.eq(true, res.hasOwnProperty("nrecords"), res);
+assert.eq(true, res.hasOwnProperty("nIndexes"), res);
 
 assert.commandWorked(coll.createIndex({x: 1}));
 
 // Shouldn't validate the newly created index here as it wasn't checkpointed yet.
 res = assert.commandWorked(db.runCommand({validate: collName, background: true}));
-assert.eq(true, res.valid);
-assert.eq(1, res.nIndexes);
+assert.eq(true, res.valid, res);
+assert.eq(1, res.nIndexes, res);
 
 forceCheckpoint();
 
 // Validating after the checkpoint should validate the newly created index.
 res = assert.commandWorked(db.runCommand({validate: collName, background: true}));
-assert.eq(true, res.valid);
-assert.eq(2, res.nIndexes);
+assert.eq(true, res.valid, res);
+assert.eq(2, res.nIndexes, res);
 
 MongoRunner.stopMongod(conn);
 }());
\ No newline at end of file
diff --git a/src/mongo/db/catalog/collection_validation.cpp b/src/mongo/db/catalog/collection_validation.cpp
index 2c6661a4f5c..6f97ed729a4 100644
--- a/src/mongo/db/catalog/collection_validation.cpp
+++ b/src/mongo/db/catalog/collection_validation.cpp
@@ -261,19 +261,20 @@ void _reportValidationResults(OperationContext* opCtx,
 
     // Report detailed index validation results gathered when using {full: true} for validated
     // indexes.
-    for (const auto& index : validateState->getIndexes()) {
-        const std::string indexName = index->descriptor()->indexName();
-        auto& indexResultsMap = results->indexResultsMap;
-        if (indexResultsMap.find(indexName) == indexResultsMap.end()) {
-            continue;
-        }
-
-        auto& vr = indexResultsMap.at(indexName);
-
+    int nIndexes = results->indexResultsMap.size();
+    for (const auto& [indexName, vr] : results->indexResultsMap) {
         if (!vr.valid) {
             results->valid = false;
         }
 
+        if (validateState->getSkippedIndexes().contains(indexName)) {
+            // Index internal state was checked and cleared, so it was reported in indexResultsMap,
+            // but we did not verify the index contents against the collection, so we should exclude
+            // it from this report.
+            --nIndexes;
+            continue;
+        }
+
         BSONObjBuilder bob(indexDetails.subobjStart(indexName));
         bob.appendBool("valid", vr.valid);
 
@@ -292,7 +293,7 @@ void _reportValidationResults(OperationContext* opCtx,
         results->errors.insert(results->errors.end(), vr.errors.begin(), vr.errors.end());
     }
 
-    output->append("nIndexes", static_cast<int>(validateState->getIndexes().size()));
+    output->append("nIndexes", nIndexes);
     output->append("keysPerIndex", keysPerIndex.done());
     output->append("indexDetails", indexDetails.done());
 }
diff --git a/src/mongo/db/catalog/collection_validation_test.cpp b/src/mongo/db/catalog/collection_validation_test.cpp
index 15f8f475aec..03f460af790 100644
--- a/src/mongo/db/catalog/collection_validation_test.cpp
+++ b/src/mongo/db/catalog/collection_validation_test.cpp
@@ -276,6 +276,19 @@ TEST_F(CollectionValidationTest, ValidateEnforceFastCount) {
                        {CollectionValidation::ValidateMode::kForegroundFullEnforceFastCount});
 }
 
+TEST_F(CollectionValidationDiskTest, ValidateIndexDetailResultsSurfaceVerifyErrors) {
+    FailPointEnableBlock fp{"WTValidateIndexStructuralDamage"};
+    auto opCtx = operationContext();
+    insertDataRange(opCtx, 0, 5);  // initialize collection
+    foregroundValidate(
+        opCtx,
+        /*valid*/ false,
+        /*numRecords*/ std::numeric_limits<int32_t>::min(),           // uninitialized
+        /*numInvalidDocuments*/ std::numeric_limits<int32_t>::min(),  // uninitialized
+        /*numErrors*/ 1,
+        {CollectionValidation::ValidateMode::kForegroundFull});
+}
+
 /**
  * Waits for a parallel running collection validation operation to start and then hang at a
  * failpoint.
diff --git a/src/mongo/db/catalog/validate_state.cpp b/src/mongo/db/catalog/validate_state.cpp
index 9f0fd873f59..5b64ada4226 100644
--- a/src/mongo/db/catalog/validate_state.cpp
+++ b/src/mongo/db/catalog/validate_state.cpp
@@ -272,6 +272,7 @@ void ValidateState::initializeCursors(OperationContext* opCtx) {
                   "checkpoint.",
                   "desc_indexName"_attr = desc->indexName(),
                   "nss"_attr = _nss);
+            _skippedIndexes.emplace(desc->indexName());
             continue;
         }
 
@@ -289,12 +290,15 @@ void ValidateState::initializeCursors(OperationContext* opCtx) {
                   "yet in a checkpoint.",
                   "desc_indexName"_attr = desc->indexName(),
                   "nss"_attr = _nss);
+            _skippedIndexes.emplace(desc->indexName());
             continue;
         }
 
         auto iam = entry->accessMethod()->asSortedData();
-        if (!iam)
+        if (!iam) {
+            _skippedIndexes.emplace(desc->indexName());
             continue;
+        }
 
         _indexCursors.emplace(
             desc->indexName(),
diff --git a/src/mongo/db/catalog/validate_state.h b/src/mongo/db/catalog/validate_state.h
index c508a6b5850..14f9008207d 100644
--- a/src/mongo/db/catalog/validate_state.h
+++ b/src/mongo/db/catalog/validate_state.h
@@ -145,6 +145,10 @@ public:
         return _indexes;
     }
 
+    const StringSet& getSkippedIndexes() const {
+        return _skippedIndexes;
+    }
+
     /**
      * Map of index names to index cursors.
      */
@@ -257,6 +261,10 @@ private:
     std::unique_ptr<SeekableRecordThrottleCursor> _traverseRecordStoreCursor;
     std::unique_ptr<SeekableRecordThrottleCursor> _seekRecordStoreCursor;
 
+    // Stores the set of indexes that will not be validated for some reason, e.g. they are not
+    // ready.
+    StringSet _skippedIndexes;
+
     RecordId _firstRecordId;
 
     DataThrottle _dataThrottle;
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_index_util.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_index_util.cpp
index 273b692c375..38fa4c4ccc8 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_index_util.cpp
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_index_util.cpp
@@ -37,6 +37,7 @@
 namespace mongo {
 
 MONGO_FAIL_POINT_DEFINE(WTCompactIndexEBUSY);
+MONGO_FAIL_POINT_DEFINE(WTValidateIndexStructuralDamage);
 
 bool WiredTigerIndexUtil::appendCustomStats(OperationContext* opCtx,
                                             BSONObjBuilder* output,
@@ -122,6 +123,15 @@ bool WiredTigerIndexUtil::validateStructure(OperationContext* opCtx,
                                             const std::string& uri,
                                             IndexValidateResults* fullResults) {
     if (fullResults && !WiredTigerRecoveryUnit::get(opCtx)->getSessionCache()->isEphemeral()) {
+        if (WTValidateIndexStructuralDamage.shouldFail()) {
+            std::string msg = str::stream() << "verify() returned an error. "
+                                            << "This indicates structural damage. "
+                                            << "Not examining individual index entries.";
+            fullResults->errors.push_back(msg);
+            fullResults->valid = false;
+            return false;
+        }
+
         int err = WiredTigerUtil::verifyTable(opCtx, uri, &(fullResults->errors));
         if (err == EBUSY) {
             std::string msg = str::stream()