SERVER-43942 track newly created indexes until they are part of a consistent WT checkpoint, and skip them in background validation

3 files changed, 40 insertions, 0 deletions

diff --git a/src/mongo/db/catalog/catalog_control_test.cpp b/src/mongo/db/catalog/catalog_control_test.cpp
index 23fc51218d2..7a8a530731d 100644
--- a/src/mongo/db/catalog/catalog_control_test.cpp
+++ b/src/mongo/db/catalog/catalog_control_test.cpp
@@ -186,6 +186,11 @@ public:
     std::unique_ptr<CheckpointLock> getCheckpointLock(OperationContext* opCtx) final {
         return nullptr;
     }
+    void addIndividuallyCheckpointedIndexToList(const std::string& ident) final {}
+    void clearIndividuallyCheckpointedIndexesList() final {}
+    bool isInIndividuallyCheckpointedIndexesList(const std::string& ident) const final {
+        return false;
+    }
 };
 
 /**
diff --git a/src/mongo/db/catalog/multi_index_block.cpp b/src/mongo/db/catalog/multi_index_block.cpp
index f206610698b..20a97d48938 100644
--- a/src/mongo/db/catalog/multi_index_block.cpp
+++ b/src/mongo/db/catalog/multi_index_block.cpp
@@ -50,6 +50,7 @@
 #include "mongo/db/query/collection_query_info.h"
 #include "mongo/db/repl/repl_set_config.h"
 #include "mongo/db/repl/replication_coordinator.h"
+#include "mongo/db/storage/durable_catalog.h"
 #include "mongo/db/storage/storage_options.h"
 #include "mongo/db/storage/write_unit_of_work.h"
 #include "mongo/logger/redaction.h"
@@ -819,6 +820,29 @@ Status MultiIndexBlock::commit(OperationContext* opCtx,
                 _indexes[i].block->getEntry()->setMultikey(opCtx, *multikeyPaths);
             }
         }
+
+        if (opCtx->getServiceContext()->getStorageEngine()->supportsCheckpoints()) {
+            // Add the new index ident to a list so that the validate cmd with {background:true}
+            // can ignore the new index until it is regularly checkpoint'ed with the rest of the
+            // storage data.
+            //
+            // Index builds use the bulk loader, which can provoke a checkpoint of just that index.
+            // This makes the checkpoint's PIT view of the collection and indexes inconsistent until
+            // the next storage-wide checkpoint is taken, at which point the list will be reset.
+            //
+            // Note that it is okay if the index commit fails: background validation will never try
+            // to look at the index and the list will be reset by the next periodic storage-wide
+            // checkpoint.
+            //
+            // TODO (SERVER-44012): to remove this workaround.
+            auto checkpointLock =
+                opCtx->getServiceContext()->getStorageEngine()->getCheckpointLock(opCtx);
+            auto indexIdent =
+                opCtx->getServiceContext()->getStorageEngine()->getCatalog()->getIndexIdent(
+                    opCtx, collection->ns(), _indexes[i].block->getIndexName());
+            opCtx->getServiceContext()->getStorageEngine()->addIndividuallyCheckpointedIndexToList(
+                indexIdent);
+        }
     }
 
     onCommit();
diff --git a/src/mongo/db/catalog/validate_state.cpp b/src/mongo/db/catalog/validate_state.cpp
index 82123f96eaf..bf9aa1f6de7 100644
--- a/src/mongo/db/catalog/validate_state.cpp
+++ b/src/mongo/db/catalog/validate_state.cpp
@@ -234,6 +234,17 @@ void ValidateState::initializeCursors(OperationContext* opCtx) {
             continue;
         }
 
+        // Skip any newly created indexes that, because they were built with a WT bulk loader, are
+        // checkpoint'ed but not yet consistent with the rest of checkpoint's PIT view of the data.
+        if (_background &&
+            opCtx->getServiceContext()->getStorageEngine()->isInIndividuallyCheckpointedIndexesList(
+                diskIndexIdent)) {
+            _indexCursors.erase(desc->indexName());
+            log() << "Skipping validation on index '" << desc->indexName() << "' in collection '"
+                  << _nss << "' because the index data is not yet consistent in the checkpoint.";
+            continue;
+        }
+
         _indexes.push_back(indexCatalog->getEntryShared(desc));
     }