New upstream version 6.0.20upstream

8 files changed, 50 insertions, 46 deletions

diff --git a/src/mongo/db/ops/SConscript b/src/mongo/db/ops/SConscript
index 3e89ae63e36..9ac3c43f9fd 100644
--- a/src/mongo/db/ops/SConscript
+++ b/src/mongo/db/ops/SConscript
@@ -15,9 +15,9 @@ env.Library(
         '$BUILD_DIR/mongo/db/catalog/collection_options',
         '$BUILD_DIR/mongo/db/catalog_raii',
         '$BUILD_DIR/mongo/db/concurrency/exception_util',
-        '$BUILD_DIR/mongo/db/curop',
         '$BUILD_DIR/mongo/db/curop_metrics',
         '$BUILD_DIR/mongo/db/dbhelpers',
+        '$BUILD_DIR/mongo/db/query/query_stats/query_stats',
         '$BUILD_DIR/mongo/db/record_id_helpers',
         '$BUILD_DIR/mongo/db/repl/oplog',
         '$BUILD_DIR/mongo/db/repl/repl_coordinator_interface',
diff --git a/src/mongo/db/ops/insert.cpp b/src/mongo/db/ops/insert.cpp
index 61d92fda6bf..8c949b333c7 100644
--- a/src/mongo/db/ops/insert.cpp
+++ b/src/mongo/db/ops/insert.cpp
@@ -87,6 +87,7 @@ Status validateDepth(const BSONObj& obj) {
 
 StatusWith<BSONObj> fixDocumentForInsert(OperationContext* opCtx,
                                          const BSONObj& doc,
+                                         bool bypassEmptyTsReplacement,
                                          bool* containsDotsAndDollarsField) {
     bool validationDisabled = DocumentValidationSettings::get(opCtx).isInternalValidationDisabled();
 
@@ -139,7 +140,8 @@ StatusWith<BSONObj> fixDocumentForInsert(OperationContext* opCtx,
             }
 
             if (!validationDisabled) {
-                if (e.type() == bsonTimestamp && e.timestampValue() == 0) {
+                if (!bypassEmptyTsReplacement && e.type() == bsonTimestamp &&
+                    e.timestampValue() == 0) {
                     // we replace Timestamp(0,0) at the top level with a correct value
                     // in the fast pass, we just mark that we want to swap
                     hasTimestampToFix = true;
@@ -187,7 +189,8 @@ StatusWith<BSONObj> fixDocumentForInsert(OperationContext* opCtx,
         BSONElement e = i.next();
         if (hadId && e.fieldNameStringData() == "_id") {
             // no-op
-        } else if (e.type() == bsonTimestamp && e.timestampValue() == 0) {
+        } else if (!bypassEmptyTsReplacement && e.type() == bsonTimestamp &&
+                   e.timestampValue() == 0) {
             auto nextTime = VectorClockMutable::get(opCtx)->tickClusterTime(1);
             b.append(e.fieldName(), nextTime.asTimestamp());
         } else {
diff --git a/src/mongo/db/ops/insert.h b/src/mongo/db/ops/insert.h
index faed6de5890..a523e499d02 100644
--- a/src/mongo/db/ops/insert.h
+++ b/src/mongo/db/ops/insert.h
@@ -47,6 +47,7 @@ class OperationContext;
  */
 StatusWith<BSONObj> fixDocumentForInsert(OperationContext* opCtx,
                                          const BSONObj& doc,
+                                         bool bypassEmptyTsReplacement = false,
                                          bool* containsDotsOrDollarsField = nullptr);
 
 /**
diff --git a/src/mongo/db/ops/parsed_update.cpp b/src/mongo/db/ops/parsed_update.cpp
index b9557e4b2fe..a2badffa2aa 100644
--- a/src/mongo/db/ops/parsed_update.cpp
+++ b/src/mongo/db/ops/parsed_update.cpp
@@ -188,6 +188,7 @@ void ParsedUpdate::parseUpdate() {
     _driver.setCollator(_expCtx->getCollator());
     _driver.setLogOp(true);
     _driver.setFromOplogApplication(_request->isFromOplogApplication());
+    _driver.setBypassEmptyTsReplacement(static_cast<bool>(_request->getBypassEmptyTsReplacement()));
     // Time-series operations will not result in any documents with dots or dollars fields.
     if (auto source = _request->source(); source == OperationSource::kTimeseriesInsert ||
         source == OperationSource::kTimeseriesUpdate) {
diff --git a/src/mongo/db/ops/update_request.h b/src/mongo/db/ops/update_request.h
index 3db331defa9..2a086d8a4ab 100644
--- a/src/mongo/db/ops/update_request.h
+++ b/src/mongo/db/ops/update_request.h
@@ -209,6 +209,14 @@ public:
         return _fromOplogApplication;
     }
 
+    void setBypassEmptyTsReplacement(OptionalBool bypassEmptyTsReplacement) {
+        _bypassEmptyTsReplacement = bypassEmptyTsReplacement;
+    }
+
+    OptionalBool getBypassEmptyTsReplacement() const {
+        return _bypassEmptyTsReplacement;
+    }
+
     void setExplain(boost::optional<ExplainOptions::Verbosity> verbosity) {
         _explain = verbosity;
     }
@@ -314,6 +322,8 @@ private:
     // The statement ids of this request.
     std::vector<StmtId> _stmtIds = {kUninitializedStmtId};
 
+    OptionalBool _bypassEmptyTsReplacement;
+
     // Flags controlling the update.
 
     // God bypasses _id checking and index generation. It is only used on behalf of system
diff --git a/src/mongo/db/ops/write_ops.cpp b/src/mongo/db/ops/write_ops.cpp
index de7df85e564..14cf48fca37 100644
--- a/src/mongo/db/ops/write_ops.cpp
+++ b/src/mongo/db/ops/write_ops.cpp
@@ -131,6 +131,12 @@ int getWriteCommandRequestBaseSize(const WriteCommandRequestBase& base) {
             encryptionInfo->toBSON().objsize() + kPerElementOverhead;
     }
 
+    if (auto bypassEmptyTsReplacement = base.getBypassEmptyTsReplacement();
+        bypassEmptyTsReplacement.has_value()) {
+        estSize += write_ops::WriteCommandRequestBase::kBypassEmptyTsReplacementFieldName.size() +
+            kBoolSize + kPerElementOverhead;
+    }
+
     return estSize;
 }
 
diff --git a/src/mongo/db/ops/write_ops.idl b/src/mongo/db/ops/write_ops.idl
index ca606f3cfad..e41af05e808 100644
--- a/src/mongo/db/ops/write_ops.idl
+++ b/src/mongo/db/ops/write_ops.idl
@@ -205,6 +205,12 @@ structs:
                 type: EncryptionInformation
                 optional: true
                 unstable: true
+            bypassEmptyTsReplacement:
+                description: "Only applicable for inserts and replacement updates. If set to true,
+                              any empty timestamps (Timestamp(0,0)) in 'documents' or 'u' will not
+                              be replaced by the current time and instead will be preserved as-is."
+                type: optionalBool
+                unstable: true
 
     UpdateOpEntry:
         description: "Parser for the entries in the 'updates' array of an update command."
@@ -539,3 +545,9 @@ commands:
                 type: EncryptionInformation
                 optional: true
                 unstable: true
+            bypassEmptyTsReplacement:
+                description: "Only applicable when 'update' is a replacement update. If set, any
+                              empty timestamps (Timestamp(0, 0)) in the update will not be replaced
+                              by the current time and instead will be preserved as-is."
+                type: optionalBool
+                unstable: true
diff --git a/src/mongo/db/ops/write_ops_exec.cpp b/src/mongo/db/ops/write_ops_exec.cpp
index 05e78ecfca6..e58146ac5cd 100644
--- a/src/mongo/db/ops/write_ops_exec.cpp
+++ b/src/mongo/db/ops/write_ops_exec.cpp
@@ -96,46 +96,6 @@
 #include "mongo/util/scopeguard.h"
 
 namespace mongo::write_ops_exec {
-class Atomic64Metric;
-}  // namespace mongo::write_ops_exec
-
-namespace mongo {
-template <>
-struct BSONObjAppendFormat<write_ops_exec::Atomic64Metric> : FormatKind<NumberLong> {};
-}  // namespace mongo
-
-
-namespace mongo::write_ops_exec {
-
-/**
- * Atomic wrapper for long long type for Metrics.
- */
-class Atomic64Metric {
-public:
-    /** Set _value to the max of the current or newMax. */
-    void setIfMax(long long newMax) {
-        /*  Note: compareAndSwap will load into val most recent value. */
-        for (long long val = _value.load(); val < newMax && !_value.compareAndSwap(&val, newMax);) {
-        }
-    }
-
-    /** store val into value. */
-    void set(long long val) {
-        _value.store(val);
-    }
-
-    /** Return the current value. */
-    long long get() const {
-        return _value.load();
-    }
-
-    operator long long() const {
-        return get();
-    }
-
-private:
-    mongo::AtomicWord<long long> _value;
-};
 
 // Convention in this file: generic helpers go in the anonymous namespace. Helpers that are for a
 // single type of operation are static functions defined above their caller.
@@ -214,7 +174,7 @@ void finishCurOp(OperationContext* opCtx, CurOp* curOp) {
     try {
         curOp->done();
         auto executionTimeMicros = duration_cast<Microseconds>(curOp->elapsedTimeExcludingPauses());
-        curOp->debug().executionTime = executionTimeMicros;
+        curOp->debug().additiveMetrics.executionTime = executionTimeMicros;
 
         recordCurOpMetrics(opCtx);
         Top::get(opCtx->getServiceContext())
@@ -713,7 +673,6 @@ bool getFleCrudProcessed(OperationContext* opCtx,
 WriteResult performInserts(OperationContext* opCtx,
                            const write_ops::InsertCommandRequest& wholeOp,
                            OperationSource source) {
-
     // Insert performs its own retries, so we should only be within a WriteUnitOfWork when run in a
     // transaction.
     auto txnParticipant = TransactionParticipant::get(opCtx);
@@ -772,11 +731,20 @@ WriteResult performInserts(OperationContext* opCtx,
     const size_t maxBatchBytes = write_ops::insertVectorMaxBytes;
     batch.reserve(std::min(wholeOp.getDocuments().size(), maxBatchSize));
 
+    // If 'wholeOp.getBypassEmptyTsReplacement()' is true or if 'source' is 'kFromMigrate', set
+    // "bypassEmptyTsReplacement=true" for fixDocumentForInsert().
+    const bool bypassEmptyTsReplacement = (source == OperationSource::kFromMigrate) ||
+        static_cast<bool>(wholeOp.getBypassEmptyTsReplacement());
+
     for (auto&& doc : wholeOp.getDocuments()) {
         const bool isLastDoc = (&doc == &wholeOp.getDocuments().back());
         bool containsDotsAndDollarsField = false;
-        auto fixedDoc = fixDocumentForInsert(opCtx, doc, &containsDotsAndDollarsField);
+
+        auto fixedDoc = fixDocumentForInsert(
+            opCtx, doc, bypassEmptyTsReplacement, &containsDotsAndDollarsField);
+
         const StmtId stmtId = getStmtIdForWriteOp(opCtx, wholeOp, stmtIdIndex++);
+
         const bool wasAlreadyExecuted = opCtx->isRetryableWrite() &&
             txnParticipant.checkStatementExecutedNoOplogEntryFetch(opCtx, stmtId);
 
@@ -1018,6 +986,7 @@ static SingleWriteResult performSingleUpdateOpWithDupKeyRetry(
     const write_ops::UpdateOpEntry& op,
     LegacyRuntimeConstants runtimeConstants,
     const boost::optional<BSONObj>& letParams,
+    const OptionalBool& bypassEmptyTsReplacement,
     OperationSource source,
     bool forgoOpCounterIncrements) {
     globalOpCounters.gotUpdate();
@@ -1046,6 +1015,7 @@ static SingleWriteResult performSingleUpdateOpWithDupKeyRetry(
     if (letParams) {
         request.setLetParameters(std::move(letParams));
     }
+    request.setBypassEmptyTsReplacement(bypassEmptyTsReplacement);
     request.setStmtIds(stmtIds);
     request.setYieldPolicy(PlanYieldPolicy::YieldPolicy::YIELD_AUTO);
     request.setSource(source);
@@ -1187,6 +1157,7 @@ WriteResult performUpdates(OperationContext* opCtx,
                                                      singleOp,
                                                      runtimeConstants,
                                                      wholeOp.getLet(),
+                                                     wholeOp.getBypassEmptyTsReplacement(),
                                                      source,
                                                      forgoOpCounterIncrements);
             out.results.emplace_back(reply);