SERVER-63811 Add check so documents stage runs when db does not exist

11 files changed, 91 insertions, 27 deletions

diff --git a/src/mongo/db/pipeline/dispatch_shard_pipeline_test.cpp b/src/mongo/db/pipeline/dispatch_shard_pipeline_test.cpp
index 069a7e2f0b2..8f04441de80 100644
--- a/src/mongo/db/pipeline/dispatch_shard_pipeline_test.cpp
+++ b/src/mongo/db/pipeline/dispatch_shard_pipeline_test.cpp
@@ -53,10 +53,11 @@ TEST_F(DispatchShardPipelineTest, DoesNotSplitPipelineIfTargetingOneShard) {
     const Document serializedCommand = aggregation_request_helper::serializeToCommandDoc(
         AggregateCommandRequest(expCtx()->ns, stages));
     const bool hasChangeStream = false;
+    const bool startsWithDocuments = false;
 
     auto future = launchAsync([&] {
         auto results = sharded_agg_helpers::dispatchShardPipeline(
-            serializedCommand, hasChangeStream, std::move(pipeline));
+            serializedCommand, hasChangeStream, startsWithDocuments, std::move(pipeline));
         ASSERT_EQ(results.remoteCursors.size(), 1UL);
         ASSERT(!results.splitPipeline);
     });
@@ -84,10 +85,11 @@ TEST_F(DispatchShardPipelineTest, DoesSplitPipelineIfMatchSpansTwoShards) {
     const Document serializedCommand = aggregation_request_helper::serializeToCommandDoc(
         AggregateCommandRequest(expCtx()->ns, stages));
     const bool hasChangeStream = false;
+    const bool startsWithDocuments = false;
 
     auto future = launchAsync([&] {
         auto results = sharded_agg_helpers::dispatchShardPipeline(
-            serializedCommand, hasChangeStream, std::move(pipeline));
+            serializedCommand, hasChangeStream, startsWithDocuments, std::move(pipeline));
         ASSERT_EQ(results.remoteCursors.size(), 2UL);
         ASSERT(bool(results.splitPipeline));
     });
@@ -118,10 +120,11 @@ TEST_F(DispatchShardPipelineTest, DispatchShardPipelineRetriesOnNetworkError) {
     const Document serializedCommand = aggregation_request_helper::serializeToCommandDoc(
         AggregateCommandRequest(expCtx()->ns, stages));
     const bool hasChangeStream = false;
+    const bool startsWithDocuments = false;
     auto future = launchAsync([&] {
         // Shouldn't throw.
         auto results = sharded_agg_helpers::dispatchShardPipeline(
-            serializedCommand, hasChangeStream, std::move(pipeline));
+            serializedCommand, hasChangeStream, startsWithDocuments, std::move(pipeline));
         ASSERT_EQ(results.remoteCursors.size(), 2UL);
         ASSERT(bool(results.splitPipeline));
     });
@@ -163,11 +166,14 @@ TEST_F(DispatchShardPipelineTest, DispatchShardPipelineDoesNotRetryOnStaleConfig
     const Document serializedCommand = aggregation_request_helper::serializeToCommandDoc(
         AggregateCommandRequest(expCtx()->ns, stages));
     const bool hasChangeStream = false;
+    const bool startsWithDocuments = false;
+
     auto future = launchAsync([&] {
-        ASSERT_THROWS_CODE(sharded_agg_helpers::dispatchShardPipeline(
-                               serializedCommand, hasChangeStream, std::move(pipeline)),
-                           AssertionException,
-                           ErrorCodes::StaleConfig);
+        ASSERT_THROWS_CODE(
+            sharded_agg_helpers::dispatchShardPipeline(
+                serializedCommand, hasChangeStream, startsWithDocuments, std::move(pipeline)),
+            AssertionException,
+            ErrorCodes::StaleConfig);
     });
 
     // Mock out an error response.
@@ -197,15 +203,17 @@ TEST_F(DispatchShardPipelineTest, WrappedDispatchDoesRetryOnStaleConfigError) {
     const Document serializedCommand = aggregation_request_helper::serializeToCommandDoc(
         AggregateCommandRequest(expCtx()->ns, stages));
     const bool hasChangeStream = false;
+    const bool startsWithDocuments = false;
     auto future = launchAsync([&] {
         // Shouldn't throw.
         sharding::router::CollectionRouter router(getServiceContext(), kTestAggregateNss);
-        auto results = router.route(operationContext(),
-                                    "dispatch shard pipeline"_sd,
-                                    [&](OperationContext* opCtx, const ChunkManager& cm) {
-                                        return sharded_agg_helpers::dispatchShardPipeline(
-                                            serializedCommand, hasChangeStream, pipeline->clone());
-                                    });
+        auto results = router.route(
+            operationContext(),
+            "dispatch shard pipeline"_sd,
+            [&](OperationContext* opCtx, const ChunkManager& cm) {
+                return sharded_agg_helpers::dispatchShardPipeline(
+                    serializedCommand, hasChangeStream, startsWithDocuments, pipeline->clone());
+            });
         ASSERT_EQ(results.remoteCursors.size(), 1UL);
         ASSERT(!bool(results.splitPipeline));
     });
diff --git a/src/mongo/db/pipeline/document_source_documents.cpp b/src/mongo/db/pipeline/document_source_documents.cpp
index 5d4e02c71ea..934116a7c27 100644
--- a/src/mongo/db/pipeline/document_source_documents.cpp
+++ b/src/mongo/db/pipeline/document_source_documents.cpp
@@ -44,7 +44,7 @@ namespace mongo {
 using boost::intrusive_ptr;
 
 REGISTER_DOCUMENT_SOURCE(documents,
-                         LiteParsedDocumentSourceDefault::parse,
+                         DocumentSourceDocuments::LiteParsed::parse,
                          DocumentSourceDocuments::createFromBson,
                          AllowedWithApiStrict::kAlways);
 
diff --git a/src/mongo/db/pipeline/document_source_documents.h b/src/mongo/db/pipeline/document_source_documents.h
index a6ab70e0c59..6048ed551bb 100644
--- a/src/mongo/db/pipeline/document_source_documents.h
+++ b/src/mongo/db/pipeline/document_source_documents.h
@@ -35,6 +35,30 @@
 namespace mongo {
 
 namespace DocumentSourceDocuments {
+class LiteParsed : public LiteParsedDocumentSource {
+public:
+    static std::unique_ptr<LiteParsed> parse(const NamespaceString& nss, const BSONElement& spec) {
+        return std::make_unique<LiteParsed>(spec.fieldName());
+    }
+
+    LiteParsed(std::string parseTimeName) : LiteParsedDocumentSource(std::move(parseTimeName)) {}
+
+    stdx::unordered_set<NamespaceString> getInvolvedNamespaces() const final {
+        return stdx::unordered_set<NamespaceString>();
+    }
+
+    PrivilegeVector requiredPrivileges(bool isMongos, bool bypassDocumentValidation) const final {
+        return {};
+    }
+
+    bool isDocuments() const final {
+        return true;
+    }
+
+    bool allowedToPassthroughFromMongos() const final {
+        return false;
+    }
+};
 
 static constexpr StringData kStageName = "$documents"_sd;
 
diff --git a/src/mongo/db/pipeline/lite_parsed_document_source.h b/src/mongo/db/pipeline/lite_parsed_document_source.h
index 34648576917..3e1dc6ea6c6 100644
--- a/src/mongo/db/pipeline/lite_parsed_document_source.h
+++ b/src/mongo/db/pipeline/lite_parsed_document_source.h
@@ -171,6 +171,13 @@ public:
     }
 
     /**
+     * Returns true if this is a $documents stage.
+     */
+    virtual bool isDocuments() const {
+        return false;
+    }
+
+    /**
      * Returns true if this stage does not require an input source.
      */
     virtual bool isInitialSource() const {
diff --git a/src/mongo/db/pipeline/lite_parsed_pipeline.h b/src/mongo/db/pipeline/lite_parsed_pipeline.h
index 712d23c32bd..d02ff8ee70e 100644
--- a/src/mongo/db/pipeline/lite_parsed_pipeline.h
+++ b/src/mongo/db/pipeline/lite_parsed_pipeline.h
@@ -123,6 +123,13 @@ public:
     }
 
     /**
+     * Returns true if the pipeline begins with a $documents stage.
+     */
+    bool startsWithDocuments() const {
+        return !_stageSpecs.empty() && _stageSpecs.front()->isDocuments();
+    }
+
+    /**
      * Returns true if the pipeline has a $changeStream stage.
      */
     bool hasChangeStream() const {
diff --git a/src/mongo/db/pipeline/sharded_agg_helpers.cpp b/src/mongo/db/pipeline/sharded_agg_helpers.cpp
index fed03ec5079..7940fc0b97b 100644
--- a/src/mongo/db/pipeline/sharded_agg_helpers.cpp
+++ b/src/mongo/db/pipeline/sharded_agg_helpers.cpp
@@ -769,9 +769,11 @@ std::unique_ptr<Pipeline, PipelineDeleter> targetShardsAndAddMergeCursors(
 
     LiteParsedPipeline liteParsedPipeline(aggRequest);
     auto hasChangeStream = liteParsedPipeline.hasChangeStream();
+    auto startsWithDocuments = liteParsedPipeline.startsWithDocuments();
     auto shardDispatchResults =
         dispatchShardPipeline(aggregation_request_helper::serializeToCommandDoc(aggRequest),
                               hasChangeStream,
+                              startsWithDocuments,
                               std::move(pipeline),
                               shardTargetingPolicy,
                               std::move(readConcern));
@@ -1007,6 +1009,7 @@ BSONObj createCommandForTargetedShards(const boost::intrusive_ptr<ExpressionCont
 DispatchShardPipelineResults dispatchShardPipeline(
     Document serializedCommand,
     bool hasChangeStream,
+    bool startsWithDocuments,
     std::unique_ptr<Pipeline, PipelineDeleter> pipeline,
     ShardTargetingPolicy shardTargetingPolicy,
     boost::optional<BSONObj> readConcern) {
@@ -1051,7 +1054,7 @@ DispatchShardPipelineResults dispatchShardPipeline(
         : expCtx->getCollatorBSON();
 
     // Determine whether we can run the entire aggregation on a single shard.
-    const bool mustRunOnAll = mustRunOnAllShards(expCtx->ns, hasChangeStream);
+    const bool mustRunOnAll = mustRunOnAllShards(expCtx->ns, hasChangeStream, startsWithDocuments);
     std::set<ShardId> shardIds = getTargetedShards(
         expCtx, mustRunOnAll, executionNsRoutingInfo, shardQuery, shardTargetingCollation);
 
@@ -1486,9 +1489,11 @@ BSONObj targetShardsForExplain(Pipeline* ownedPipeline) {
     AggregateCommandRequest aggRequest(expCtx->ns, rawStages);
     LiteParsedPipeline liteParsedPipeline(aggRequest);
     auto hasChangeStream = liteParsedPipeline.hasChangeStream();
+    auto startsWithDocuments = liteParsedPipeline.startsWithDocuments();
     auto shardDispatchResults =
         dispatchShardPipeline(aggregation_request_helper::serializeToCommandDoc(aggRequest),
                               hasChangeStream,
+                              startsWithDocuments,
                               std::move(pipeline));
     BSONObjBuilder explainBuilder;
     auto appendStatus =
@@ -1523,11 +1528,13 @@ Shard::RetryPolicy getDesiredRetryPolicy(OperationContext* opCtx) {
     return Shard::RetryPolicy::kIdempotent;
 }
 
-bool mustRunOnAllShards(const NamespaceString& nss, bool hasChangeStream) {
+bool mustRunOnAllShards(const NamespaceString& nss,
+                        bool hasChangeStream,
+                        bool startsWithDocuments) {
     // The following aggregations must be routed to all shards:
     // - Any collectionless aggregation, such as non-localOps $currentOp.
     // - Any aggregation which begins with a $changeStream stage.
-    return nss.isCollectionlessAggregateNS() || hasChangeStream;
+    return !startsWithDocuments && (nss.isCollectionlessAggregateNS() || hasChangeStream);
 }
 
 std::unique_ptr<Pipeline, PipelineDeleter> attachCursorToPipeline(
diff --git a/src/mongo/db/pipeline/sharded_agg_helpers.h b/src/mongo/db/pipeline/sharded_agg_helpers.h
index 5f30f6f6a93..f69d422da0c 100644
--- a/src/mongo/db/pipeline/sharded_agg_helpers.h
+++ b/src/mongo/db/pipeline/sharded_agg_helpers.h
@@ -124,6 +124,7 @@ SplitPipeline splitPipeline(std::unique_ptr<Pipeline, PipelineDeleter> pipeline)
 DispatchShardPipelineResults dispatchShardPipeline(
     Document serializedCommand,
     bool hasChangeStream,
+    bool startsWithDocuments,
     std::unique_ptr<Pipeline, PipelineDeleter> pipeline,
     ShardTargetingPolicy shardTargetingPolicy = ShardTargetingPolicy::kAllowed,
     boost::optional<BSONObj> readConcern = boost::none);
@@ -179,7 +180,7 @@ StatusWith<ChunkManager> getExecutionNsRoutingInfo(OperationContext* opCtx,
 /**
  * Returns true if an aggregation over 'nss' must run on all shards.
  */
-bool mustRunOnAllShards(const NamespaceString& nss, bool hasChangeStream);
+bool mustRunOnAllShards(const NamespaceString& nss, bool hasChangeStream, bool startsWithDocuments);
 
 /**
  * Retrieves the desired retry policy based on whether the default writeConcern is set on 'opCtx'.
diff --git a/src/mongo/s/commands/cluster_map_reduce_agg.cpp b/src/mongo/s/commands/cluster_map_reduce_agg.cpp
index 52a61ab9325..3591aa20b17 100644
--- a/src/mongo/s/commands/cluster_map_reduce_agg.cpp
+++ b/src/mongo/s/commands/cluster_map_reduce_agg.cpp
@@ -187,6 +187,7 @@ bool runAggregationMapReduce(OperationContext* opCtx,
                                                                cm,
                                                                involvedNamespaces,
                                                                false,   // hasChangeStream
+                                                               false,   // startsWithDocuments
                                                                true,    // allowedToPassthrough
                                                                false);  // perShardCursor
     try {
@@ -230,7 +231,8 @@ bool runAggregationMapReduce(OperationContext* opCtx,
                     namespaces,
                     privileges,
                     &tempResults,
-                    false));  // hasChangeStream
+                    false,    // hasChangeStream
+                    false));  // startsWithDocuments
                 break;
             }
 
diff --git a/src/mongo/s/query/cluster_aggregate.cpp b/src/mongo/s/query/cluster_aggregate.cpp
index 7c09e31d1ef..6a0b135fef4 100644
--- a/src/mongo/s/query/cluster_aggregate.cpp
+++ b/src/mongo/s/query/cluster_aggregate.cpp
@@ -310,6 +310,7 @@ Status ClusterAggregate::runAggregate(OperationContext* opCtx,
     auto hasChangeStream = liteParsedPipeline.hasChangeStream();
     auto involvedNamespaces = liteParsedPipeline.getInvolvedNamespaces();
     auto shouldDoFLERewrite = ::mongo::shouldDoFLERewrite(request);
+    auto startsWithDocuments = liteParsedPipeline.startsWithDocuments();
 
     // If the routing table is not already taken by the higher level, fill it now.
     if (!cm) {
@@ -339,7 +340,7 @@ Status ClusterAggregate::runAggregate(OperationContext* opCtx,
 
         if (executionNsRoutingInfoStatus.isOK()) {
             cm = std::move(executionNsRoutingInfoStatus.getValue());
-        } else if (!(hasChangeStream &&
+        } else if (!((hasChangeStream || startsWithDocuments) &&
                      executionNsRoutingInfoStatus == ErrorCodes::NamespaceNotFound)) {
             appendEmptyResultSetWithStatus(
                 opCtx, namespaces.requestedNss, executionNsRoutingInfoStatus.getStatus(), result);
@@ -403,6 +404,7 @@ Status ClusterAggregate::runAggregate(OperationContext* opCtx,
         cm,
         involvedNamespaces,
         hasChangeStream,
+        startsWithDocuments,
         allowedToPassthrough,
         request.getPassthroughToShard().has_value());
 
@@ -477,7 +479,8 @@ Status ClusterAggregate::runAggregate(OperationContext* opCtx,
                     namespaces,
                     privileges,
                     result,
-                    hasChangeStream);
+                    hasChangeStream,
+                    startsWithDocuments);
             }
             case cluster_aggregation_planner::AggregationTargeter::TargetingPolicy::
                 kSpecificShardOnly: {
diff --git a/src/mongo/s/query/cluster_aggregation_planner.cpp b/src/mongo/s/query/cluster_aggregation_planner.cpp
index 3f970f2f8c1..5788b30a7d4 100644
--- a/src/mongo/s/query/cluster_aggregation_planner.cpp
+++ b/src/mongo/s/query/cluster_aggregation_planner.cpp
@@ -563,6 +563,7 @@ AggregationTargeter AggregationTargeter::make(
     boost::optional<ChunkManager> cm,
     stdx::unordered_set<NamespaceString> involvedNamespaces,
     bool hasChangeStream,
+    bool startsWithDocuments,
     bool allowedToPassthrough,
     bool perShardCursor) {
     if (perShardCursor) {
@@ -583,10 +584,11 @@ AggregationTargeter AggregationTargeter::make(
 
     // Determine whether this aggregation must be dispatched to all shards in the cluster.
     const bool mustRunOnAll =
-        sharded_agg_helpers::mustRunOnAllShards(executionNss, hasChangeStream);
+        sharded_agg_helpers::mustRunOnAllShards(executionNss, hasChangeStream, startsWithDocuments);
 
-    // If we don't have a routing table, then this is a $changeStream which must run on all shards.
-    invariant(cm || (mustRunOnAll && hasChangeStream));
+    // If we don't have a routing table, then this is either a $changeStream which must run on all
+    // shards or a $documents stage which must not.
+    invariant(cm || (mustRunOnAll && hasChangeStream) || (startsWithDocuments && !mustRunOnAll));
 
     // A pipeline is allowed to passthrough to the primary shard iff the following conditions are
     // met:
@@ -663,11 +665,12 @@ Status dispatchPipelineAndMerge(OperationContext* opCtx,
                                 const ClusterAggregate::Namespaces& namespaces,
                                 const PrivilegeVector& privileges,
                                 BSONObjBuilder* result,
-                                bool hasChangeStream) {
+                                bool hasChangeStream,
+                                bool startsWithDocuments) {
     auto expCtx = targeter.pipeline->getContext();
     // If not, split the pipeline as necessary and dispatch to the relevant shards.
     auto shardDispatchResults = sharded_agg_helpers::dispatchShardPipeline(
-        serializedCommand, hasChangeStream, std::move(targeter.pipeline));
+        serializedCommand, hasChangeStream, startsWithDocuments, std::move(targeter.pipeline));
 
     // If the operation is an explain, then we verify that it succeeded on all targeted
     // shards, write the results to the output builder, and return immediately.
diff --git a/src/mongo/s/query/cluster_aggregation_planner.h b/src/mongo/s/query/cluster_aggregation_planner.h
index 046d8eab6ca..78c919192db 100644
--- a/src/mongo/s/query/cluster_aggregation_planner.h
+++ b/src/mongo/s/query/cluster_aggregation_planner.h
@@ -82,6 +82,7 @@ struct AggregationTargeter {
         boost::optional<ChunkManager> cm,
         stdx::unordered_set<NamespaceString> involvedNamespaces,
         bool hasChangeStream,
+        bool startsWithDocuments,
         bool allowedToPassthrough,
         bool perShardCursor);
 
@@ -125,7 +126,8 @@ Status dispatchPipelineAndMerge(OperationContext* opCtx,
                                 const ClusterAggregate::Namespaces& namespaces,
                                 const PrivilegeVector& privileges,
                                 BSONObjBuilder* result,
-                                bool hasChangeStream);
+                                bool hasChangeStream,
+                                bool startsWithDocuments);
 
 /**
  * Similar to runPipelineOnPrimaryShard but allows $changeStreams. Intended for use by per shard