SERVER-19643 Aggregation directly from a shard does not use sharding filter.

4 files changed, 34 insertions, 5 deletions

diff --git a/jstests/aggregation/testshard1.js b/jstests/aggregation/testshard1.js
index 64cc46ffaa6..ef9f91bae70 100644
--- a/jstests/aggregation/testshard1.js
+++ b/jstests/aggregation/testshard1.js
@@ -1,4 +1,5 @@
 load('jstests/aggregation/extras/utils.js');
+load('jstests/libs/analyze_plan.js'); // For planHasStage.
 
 // Use this for aggregations that only have arrays or results of specified order.
 // It will check that cursors return the same results as non-cursors.
@@ -267,6 +268,16 @@ for (var shardName in res.shards) {
 }
 
 (function() {
+    jsTestLog("Do a sharded explain from a mongod, not mongos, to ensure that it does not have " +
+              "a SHARDING_FILTER stage.");
+    var shardDb = shardedAggTest.shard0.getDB('aggShard');
+    var res = shardDb.ts1.aggregate([{$match: {}}], {explain: true});
+    printjson(res);
+    assert.commandWorked(res);
+    assert(!planHasStage(res.stages[0].$cursor.queryPlanner.winningPlan, "SHARDING_FILTER"));
+}());
+
+(function() {
     jsTestLog('Testing a $match stage on the shard key.');
 
     var outCollection = 'testShardKeyMatchOut';
diff --git a/src/mongo/db/commands/pipeline_command.cpp b/src/mongo/db/commands/pipeline_command.cpp
index fc3537d1a49..31b80d4fda6 100644
--- a/src/mongo/db/commands/pipeline_command.cpp
+++ b/src/mongo/db/commands/pipeline_command.cpp
@@ -224,7 +224,7 @@ public:
             // This does mongod-specific stuff like creating the input PlanExecutor and adding
             // it to the front of the pipeline if needed.
             std::shared_ptr<PlanExecutor> input =
-                PipelineD::prepareCursorSource(txn, collection, pPipeline, pCtx);
+                PipelineD::prepareCursorSource(txn, collection, nss, pPipeline, pCtx);
             pPipeline->stitch();
 
             if (collection && input) {
diff --git a/src/mongo/db/pipeline/pipeline_d.cpp b/src/mongo/db/pipeline/pipeline_d.cpp
index 06114c90404..4cebf691c26 100644
--- a/src/mongo/db/pipeline/pipeline_d.cpp
+++ b/src/mongo/db/pipeline/pipeline_d.cpp
@@ -214,6 +214,7 @@ StatusWith<std::unique_ptr<PlanExecutor>> attemptToGetExecutor(
 shared_ptr<PlanExecutor> PipelineD::prepareCursorSource(
     OperationContext* txn,
     Collection* collection,
+    const NamespaceString& nss,
     const intrusive_ptr<Pipeline>& pPipeline,
     const intrusive_ptr<ExpressionContext>& pExpCtx) {
     // We will be modifying the source vector as we go.
@@ -292,8 +293,16 @@ shared_ptr<PlanExecutor> PipelineD::prepareCursorSource(
     }
 
     // Create the PlanExecutor.
-    auto exec = prepareExecutor(
-        txn, collection, pPipeline, pExpCtx, sortStage, deps, queryObj, &sortObj, &projForQuery);
+    auto exec = prepareExecutor(txn,
+                                collection,
+                                nss,
+                                pPipeline,
+                                pExpCtx,
+                                sortStage,
+                                deps,
+                                queryObj,
+                                &sortObj,
+                                &projForQuery);
 
     return addCursorSource(pPipeline, pExpCtx, exec, deps, queryObj, sortObj, projForQuery);
 }
@@ -301,6 +310,7 @@ shared_ptr<PlanExecutor> PipelineD::prepareCursorSource(
 std::shared_ptr<PlanExecutor> PipelineD::prepareExecutor(
     OperationContext* txn,
     Collection* collection,
+    const NamespaceString& nss,
     const intrusive_ptr<Pipeline>& pipeline,
     const intrusive_ptr<ExpressionContext>& expCtx,
     const intrusive_ptr<DocumentSourceSort>& sortStage,
@@ -326,8 +336,13 @@ std::shared_ptr<PlanExecutor> PipelineD::prepareExecutor(
     //
     // LATER - We should attempt to determine if the results from the query are returned in some
     // order so we can then apply other optimizations there are tickets for, such as SERVER-4507.
-    size_t plannerOpts = QueryPlannerParams::DEFAULT | QueryPlannerParams::INCLUDE_SHARD_FILTER |
-        QueryPlannerParams::NO_BLOCKING_SORT;
+    size_t plannerOpts = QueryPlannerParams::DEFAULT | QueryPlannerParams::NO_BLOCKING_SORT;
+
+    // If we are connecting directly to the shard rather than through a mongos, don't filter out
+    // orphaned documents.
+    if (ShardingState::get(txn)->needCollectionMetadata(txn, nss.ns())) {
+        plannerOpts |= QueryPlannerParams::INCLUDE_SHARD_FILTER;
+    }
 
     // The only way to get a text score is to let the query system handle the projection. In all
     // other cases, unless the query system can do an index-covered projection and avoid going to
diff --git a/src/mongo/db/pipeline/pipeline_d.h b/src/mongo/db/pipeline/pipeline_d.h
index a9993c6331b..95a08719b41 100644
--- a/src/mongo/db/pipeline/pipeline_d.h
+++ b/src/mongo/db/pipeline/pipeline_d.h
@@ -32,6 +32,7 @@
 #include <memory>
 
 #include "mongo/bson/bsonobj.h"
+#include "mongo/db/namespace_string.h"
 
 namespace mongo {
 class Collection;
@@ -79,6 +80,7 @@ public:
     static std::shared_ptr<PlanExecutor> prepareCursorSource(
         OperationContext* txn,
         Collection* collection,
+        const NamespaceString& nss,
         const boost::intrusive_ptr<Pipeline>& pPipeline,
         const boost::intrusive_ptr<ExpressionContext>& pExpCtx);
 
@@ -97,6 +99,7 @@ private:
     static std::shared_ptr<PlanExecutor> prepareExecutor(
         OperationContext* txn,
         Collection* collection,
+        const NamespaceString& nss,
         const boost::intrusive_ptr<Pipeline>& pipeline,
         const boost::intrusive_ptr<ExpressionContext>& expCtx,
         const boost::intrusive_ptr<DocumentSourceSort>& sortStage,