SERVER-65895 Also create applier metrics for donor shards with no progress document

2 files changed, 89 insertions, 25 deletions

diff --git a/src/mongo/db/s/resharding/resharding_recipient_service.cpp b/src/mongo/db/s/resharding/resharding_recipient_service.cpp
index 01d313c73a5..97fc1e9c6f2 100644
--- a/src/mongo/db/s/resharding/resharding_recipient_service.cpp
+++ b/src/mongo/db/s/resharding/resharding_recipient_service.cpp
@@ -559,7 +559,9 @@ ReshardingRecipientService::RecipientStateMachine::_makeDataReplication(Operatio
                 std::make_unique<ReshardingOplogApplierMetrics>(_metricsNew.get(), boost::none));
         }
     } else {
-        invariant(_applierMetricsMap.size() == _donorShards.size());
+        invariant(_applierMetricsMap.size() == _donorShards.size(),
+                  str::stream() << "applier metrics map size: " << _applierMetricsMap.size()
+                                << " != donor shards count: " << _donorShards.size());
     }
 
     return _dataReplicationFactory(opCtx,
@@ -1135,32 +1137,29 @@ void ReshardingRecipientService::RecipientStateMachine::_restoreMetrics(
             }
         }
 
-        {
-            AutoGetCollection progressApplierColl(
-                opCtx.get(), NamespaceString::kReshardingApplierProgressNamespace, MODE_IS);
-            if (progressApplierColl) {
-                BSONObj result;
-                Helpers::findOne(
-                    opCtx.get(),
-                    progressApplierColl.getCollection(),
-                    BSON(ReshardingOplogApplierProgress::kOplogSourceIdFieldName
-                         << (ReshardingSourceId{_metadata.getReshardingUUID(), donor.getShardId()})
-                                .toBSON()),
-                    result);
-
-                if (!result.isEmpty()) {
-                    auto progressDoc = ReshardingOplogApplierProgress::parse(
-                        IDLParserErrorContext("resharding-recipient-service-progress-doc"), result);
-                    oplogEntriesApplied += progressDoc.getNumEntriesApplied();
-
-                    if (ShardingDataTransformMetrics::isEnabled()) {
-                        progressDocList.emplace_back(donor.getShardId(), progressDoc);
-                    }
-                }
-            } else {
-                progressDocList.emplace_back(donor.getShardId(), boost::none);
+        boost::optional<ReshardingOplogApplierProgress> progressDoc;
+        AutoGetCollection progressApplierColl(
+            opCtx.get(), NamespaceString::kReshardingApplierProgressNamespace, MODE_IS);
+        if (progressApplierColl) {
+            BSONObj result;
+            Helpers::findOne(
+                opCtx.get(),
+                progressApplierColl.getCollection(),
+                BSON(ReshardingOplogApplierProgress::kOplogSourceIdFieldName
+                     << (ReshardingSourceId{_metadata.getReshardingUUID(), donor.getShardId()})
+                            .toBSON()),
+                result);
+
+            if (!result.isEmpty()) {
+                progressDoc = ReshardingOplogApplierProgress::parse(
+                    IDLParserErrorContext("resharding-recipient-service-progress-doc"), result);
+                oplogEntriesApplied += progressDoc->getNumEntriesApplied();
             }
         }
+
+        if (ShardingDataTransformMetrics::isEnabled()) {
+            progressDocList.emplace_back(donor.getShardId(), progressDoc);
+        }
     }
 
     // Restore stats here where interrupts will never occur, this is to ensure we will only update
diff --git a/src/mongo/db/s/resharding/resharding_recipient_service_test.cpp b/src/mongo/db/s/resharding/resharding_recipient_service_test.cpp
index 518df23e17b..6bddf3386c2 100644
--- a/src/mongo/db/s/resharding/resharding_recipient_service_test.cpp
+++ b/src/mongo/db/s/resharding/resharding_recipient_service_test.cpp
@@ -31,6 +31,7 @@
 
 #include "mongo/platform/basic.h"
 
+#include "mongo/bson/timestamp.h"
 #include "mongo/db/dbdirectclient.h"
 #include "mongo/db/op_observer_noop.h"
 #include "mongo/db/op_observer_registry.h"
@@ -48,6 +49,7 @@
 #include "mongo/db/s/resharding/resharding_recipient_service.h"
 #include "mongo/db/s/resharding/resharding_recipient_service_external_state.h"
 #include "mongo/db/s/resharding/resharding_service_test_helpers.h"
+#include "mongo/idl/server_parameter_test_util.h"
 #include "mongo/logv2/log.h"
 #include "mongo/unittest/death_test.h"
 #include "mongo/util/fail_point.h"
@@ -267,6 +269,15 @@ public:
         resharding::data_copy::ensureCollectionExists(opCtx, recipientDoc.getSourceNss(), options);
     }
 
+    void createTempReshardingCollection(OperationContext* opCtx,
+                                        const ReshardingRecipientDocument& recipientDoc) {
+        CollectionOptions options;
+        options.uuid = recipientDoc.getReshardingUUID();
+        resharding::data_copy::ensureCollectionDropped(opCtx, recipientDoc.getTempReshardingNss());
+        resharding::data_copy::ensureCollectionExists(
+            opCtx, recipientDoc.getTempReshardingNss(), options);
+    }
+
     void notifyToStartCloning(OperationContext* opCtx,
                               RecipientStateMachine& recipient,
                               const ReshardingRecipientDocument& recipientDoc) {
@@ -859,5 +870,59 @@ TEST_F(ReshardingRecipientServiceTest, RestoreMetricsAfterStepUp) {
     }
 }
 
+TEST_F(ReshardingRecipientServiceTest, RestoreMetricsAfterStepUpWithMissingProgressDoc) {
+    RAIIServerParameterControllerForTest featureFlagController(
+        "featureFlagShardingDataTransformMetrics", true);
+
+    auto doc = makeStateDocument(false);
+    auto instanceId =
+        BSON(ReshardingRecipientDocument::kReshardingUUIDFieldName << doc.getReshardingUUID());
+    auto opCtx = makeOperationContext();
+
+    auto donorShards = doc.getDonorShards();
+    auto insertFn = [&](const NamespaceString nss, const InsertStatement insertStatement) {
+        resharding::data_copy::ensureCollectionExists(opCtx.get(), nss, CollectionOptions());
+
+        std::vector<InsertStatement> inserts{insertStatement};
+        resharding::data_copy::insertBatch(opCtx.get(), nss, inserts);
+    };
+
+    for (unsigned i = 0; i < donorShards.size(); i++) {
+        if (i == 0) {
+            continue;
+        }
+
+        const auto& donor = donorShards[i];
+
+        // Setup oplogBuffer collection.
+        ReshardingDonorOplogId donorOplogId{{20, i}, {19, 0}};
+        insertFn(getLocalOplogBufferNamespace(doc.getSourceUUID(), donor.getShardId()),
+                 InsertStatement{BSON("_id" << donorOplogId.toBSON())});
+
+        // Setup reshardingApplierProgress collection.
+        ReshardingOplogApplierProgress progressDoc(
+            {doc.getReshardingUUID(), donor.getShardId()}, donorOplogId, 10 /* numOplogApplied */);
+        insertFn(NamespaceString::kReshardingApplierProgressNamespace,
+                 InsertStatement{progressDoc.toBSON()});
+    }
+
+    auto mutableState = doc.getMutableState();
+    mutableState.setState(RecipientStateEnum::kApplying);
+    doc.setMutableState(mutableState);
+    doc.setCloneTimestamp(Timestamp{10, 0});
+    doc.setStartConfigTxnCloneTime(Date_t::now());
+
+    auto metadata = doc.getCommonReshardingMetadata();
+    metadata.setStartTime(Date_t::now());
+    doc.setCommonReshardingMetadata(metadata);
+
+    createTempReshardingCollection(opCtx.get(), doc);
+
+    RecipientStateMachine::insertStateDocument(opCtx.get(), doc);
+    auto recipient = RecipientStateMachine::getOrCreate(opCtx.get(), _service, doc.toBSON());
+    notifyReshardingCommitting(opCtx.get(), *recipient, doc);
+    ASSERT_OK(recipient->getCompletionFuture().getNoThrow());
+}
+
 }  // namespace
 }  // namespace mongo