Update upstream source from tag 'upstream/6.0.0'master

Update to upstream version '6.0.0'
with Debian dir 5604a80ec1c96ca76f25f40d78e6ef855abec322

1 files changed, 35 insertions, 64 deletions

diff --git a/src/mongo/s/chunk_manager_targeter.cpp b/src/mongo/s/chunk_manager_targeter.cpp
index f750554ea56..bdb386420c6 100644
--- a/src/mongo/s/chunk_manager_targeter.cpp
+++ b/src/mongo/s/chunk_manager_targeter.cpp
@@ -43,8 +43,6 @@
 #include "mongo/db/query/canonical_query.h"
 #include "mongo/db/query/collation/collation_index_key.h"
 #include "mongo/db/query/collation/collator_factory_interface.h"
-#include "mongo/db/stats/counters.h"
-#include "mongo/db/timeseries/metadata.h"
 #include "mongo/db/timeseries/timeseries_constants.h"
 #include "mongo/db/timeseries/timeseries_options.h"
 #include "mongo/db/timeseries/timeseries_update_delete_util.h"
@@ -65,8 +63,6 @@
 namespace mongo {
 namespace {
 
-MONGO_FAIL_POINT_DEFINE(waitForDatabaseToBeDropped);
-
 enum CompareResult { CompareResult_Unknown, CompareResult_GTE, CompareResult_LT };
 
 constexpr auto kIdFieldName = "_id"_sd;
@@ -230,8 +226,6 @@ bool isMetadataDifferent(const ChunkManager& managerA, const ChunkManager& manag
 
 }  // namespace
 
-const size_t ChunkManagerTargeter::kMaxDatabaseCreationAttempts = 3;
-
 ChunkManagerTargeter::ChunkManagerTargeter(OperationContext* opCtx,
                                            const NamespaceString& nss,
                                            boost::optional<OID> targetEpoch)
@@ -245,45 +239,13 @@ ChunkManagerTargeter::ChunkManagerTargeter(OperationContext* opCtx,
  * user request is on the view namespace, we implicity tranform the request to the buckets namepace.
  */
 ChunkManager ChunkManagerTargeter::_init(OperationContext* opCtx, bool refresh) {
-    const auto createDatabaseAndGetRoutingInfo = [&opCtx, &refresh](const NamespaceString& nss) {
-        size_t attempts = 1;
-        while (true) {
-            try {
-                cluster::createDatabase(opCtx, nss.db());
-
-                if (refresh) {
-                    uassertStatusOK(
-                        Grid::get(opCtx)->catalogCache()->getCollectionRoutingInfoWithRefresh(opCtx,
-                                                                                              nss));
-                }
-
-                if (MONGO_unlikely(waitForDatabaseToBeDropped.shouldFail())) {
-                    LOGV2(8314600, "Hanging due to waitForDatabaseToBeDropped fail point");
-                    waitForDatabaseToBeDropped.pauseWhileSet(opCtx);
-                }
-
-                return uassertStatusOK(getCollectionRoutingInfoForTxnCmd(opCtx, nss));
-            } catch (const ExceptionFor<ErrorCodes::NamespaceNotFound>&) {
-                LOGV2_INFO(8314601,
-                           "Failed initialization of routing info because the database has been "
-                           "concurrently dropped",
-                           logAttrs(nss),
-                           "attemptNumber"_attr = attempts,
-                           "maxAttempts"_attr = kMaxDatabaseCreationAttempts);
-
-                if (attempts++ >= kMaxDatabaseCreationAttempts) {
-                    // The maximum number of attempts has been reached, so the procedure fails as it
-                    // could be a logical error. At this point, it is unlikely that the error is
-                    // caused by concurrent drop database operations.
-                    throw;
-                }
-            }
-        }
-    };
+    cluster::createDatabase(opCtx, _nss.db());
 
-    createDatabaseAndGetRoutingInfo(_nss);
-
-    auto cm = createDatabaseAndGetRoutingInfo(_nss);
+    if (refresh) {
+        uassertStatusOK(
+            Grid::get(opCtx)->catalogCache()->getCollectionRoutingInfoWithRefresh(opCtx, _nss));
+    }
+    auto cm = uassertStatusOK(getCollectionRoutingInfoForTxnCmd(opCtx, _nss));
 
     // For a sharded time-series collection, only the underlying buckets collection is stored on the
     // config servers. If the user operation is on the time-series view namespace, we should check
@@ -299,18 +261,27 @@ ChunkManager ChunkManagerTargeter::_init(OperationContext* opCtx, bool refresh)
     //    back to the view namespace and reset '_isRequestOnTimeseriesViewNamespace'.
     if (!cm.isSharded() && !_nss.isTimeseriesBucketsCollection()) {
         auto bucketsNs = _nss.makeTimeseriesBucketsNamespace();
-        auto bucketsPlacementInfo = createDatabaseAndGetRoutingInfo(bucketsNs);
-        if (bucketsPlacementInfo.isSharded()) {
+        if (refresh) {
+            uassertStatusOK(Grid::get(opCtx)->catalogCache()->getCollectionRoutingInfoWithRefresh(
+                opCtx, bucketsNs));
+        }
+        auto bucketsRoutingInfo =
+            uassertStatusOK(getCollectionRoutingInfoForTxnCmd(opCtx, bucketsNs));
+        if (bucketsRoutingInfo.isSharded()) {
             _nss = bucketsNs;
-            cm = std::move(bucketsPlacementInfo);
+            cm = std::move(bucketsRoutingInfo);
             _isRequestOnTimeseriesViewNamespace = true;
         }
     } else if (!cm.isSharded() && _isRequestOnTimeseriesViewNamespace) {
         // This can happen if a sharded time-series collection is dropped and re-created. Then we
         // need to reset the namepace to the original namespace.
         _nss = _nss.getTimeseriesViewNamespace();
-        auto newCm = createDatabaseAndGetRoutingInfo(_nss);
-        cm = std::move(newCm);
+
+        if (refresh) {
+            uassertStatusOK(
+                Grid::get(opCtx)->catalogCache()->getCollectionRoutingInfoWithRefresh(opCtx, _nss));
+        }
+        cm = uassertStatusOK(getCollectionRoutingInfoForTxnCmd(opCtx, _nss));
         _isRequestOnTimeseriesViewNamespace = false;
     }
 
@@ -352,7 +323,7 @@ BSONObj ChunkManagerTargeter::extractBucketsShardKeyFromTimeseriesDoc(
 
     if (auto metaField = timeseriesOptions.getMetaField(); metaField) {
         if (auto metaElement = doc.getField(*metaField); !metaElement.eoo()) {
-            timeseries::metadata::normalize(metaElement, builder, timeseries::kBucketMetaFieldName);
+            builder.appendAs(metaElement, timeseries::kBucketMetaFieldName);
         }
     }
 
@@ -415,10 +386,6 @@ std::vector<ShardEndpoint> ChunkManagerTargeter::targetUpdate(OperationContext*
 
     const auto& updateOp = itemRef.getUpdate();
 
-    if (updateOp.getMulti()) {
-        updateManyCount.increment(1);
-    }
-
     // If the collection is not sharded, forward the update to the primary shard.
     if (!_cm.isSharded()) {
         // TODO (SERVER-51070): Remove the boost::none when the config server can support
@@ -535,11 +502,8 @@ std::vector<ShardEndpoint> ChunkManagerTargeter::targetDelete(OperationContext*
                                                                itemRef.getLet(),
                                                                itemRef.getLegacyRuntimeConstants());
 
-    if (deleteOp.getMulti()) {
-        deleteManyCount.increment(1);
-    }
-
     BSONObj deleteQuery = deleteOp.getQ();
+    BSONObj shardKey;
     if (_cm.isSharded()) {
         if (_isRequestOnTimeseriesViewNamespace) {
             uassert(ErrorCodes::NotImplemented,
@@ -565,13 +529,20 @@ std::vector<ShardEndpoint> ChunkManagerTargeter::targetDelete(OperationContext*
                 deleteQuery = BSONObj();
             }
         }
+
+        // Sharded collections have the following further requirements for targeting:
+        //
+        // Limit-1 deletes must be targeted exactly by shard key *or* exact _id
+        shardKey =
+            uassertStatusOK(_cm.getShardKeyPattern().extractShardKeyFromQuery(expCtx, deleteQuery));
     }
 
-    // We first try to target based on the delete's query. It is always valid to forward any delete
-    // to a single shard, so return immediately if we are able to target a single shard.
-    auto endpoints = uassertStatusOK(_targetQuery(expCtx, deleteQuery, collation));
-    if (endpoints.size() == 1) {
-        return endpoints;
+    // Target the shard key or delete query
+    if (!shardKey.isEmpty()) {
+        auto swEndpoint = _targetShardKey(shardKey, collation);
+        if (swEndpoint.isOK()) {
+            return std::vector{std::move(swEndpoint.getValue())};
+        }
     }
 
     // We failed to target a single shard.
@@ -600,7 +571,7 @@ std::vector<ShardEndpoint> ChunkManagerTargeter::targetDelete(OperationContext*
                           << ", shard key pattern: " << _cm.getShardKeyPattern().toString(),
             !_cm.isSharded() || deleteOp.getMulti() || isExactIdQuery(opCtx, *cq, _cm));
 
-    return endpoints;
+    return uassertStatusOK(_targetQuery(expCtx, deleteQuery, collation));
 }
 
 StatusWith<std::vector<ShardEndpoint>> ChunkManagerTargeter::_targetQuery(