SERVER-37904 Add enableOverrideClusterChainingSetting server parameter

(cherry picked from commit 7422c1b638d2e4b200e7b72041d76f5702364d2b)

5 files changed, 60 insertions, 5 deletions

diff --git a/etc/backports_required_for_multiversion_tests.yml b/etc/backports_required_for_multiversion_tests.yml
index 572e3d98a42..da3236a4886 100644
--- a/etc/backports_required_for_multiversion_tests.yml
+++ b/etc/backports_required_for_multiversion_tests.yml
@@ -159,6 +159,8 @@ all:
     test_file: jstests/sharding/cleanup_orphaned_cmd_during_movechunk.js
   - ticket: SERVER-52906
     test_file: jstests/sharding/cleanup_orphaned_cmd_during_movechunk_hashed.js
+  - ticket: SERVER-37904
+    test_file: jstests/replsets/cluster_chaining_override.js
 
 suites:
 
diff --git a/jstests/replsets/cluster_chaining_override.js b/jstests/replsets/cluster_chaining_override.js
new file mode 100644
index 00000000000..5a913c3096e
--- /dev/null
+++ b/jstests/replsets/cluster_chaining_override.js
@@ -0,0 +1,40 @@
+/**
+ * Tests that if chaining is disabled, enabling the server parameter
+ * 'enableOverrideClusterChainingSetting' will allow the node to chain anyway.
+ */
+
+(function() {
+"use strict";
+load("jstests/replsets/rslib.js");
+
+let rst = new ReplSetTest({
+    nodes: {
+        n0: {},
+        n1: {rsConfig: {priority: 0}},
+        n2: {rsConfig: {priority: 0}, setParameter: {enableOverrideClusterChainingSetting: true}}
+    },
+    // Enable the periodic noop writer to aid sync source selection.
+    nodeOptions: {setParameter: {writePeriodicNoops: true}},
+    settings: {chainingAllowed: false},
+    useBridge: true
+});
+rst.startSet();
+rst.initiate();
+
+const primary = rst.getPrimary();
+rst.awaitSecondaryNodes();
+const [n1, n2] = rst.getSecondaries();
+
+// Create a network partition between n2 and the primary.
+n2.disconnect(primary);
+
+// Since n2 has enabled the parameter 'enableOverrideClusterChainingSetting', n2 should eventually
+// chain from n1.
+rst.awaitSyncSource(n2, n1);
+
+// A write with write concern {w:3} should still reach n2.
+var options = {writeConcern: {w: 3, wtimeout: ReplSetTest.kDefaultTimeoutMS}};
+assert.commandWorked(primary.getDB("admin").foo.insert({x: 1}, options));
+
+rst.stopSet();
+}());
\ No newline at end of file
diff --git a/src/mongo/db/repl/SConscript b/src/mongo/db/repl/SConscript
index dadd94cb10d..cda28665b2e 100644
--- a/src/mongo/db/repl/SConscript
+++ b/src/mongo/db/repl/SConscript
@@ -667,6 +667,7 @@ env.Library('topology_coordinator',
     LIBDEPS_PRIVATE=[
         '$BUILD_DIR/mongo/db/catalog/commit_quorum_options',
         '$BUILD_DIR/mongo/idl/server_parameter',
+        'repl_server_parameters'
     ])
 
 env.Library(
diff --git a/src/mongo/db/repl/repl_server_parameters.idl b/src/mongo/db/repl/repl_server_parameters.idl
index befa9891c42..2e06b7101f7 100644
--- a/src/mongo/db/repl/repl_server_parameters.idl
+++ b/src/mongo/db/repl/repl_server_parameters.idl
@@ -307,6 +307,16 @@ server_parameters:
         cpp_varname: assertStableTimestampEqualsAppliedThroughOnRecovery
         default: false
 
+    enableOverrideClusterChainingSetting:
+        description: >-
+            When enabled, allows a node to override the cluster-wide chainingAllowed setting. 
+            If chaining is disabled in the replica set, enabling this parameter allows the node
+            to chain regardless. 
+        set_at: [ startup ]
+        cpp_vartype: AtomicWord<bool>
+        cpp_varname: enableOverrideClusterChainingSetting
+        default: false
+
     startupRecoveryForRestore:
         description: >-
             When set, do startup recovery in such a way that the history of the recovered
diff --git a/src/mongo/db/repl/topology_coordinator.cpp b/src/mongo/db/repl/topology_coordinator.cpp
index ea679f51ba0..6dd1b2d6117 100644
--- a/src/mongo/db/repl/topology_coordinator.cpp
+++ b/src/mongo/db/repl/topology_coordinator.cpp
@@ -55,6 +55,7 @@
 #include "mongo/db/repl/heartbeat_response_action.h"
 #include "mongo/db/repl/isself.h"
 #include "mongo/db/repl/member_data.h"
+#include "mongo/db/repl/repl_server_parameters_gen.h"
 #include "mongo/logv2/log.h"
 #include "mongo/rpc/metadata/oplog_query_metadata.h"
 #include "mongo/rpc/metadata/repl_set_metadata.h"
@@ -213,7 +214,7 @@ HostAndPort TopologyCoordinator::chooseNewSyncSource(Date_t now,
     // If we are only allowed to sync from the primary, use it as the sync source if possible.
     if (readPreference == ReadPreference::PrimaryOnly ||
         (chainingPreference == ChainingPreference::kUseConfiguration &&
-         !_rsConfig.isChainingAllowed())) {
+         !_rsConfig.isChainingAllowed() && !enableOverrideClusterChainingSetting.load())) {
         if (readPreference == ReadPreference::SecondaryOnly) {
             LOGV2_FATAL(
                 3873102,
@@ -2951,10 +2952,11 @@ bool TopologyCoordinator::shouldChangeSyncSource(const HostAndPort& currentSourc
     bool sourceIsPrimary =
         replMetadata.getIsPrimary().value_or(oqMetadata.getPrimaryIndex() == currentSourceIndex);
 
-    // Change sync source if chaining is disabled, we are not syncing from the primary, and we know
-    // who the new primary is. We do not consider chaining disabled if we are the primary, since
-    // we are in catchup mode.
-    auto chainingDisabled = !_rsConfig.isChainingAllowed() && _currentPrimaryIndex != _selfIndex;
+    // Change sync source if chaining is disabled (without overrides), we are not syncing from the
+    // primary, and we know who the new primary is. We do not consider chaining disabled if we are
+    // the primary, since we are in catchup mode.
+    auto chainingDisabled = !_rsConfig.isChainingAllowed() &&
+        !enableOverrideClusterChainingSetting.load() && _currentPrimaryIndex != _selfIndex;
     auto foundNewPrimary = _currentPrimaryIndex != -1 && _currentPrimaryIndex != currentSourceIndex;
     if (!sourceIsPrimary && chainingDisabled && foundNewPrimary) {
         auto newPrimary = _rsConfig.getMemberAt(_currentPrimaryIndex).getHostAndPort();