SERVER-46958 add mongosShutdownTimeoutMillisForSignaledShutdown server parameter

6 files changed, 48 insertions, 18 deletions

diff --git a/src/mongo/s/mongos_server_parameters.idl b/src/mongo/s/mongos_server_parameters.idl
index bb659f7cf93..949b4d6c627 100644
--- a/src/mongo/s/mongos_server_parameters.idl
+++ b/src/mongo/s/mongos_server_parameters.idl
@@ -86,3 +86,13 @@ server_parameters:
     cpp_vartype: bool
     cpp_varname: "gEnableFinerGrainedCatalogCacheRefresh"
     default: true
+
+  mongosShutdownTimeoutMillisForSignaledShutdown:
+    description: >-
+        The time taken for quiesce mode at shutdown in response to SIGTERM.
+    set_at: [ startup, runtime ]
+    cpp_vartype: AtomicWord<int>
+    cpp_varname: "mongosShutdownTimeoutMillisForSignaledShutdown"
+    default: 15000
+    validator:
+        gte: 0
\ No newline at end of file
diff --git a/src/mongo/s/mongos_topology_coordinator.cpp b/src/mongo/s/mongos_topology_coordinator.cpp
index 4f889b65b83..1d18c96a6d3 100644
--- a/src/mongo/s/mongos_topology_coordinator.cpp
+++ b/src/mongo/s/mongos_topology_coordinator.cpp
@@ -192,7 +192,8 @@ void MongosTopologyCoordinator::enterQuiesceMode() {
     IsMasterMetrics::get(getGlobalServiceContext())->resetNumAwaitingTopologyChanges();
 }
 
-void MongosTopologyCoordinator::enterQuiesceModeAndWait(OperationContext* opCtx) {
+void MongosTopologyCoordinator::enterQuiesceModeAndWait(OperationContext* opCtx,
+                                                        Milliseconds quiesceTime) {
     enterQuiesceMode();
 
     if (MONGO_unlikely(hangDuringQuiesceMode.shouldFail())) {
@@ -200,11 +201,8 @@ void MongosTopologyCoordinator::enterQuiesceModeAndWait(OperationContext* opCtx)
         hangDuringQuiesceMode.pauseWhileSet(opCtx);
     }
 
-    // TODO SERVER-46958: Determine what the quiesce time should be by checking the
-    // shutdownTimeoutMillisForSignaledShutdown mongos server parameter.
-    auto timeout = Milliseconds(100);
-    LOGV2(4695701, "Entering quiesce mode for mongos shutdown", "quiesceTime"_attr = timeout);
-    opCtx->sleepFor(timeout);
+    LOGV2(4695701, "Entering quiesce mode for mongos shutdown", "quiesceTime"_attr = quiesceTime);
+    opCtx->sleepFor(quiesceTime);
     LOGV2(4695702, "Exiting quiesce mode for mongos shutdown");
 }
 
diff --git a/src/mongo/s/mongos_topology_coordinator.h b/src/mongo/s/mongos_topology_coordinator.h
index 708de6a1241..eccf83c8a31 100644
--- a/src/mongo/s/mongos_topology_coordinator.h
+++ b/src/mongo/s/mongos_topology_coordinator.h
@@ -70,12 +70,11 @@ public:
     void enterQuiesceMode();
 
     /**
-     * While in quiesce mode, we will sleep for 100ms. This allows short running operations to
-     * continue. We will also accept new operations, but we fail isMaster requests with
+     * While in quiesce mode, we will sleep for quiesceTime. This allows short running operations
+     * to continue. We will also accept new operations, but we fail isMaster requests with
      * ShutdownInProgress.
-     * TODO SERVER-46958: Modify comment with correct timeout value.
      */
-    void enterQuiesceModeAndWait(OperationContext* opCtx);
+    void enterQuiesceModeAndWait(OperationContext* opCtx, Milliseconds quiesceTime);
 
     TopologyVersion getTopologyVersion() const {
         stdx::lock_guard lk(_mutex);
diff --git a/src/mongo/s/server.cpp b/src/mongo/s/server.cpp
index 5f0589aded1..05d228e5ba3 100644
--- a/src/mongo/s/server.cpp
+++ b/src/mongo/s/server.cpp
@@ -85,6 +85,7 @@
 #include "mongo/s/grid.h"
 #include "mongo/s/is_mongos.h"
 #include "mongo/s/mongos_options.h"
+#include "mongo/s/mongos_server_parameters_gen.h"
 #include "mongo/s/mongos_topology_coordinator.h"
 #include "mongo/s/query/cluster_cursor_cleanup_job.h"
 #include "mongo/s/query/cluster_cursor_manager.h"
@@ -254,7 +255,8 @@ void implicitlyAbortAllTransactions(OperationContext* opCtx) {
  * NOTE: This function may be called at any time after registerShutdownTask is called below. It must
  * not depend on the prior execution of mongo initializers or the existence of threads.
  */
-void cleanupTask(ServiceContext* serviceContext) {
+void cleanupTask(const ShutdownTaskArgs& shutdownArgs) {
+    const auto serviceContext = getGlobalServiceContext();
     {
         // This client initiation pattern is only to be used here, with plans to eliminate this
         // pattern down the line.
@@ -269,11 +271,20 @@ void cleanupTask(ServiceContext* serviceContext) {
             opCtx = uniqueTxn.get();
         }
 
+        Milliseconds quiesceTime;
+        if (shutdownArgs.quiesceTime) {
+            quiesceTime = *shutdownArgs.quiesceTime;
+        } else {
+            // IDL gaurantees that quiesceTime is populated.
+            invariant(!shutdownArgs.isUserInitiated);
+            quiesceTime = Milliseconds(mongosShutdownTimeoutMillisForSignaledShutdown.load());
+        }
+
         // Enter quiesce mode so that existing and new short operations are allowed to finish.
         // At this point, we will start responding to any isMaster request with ShutdownInProgress
         // so that clients can re-route their operations.
         if (auto mongosTopCoord = MongosTopologyCoordinator::get(opCtx)) {
-            mongosTopCoord->enterQuiesceModeAndWait(opCtx);
+            mongosTopCoord->enterQuiesceModeAndWait(opCtx, quiesceTime);
         }
 
         // Shutdown the TransportLayer so that new connections aren't accepted
@@ -928,9 +939,9 @@ ExitCode mongoSMain(int argc, char* argv[], char** envp) {
         return EXIT_ABRUPT;
     }
 
-    const auto service = getGlobalServiceContext();
+    registerShutdownTask(cleanupTask);
 
-    registerShutdownTask([service]() { cleanupTask(service); });
+    const auto service = getGlobalServiceContext();
 
     ErrorExtraInfo::invariantHaveAllParsers();
 
diff --git a/src/mongo/shell/servers.js b/src/mongo/shell/servers.js
index c5cf428d778..94638d450c7 100644
--- a/src/mongo/shell/servers.js
+++ b/src/mongo/shell/servers.js
@@ -626,14 +626,17 @@ var _isMongodVersionEqualOrAfter = function(version1, version2) {
     return false;
 };
 
-// Removes a setParameter parameter from mongods running a version that won't recognize them.
-var _removeSetParameterIfBeforeVersion = function(opts, parameterName, requiredVersion) {
+// Removes a setParameter parameter from mongods or mongoses running a version that won't recognize
+// them.
+var _removeSetParameterIfBeforeVersion = function(
+    opts, parameterName, requiredVersion, isMongos = false) {
+    var processString = isMongos ? "mongos" : "mongod";
     var versionCompatible = (opts.binVersion === "" || opts.binVersion === undefined ||
                              _isMongodVersionEqualOrAfter(requiredVersion, opts.binVersion));
     if (!versionCompatible && opts.setParameter && opts.setParameter[parameterName] != undefined) {
         print("Removing '" + parameterName + "' setParameter with value " +
-              opts.setParameter[parameterName] +
-              " because it isn't compatibile with mongod running version " + opts.binVersion);
+              opts.setParameter[parameterName] + " because it isn't compatible with " +
+              processString + " running version " + opts.binVersion);
         delete opts.setParameter[parameterName];
     }
 };
@@ -793,6 +796,9 @@ MongoRunner.mongosOptions = function(opts) {
         opts.binVersion = MongoRunner.getBinVersionFor(testOptions.mongosBinVersion);
     }
 
+    _removeSetParameterIfBeforeVersion(
+        opts, "mongosShutdownTimeoutMillisForSignaledShutdown", "4.5.0", true);
+
     // If the mongos is being restarted with a newer version, make sure we remove any options
     // that no longer exist in the newer version.
     if (opts.restart && MongoRunner.areBinVersionsTheSame('latest', opts.binVersion)) {
diff --git a/src/mongo/shell/shardingtest.js b/src/mongo/shell/shardingtest.js
index af86706a6aa..fa3bf67b546 100644
--- a/src/mongo/shell/shardingtest.js
+++ b/src/mongo/shell/shardingtest.js
@@ -1544,6 +1544,12 @@ var ShardingTest = function(params) {
         options = Object.merge(options, otherParams.mongosOptions);
         options = Object.merge(options, otherParams["s" + i]);
 
+        // The default time for mongos quiesce mode in response to SIGTERM is 15 seconds.
+        // Reduce this to 100ms for faster shutdown.
+        options.setParameter = options.setParameter || {};
+        options.setParameter.mongosShutdownTimeoutMillisForSignaledShutdown =
+            options.setParameter.mongosShutdownTimeoutMillisForSignaledShutdown || 100;
+
         options.port = options.port || _allocatePortForMongos();
 
         mongosOptions.push(options);