SERVER-15683: add command to introspect OperationContext and RecoveryUnit

5 files changed, 75 insertions, 22 deletions

diff --git a/src/mongo/db/clientlistplugin.cpp b/src/mongo/db/clientlistplugin.cpp
index 18d7ae04156..a7fd007d28c 100644
--- a/src/mongo/db/clientlistplugin.cpp
+++ b/src/mongo/db/clientlistplugin.cpp
@@ -28,7 +28,9 @@
 
 #include "mongo/platform/basic.h"
 
+#include "mongo/bson/bsonobjbuilder.h"
 #include "mongo/db/client.h"
+#include "mongo/db/commands.h"
 #include "mongo/db/curop.h"
 #include "mongo/db/global_environment_experiment.h"
 #include "mongo/db/operation_context.h"
@@ -116,5 +118,49 @@ namespace {
         }
 
     } clientListPlugin;
+
+    class CommandHelper : public GlobalEnvironmentExperiment::ProcessOperationContext {
+    public:
+        virtual void processOpContext(OperationContext* txn) {
+            BSONObjBuilder b;
+            if ( txn->getClient() )
+                txn->getClient()->reportState( b );
+            if ( txn->getCurOp() )
+                txn->getCurOp()->reportState( &b );
+            if ( txn->lockState() )
+                b.append( "lockState", txn->lockState()->reportState() );
+            if ( txn->recoveryUnit() )
+                txn->recoveryUnit()->reportState( &b );
+            array.append( b.obj() );
+        }
+
+        BSONArrayBuilder array;
+    };
+
+    class CurrentOpContexts : public Command {
+    public:
+        CurrentOpContexts() : Command( "currentOpCtx" ) { }
+
+        virtual bool isWriteCommandForConfigServer() const { return false; }
+
+        virtual bool slaveOk() const { return true; }
+
+        bool run( OperationContext* txn,
+                  const string& dbname,
+                  BSONObj& cmdObj,
+                  int,
+                  string& errmsg,
+                  BSONObjBuilder& result,
+                  bool fromRepl) {
+
+            CommandHelper helper;
+            getGlobalEnvironment()->forEachOperationContext(&helper);
+
+            result.appendArray( "operations", helper.array.arr() );
+
+            return true;
+        }
+    } currentOpContexts;
+
 }  // namespace
 }  // namespace mongo
diff --git a/src/mongo/db/global_environment_d.cpp b/src/mongo/db/global_environment_d.cpp
index 33fe95a6bfe..cae5ef8ccad 100644
--- a/src/mongo/db/global_environment_d.cpp
+++ b/src/mongo/db/global_environment_d.cpp
@@ -176,8 +176,8 @@ namespace mongo {
     void GlobalEnvironmentMongoD::forEachOperationContext(ProcessOperationContext* procOpCtx) {
         scoped_lock lock(_registeredOpContextsMutex);
 
-        OperationContextSet::iterator it;
-        for (it = _registeredOpContexts.begin(); it != _registeredOpContexts.end(); it++) {
+        OperationContextSet::const_iterator it;
+        for (it = _registeredOpContexts.begin(); it != _registeredOpContexts.end(); ++it) {
             procOpCtx->processOpContext(*it);
         }
     }
diff --git a/src/mongo/db/operation_context_impl.cpp b/src/mongo/db/operation_context_impl.cpp
index 91858ff64d7..47310d3b0f5 100644
--- a/src/mongo/db/operation_context_impl.cpp
+++ b/src/mongo/db/operation_context_impl.cpp
@@ -57,6 +57,8 @@ namespace mongo {
             _locker.reset(new LockerImpl<true>());
         }
 
+        _client = &cc();
+
         getGlobalEnvironment()->registerOperationContext(this);
     }
 
@@ -92,15 +94,15 @@ namespace mongo {
     }
 
     bool OperationContextImpl::isGod() const {
-        return cc().isGod();
+        return getClient()->isGod();
     }
 
     Client* OperationContextImpl::getClient() const {
-        return &cc();
+        return _client;
     }
 
     CurOp* OperationContextImpl::getCurOp() const {
-        return cc().curop();
+        return getClient()->curop();
     }
 
     unsigned int OperationContextImpl::getOpID() const {
@@ -153,9 +155,9 @@ namespace mongo {
     } // namespace
 
     void OperationContextImpl::checkForInterrupt(bool heedMutex) const {
-        Client& c = cc();
+        Client* c = getClient();
 
-        if (heedMutex && lockState()->isWriteLocked() && c.hasWrittenSinceCheckpoint()) {
+        if (heedMutex && lockState()->isWriteLocked() && c->hasWrittenSinceCheckpoint()) {
             return;
         }
 
@@ -163,46 +165,46 @@ namespace mongo {
                 "interrupted at shutdown",
                 !getGlobalEnvironment()->getKillAllOperations());
 
-        if (c.curop()->maxTimeHasExpired()) {
-            c.curop()->kill();
+        if (c->curop()->maxTimeHasExpired()) {
+            c->curop()->kill();
             uasserted(ErrorCodes::ExceededTimeLimit, "operation exceeded time limit");
         }
 
         MONGO_FAIL_POINT_BLOCK(checkForInterruptFail, scopedFailPoint) {
-            if (opShouldFail(c, scopedFailPoint.getData())) {
-                log() << "set pending kill on " << (c.curop()->parent() ? "nested" : "top-level")
-                      << " op " << c.curop()->opNum() << ", for checkForInterruptFail";
-                c.curop()->kill();
+            if (opShouldFail(*c, scopedFailPoint.getData())) {
+                log() << "set pending kill on " << (c->curop()->parent() ? "nested" : "top-level")
+                      << " op " << c->curop()->opNum() << ", for checkForInterruptFail";
+                c->curop()->kill();
             }
         }
 
-        if (c.curop()->killPending()) {
+        if (c->curop()->killPending()) {
             uasserted(ErrorCodes::Interrupted, "operation was interrupted");
         }
     }
 
     Status OperationContextImpl::checkForInterruptNoAssert() const {
         // TODO(spencer): Unify error codes and implementation with checkForInterrupt()
-        Client& c = cc();
+        Client* c = getClient();
 
         if (getGlobalEnvironment()->getKillAllOperations()) {
             return Status(ErrorCodes::Interrupted, "interrupted at shutdown");
         }
 
-        if (c.curop()->maxTimeHasExpired()) {
-            c.curop()->kill();
+        if (c->curop()->maxTimeHasExpired()) {
+            c->curop()->kill();
             return Status(ErrorCodes::Interrupted, "exceeded time limit");
         }
 
         MONGO_FAIL_POINT_BLOCK(checkForInterruptFail, scopedFailPoint) {
-            if (opShouldFail(c, scopedFailPoint.getData())) {
-                log() << "set pending kill on " << (c.curop()->parent() ? "nested" : "top-level")
-                      << " op " << c.curop()->opNum() << ", for checkForInterruptFail";
-                c.curop()->kill();
+            if (opShouldFail(*c, scopedFailPoint.getData())) {
+                log() << "set pending kill on " << (c->curop()->parent() ? "nested" : "top-level")
+                      << " op " << c->curop()->opNum() << ", for checkForInterruptFail";
+                c->curop()->kill();
             }
         }
 
-        if (c.curop()->killPending()) {
+        if (c->curop()->killPending()) {
             return Status(ErrorCodes::Interrupted, "interrupted");
         }
 
diff --git a/src/mongo/db/operation_context_impl.h b/src/mongo/db/operation_context_impl.h
index b8ae53a2185..616380e174f 100644
--- a/src/mongo/db/operation_context_impl.h
+++ b/src/mongo/db/operation_context_impl.h
@@ -73,6 +73,7 @@ namespace mongo {
     private:
         std::auto_ptr<RecoveryUnit> _recovery;
         std::auto_ptr<Locker> _locker;
+        Client* _client; // cached, not owned
     };
 
 }  // namespace mongo
diff --git a/src/mongo/db/storage/recovery_unit.h b/src/mongo/db/storage/recovery_unit.h
index 4657a85c7f6..39fb62b47cd 100644
--- a/src/mongo/db/storage/recovery_unit.h
+++ b/src/mongo/db/storage/recovery_unit.h
@@ -35,6 +35,8 @@
 
 namespace mongo {
 
+    class BSONObjBuilder;
+
     /**
      * A RecoveryUnit is responsible for ensuring that data is persisted.
      * All on-disk information must be mutated through this interface.
@@ -44,6 +46,8 @@ namespace mongo {
     public:
         virtual ~RecoveryUnit() { }
 
+        virtual void reportState( BSONObjBuilder* b ) const { }
+
         /**
          * These should be called through WriteUnitOfWork rather than directly.
          *