2025-11-28 00:35:46 +09:00

553 lines
16 KiB
C++

// --------------------------------------------------------------------
//
// Copyright (c) Microsoft Corporation. All rights reserved
//
// THIS CODE AND INFORMATION IS PROVIDED "AS IS" WITHOUT WARRANTY OF
// ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING BUT NOT LIMITED TO
// THE IMPLIED WARRANTIES OF MERCHANTABILITY AND/OR FITNESS FOR A
// PARTICULAR PURPOSE.
//
// --------------------------------------------------------------------
//
// Purpose:
// This console application sample demonstrates how to use NT
// completion ports in MSMQ to asynchronously receive messages in an
// efficient manner. This mechanism is scalable in the number of
// queues and messages by adding more processors and threads.
// Similarly, generic completion port handlers can be provided to
// handle other NT resources as well as queues.
//
// Both the MSMQ COM components and the MQ API are used in this program.
// Note that the COM components are used for creating a queue, open
// the queue, and sending messages. Conversely, the MQ API is used
// to implement completion port-based asynchronous receive.
//
// Requirements:
// VC5 is required.
// MSMQ must be installed. Specifically mqoa.dll must be registered
// and on the path.
// Project settings:
// - The include path must include the location of mq.h:
// e.g. ...\msmq\sdk\include
// - The link library path must include the location of mqrt.lib:
// e.g ...\msmq\sdk\lib
//
// Overview:
// The following steps comprise this sample:
// - A global MSMQQueueInfo object is used to reference the sample's
// single queue.
// - Initialize OLE.
// - Create a new completion port.
// - Create a bunch of threads with a generic CompletionPortThread start
// routine parameterized with the completion port handle from the previous step.
// - Open the queue and associate its handle with the completion port.
// - Note that the queue is deleted and recreated if already exists;
// otherwise, a new queue is created.
// - Enable a bunch of asynchronous message receive requests on the queue.
// Since the queue is associated with the completion port, each of these
// requests results in the CompletionPortThread handler being notified
// asynchronously by NT when an async receive message operation "completes."
// - Note that the NT scheduler will select the "best" available completion
// port thread that is synchronously waiting for a completion notification.
// - Finally, to test the completion port handlers, a bunch of messages is sent
// to the queue and the program hibernates.
// - To exit, just kill the console application window.
//
// Warning: *** Only limited error checking and handling are provided. ***
//
#include <windows.h>
#include <stdio.h>
#import <mqoa.tlb> no_namespace
#include <mq.h>
//
// Global queue object.
//
IMSMQQueueInfoPtr g_qinfo;
//
// Structure containing both an OVERLAPPED structure and other MSMQ
// application-specific data.
//
struct RECEIVE_CONTEXT
{
OVERLAPPED ov;
HANDLE hQueue;
MQMSGPROPS *pmsgprops;
};
//
// AllocMsgProps
// Parameters:
// prc IN receive context containing an OVERLAPPED structure and
// additional MSMQ-specific information: queue handle and
// message properties.
//
// Purpose:
// Allocates a property identifier array for the PROPID_M_APPSPECIFIC property.
// This is just because we're lazy and don't want to allocate/deallocate
// buffers for the message body, label, etc. The PROPID_M_APPSPECIFIC
// property allows us to "uniquely" stamp each messsages for identification
// purposes (in this case, with an ordinal number).
//
void AllocMsgProps(RECEIVE_CONTEXT *prc)
{
int cProp = 1;
prc->pmsgprops->aPropID = new MSGPROPID[cProp];
prc->pmsgprops->aPropVar = new MQPROPVARIANT[cProp];
prc->pmsgprops->aStatus = new HRESULT[cProp];
int iProp = 0;
prc->pmsgprops->aPropID[iProp] = PROPID_M_APPSPECIFIC;
prc->pmsgprops->aPropVar[iProp].vt = VT_UI4;
prc->pmsgprops->cProp = cProp;
}
//
// HandleReceivedMessage:
// Parameters:
// prc IN receive context containing an OVERLAPPED structure and
// additional MSMQ-specific information: queue handle and
// message properties.
//
// Purpose:
// Inspects the value of HRESULT returned by the MSMQ device driver
// in the OVERLAPPED structure. Provides more detailed last-error
// information. Then displays the message -- in this case just the
// message's PROPID_M_APPSPECIFIC property.
//
void HandleReceivedMessage(RECEIVE_CONTEXT* prc)
{
//
// Get receive message final status.
//
HRESULT rc = MQGetOverlappedResult(&prc->ov);
//
// Handle the status and message.
//
if (SUCCEEDED(rc))
{
//
// Get the received message. PROPID_M_ APPSPECIFIC is the single property
// that we set and now retrieve.
//
long lAppSpecific = prc->pmsgprops->aPropVar[0].lVal;
printf("Thread id: %x Message received with the app-specific data: %d\n",
GetCurrentThreadId(),
lAppSpecific);
}
}
//
// HandleErrors
// Parameters:
// _com_error
//
// Purpose:
// Displays an error and aborts further execution.
//
void HandleErrors(_com_error comerr)
{
HRESULT hr = comerr.Error();
printf("An error occurred. Error code: %x\nExiting...", hr);
exit(hr);
};
//
// EnableAsyncReceive:
// Parameters:
// prc IN receive context containing an OVERLAPPED structure and
// additional MSMQ-specific information: queue handle and
// message properties.
//
// Purpose:
// Makes an MSMQ asynchronous receive request, specifying
// an OVERLAPPED structure with the appropriate queue, which
// has been associated already with a completion port.
//
HRESULT EnableAsyncReceive(RECEIVE_CONTEXT* prc)
{
//
// Re-enable asynchronous receiving.
//
return MQReceiveMessage(
prc->hQueue,
INFINITE, // Time-out interval
MQ_ACTION_RECEIVE,
prc->pmsgprops,
&prc->ov, // OVERLAPPED
NULL, // No callback function
NULL, // No cursor
NULL); // No transaction pointer
}
//
// CompletionPortThread:
// Parameters:
// lParam IN completion port handle.
//
// Purpose:
// Start routine for each worker thread.
// Waits for the completion port to complete. When notification arrives,
// handles the received message and re-enables MSMQ asynchronous receiving.
//
DWORD WINAPI CompletionPortThread(LPVOID lParam)
{
HANDLE hPort = (HANDLE)lParam;
HRESULT hr = NOERROR;
for (;;)
{
//
// Wait for completion notification.
//
DWORD dwNoOfBytes;
ULONG_PTR dwKey;
OVERLAPPED* pov = NULL;
BOOL fSuccess = GetQueuedCompletionStatus(
hPort,
&dwNoOfBytes,
&dwKey,
&pov,
INFINITE // Notification time-out interval
);
//
// A NULL value of pov is returned if completion port notification
// failed. In this case, fSuccess is guaranteed to be FALSE.
// When fSuccess is TRUE, the OVERLAPPED structure may still
// contain an error code. It is inspected in HandleReceivedMessage.
//
if (pov == NULL)
{
//
// Unrecoverable error occurred in the completion port. Wait for the next notification.
//
continue;
}
RECEIVE_CONTEXT* prc = CONTAINING_RECORD(pov, RECEIVE_CONTEXT, ov);
HandleReceivedMessage(prc);
//
// Start the next message receive operation.
//
hr = EnableAsyncReceive(prc);
if (FAILED(hr))
return hr;
}
//
// The queue is unreachable.
//
return 0;
}
//
// CreateWorkingThreads
// Parameters:
// hPort IN completion port handle.
//
// Purpose:
// Creates a certain number of worker threads whose start routine
// is CompletionPortThread, associating each of them with
// the incoming port handle.
//
void CreateWorkingThreads(HANDLE hPort)
{
//
// Start several threads to handle the completion port.
// The number of threads that you create depends on number of processors
// in the system and the expected serialization in the working thread logic.
//
const int xNumberOfProcessors = 1;
const int xNumberOfThreads = 2 * xNumberOfProcessors;
for (int i = 0; i < xNumberOfThreads; i++)
{
DWORD dwThreadId;
HANDLE hThread = CreateThread(
NULL, // Thread security
0, // Default stack
CompletionPortThread, // Start routine
hPort, // Start routine parameter
0, // Run immediately
&dwThreadId // Thread identifier
);
CloseHandle(hThread);
}
}
//
// OpenQueueForAsyncReceiveWithCompletionPort
// Parameters:
// hPort IN completion port handle
//
// Purpose:
// Deletes and recreates the existing queue if already exists;
// otherwise, creates a new queue which is then opened for receive.
// Finally, the newly opened queue handle is associated with the incoming
// completion port.
//
IMSMQQueuePtr OpenQueueForAsyncReceiveWithCompletionPort(
HANDLE hPort)
{
IMSMQQueuePtr qRec;
//
// Delete the existing queue, ignoring errors.
//
try {
g_qinfo->Delete();
}
catch(_com_error comerr) {
};
//
// Recreate the queue.
//
g_qinfo->Create();
//
// Open the queue to receive messages from it.
//
qRec = g_qinfo->Open(MQ_RECEIVE_ACCESS, MQ_DENY_NONE);
//
// Associate the queue with the completion port.
//
CreateIoCompletionPort(
(HANDLE)qRec->Handle, // Queue to associate port with
hPort, // Port handle
0,
0);
return qRec;
}
//
// SendSomeMessages
// Parameters:
// cMsgs IN number of messages to send.
//
// Purpose:
// Sends a bunch of messages to the queue.
//
void SendSomeMessages(int cMsgs)
{
IMSMQQueuePtr qSend;
IMSMQMessagePtr msg("MSMQ.MSMQMessage");
//
// Open the queue for sending messages.
//
qSend = g_qinfo->Open(MQ_SEND_ACCESS, MQ_DENY_NONE);
//
// Send a bunch of messages.
//
for (int i = 0; i < cMsgs; i++)
{
msg->AppSpecific = i;
msg->Send(qSend);
}
}
//
// InitiateAsyncReceiveWithCompletionPort
// Parameters:
// q IN queue object to enable asynchronous messaging on
//
// Purpose:
// Creates several (a bunch!) MSMQ asynchronous message receive requests. Each
// one is completion-port based.
//
HRESULT InitiateAsyncReceiveWithCompletionPort(IMSMQQueuePtr q)
{
HRESULT hr = NOERROR;
//
// Kick off several overlapped receives.
//
const DWORD cOverlappedReceives = 1;
for (int i = 0; i < cOverlappedReceives; i++)
{
//
// Allocate and set a receive context.
//
RECEIVE_CONTEXT* prc = new RECEIVE_CONTEXT;
memset(prc, 0, sizeof(RECEIVE_CONTEXT));
prc->hQueue = (HANDLE)q->Handle;
prc->pmsgprops = new MQMSGPROPS;
AllocMsgProps(prc);
hr = EnableAsyncReceive(prc);
if (FAILED(hr)) {
return hr;
}
}
return hr;
}
//
// Check whether the local computer has access to the directory service (DS).
//
short DetectDsConnection(void)
{
HRESULT hresult;
IMSMQApplication2Ptr pApp;
hresult = CoInitialize(NULL);
if (FAILED(hresult))
{
printf("\nCOM cannot be initialized. Error code: %d\n", hresult);
exit(1);
}
hresult = pApp.CreateInstance(__uuidof(MSMQApplication));
if (FAILED(hresult))
{
printf("\nThe MSMQApplication object cannot be created. Error code: %d\n", hresult);
exit(1);
}
return pApp->GetIsDsEnabled();
}
//
// MAIN
//
int main()
{
HRESULT hr;
char strPathName[522];
char strComputerName[256];
char strQueueName[256];
int dNoOfMessages;
//
// Initialize OLE.
//
hr = OleInitialize(NULL);
if (FAILED(hr))
return hr;
try
{
g_qinfo = IMSMQQueueInfoPtr("MSMQ.MSMQQueueInfo");
IMSMQQueuePtr qRec;
//
// Create a new completion port.
//
HANDLE hPort = CreateIoCompletionPort(
INVALID_HANDLE_VALUE, // Create a new port.
NULL, // Create a new port.
0,
0);
//
// Create some worker threads to handle asynchronous receive operations.
//
CreateWorkingThreads(hPort);
//
// Get a queue path name.
//
printf("\nEnter queue computer name: ");
if (fgets(strComputerName, 255, stdin) == NULL)
{
printf("\nInvalid input was entered.\n");
exit(1);
}
//
// Remove the \n at the end of the string created by fgets.
//
if(strComputerName[strlen(strComputerName) - 1] == '\n')
{
strComputerName[strlen(strComputerName) - 1] = '\0';
}
printf("\nEnter queue name: ");
if (fgets(strQueueName, 255, stdin) == NULL)
{
printf("\nInvalid input was entered.\n");
exit(1);
}
if(strQueueName[strlen(strQueueName) - 1] == '\n')
{
strQueueName[strlen(strQueueName) - 1] = '\0';
}
if(strComputerName[0] == 0 || strQueueName[0] == 0)
{
printf("\nInvalid parameters were supplied. Exiting...\n");
exit(1);
}
//
// Open the queue and associate it with the completion port.
//
int nRes, nSize;
nSize = (sizeof(strComputerName)/sizeof(strComputerName[0])) + (sizeof(strQueueName)/sizeof(strQueueName[0])) -1;
if(DetectDsConnection())
//
// Access to the DS is enabled, so work with a public queue.
//
{
nRes = _snprintf_s(strPathName, sizeof(strPathName), nSize+1, "%s\\%s", strComputerName, strQueueName);
}
else
//
// Access to the DS is disabled, so work with a private queue.
//
{
nRes = _snprintf_s(strPathName, sizeof(strPathName), nSize+10, "%s\\private$\\%s", strComputerName, strQueueName);
}
if(nRes < 0)
{
printf("The path name is too long for the buffer specified.\n");
exit(1);
}
strPathName[nRes] = '\0';
g_qinfo->PathName = strPathName;
qRec = OpenQueueForAsyncReceiveWithCompletionPort(hPort);
//
// Invoke MSMQ to start an asynchronous receive operation on the queue several times.
//
hr = InitiateAsyncReceiveWithCompletionPort(qRec);
if (FAILED(hr)) {
return hr;
}
//
// Send some messages.
//
printf("\nEnter the number of messages to send: ");
int iRes = scanf_s("%d", &dNoOfMessages);
if (iRes == 0 || iRes == EOF)
{
printf("\nInvalid input was entered.\n");
exit(1);
}
SendSomeMessages(dNoOfMessages);
//
// Wait forever. Kill the application manually to exit.
//
printf("\nKill the application manually to exit.\n");
Sleep(INFINITE);
}
catch(_com_error comerr) {
HandleErrors(comerr);
};
return 0;
}