router & dealer
这对模型是zmq最重要的模型，由这对模型也衍生出一些其他模型，比如接下来要说到的req-rep模型。再分析router-dealer之前先看一下zmq的两个队列 : fq和 lb。前者是一个用于接收消息的fair queueing队列，后者是一个用于发送消息的load balance队列。两者都使用了一个array_t来作为自己的队列数据结构，array_t的实现比较简单，主要是数组的扩展，每个元素用一个index来标记自己在数组中的位置，这样可以快速标记一个元素在数组中的位置，方便操作，用空间换取时间。下面是fq的一段主要函数：
int zmq::fq_t::recvpipe (msg_t *msg_, pipe_t **pipe_)
{
// Deallocate old content of the message.
int rc = msg_->close ();
errno_assert (rc == 0);
// Round-robin over the pipes to get the next message.
while (active > 0) {
// Try to fetch new message. If we've already read part of the message
// subsequent part should be immediately available.
bool fetched = pipes [current]->read (msg_);
// Note that when message is not fetched, current pipe is deactivated
// and replaced by another active pipe. Thus we don't have to increase
// the 'current' pointer.
if (fetched) {
if (pipe_)
*pipe_ = pipes [current];
more = msg_->flags () & msg_t::more? true: false;
if (!more) {
last_in = pipes [current];
current = (current + 1) % active;
}
return 0;
}
// Check the atomicity of the message.
// If we've already received the first part of the message
// we should get the remaining parts without blocking.
zmq_assert (!more);
active--;
pipes.swap (current, active);
if (current == active)
current = 0;
}
// No message is available. Initialise the output parameter
// to be a 0-byte message.
rc = msg_->init ();
errno_assert (rc == 0);
errno = EAGAIN;
return -1;
}
fq把所有active状态的pipe放在数组的前面，当有新的pipe或者是是已有pipe转入非active状态时都通过array_t的swap来进行快速移动。另外fq用一个游标记录当前正在读取的pipe，每次从游标标记的pipe中读取数据，当完整读取一条数据时（msg的flag没有more标记），移动游标，以此达到fair queueing的目标。lb的实现与fq类似。fq和lb必须要完整处理一条数据之后才能移动到下一个位置，即使其他pipe有数据也需要等待。说完fq和lb的实现，接下来看router和dealer是怎么使用他们的。
router类使用fq管理pipe，router的recv方法会调用fq的recvpipe来达到fair queueing的目的。router的send方法则需要指定管道，实现方式是强制第一条发送的msg必须是identity类型的msg以便指定对应的管道。
下面是router类的成员变量:
// Receive peer id and update lookup map
bool identify_peer (pipe_t *pipe_);
// Fair queueing object for inbound pipes.
fq_t fq;
// True iff there is a message held in the pre-fetch buffer.
bool prefetched;
// If true, the receiver got the message part with
// the peer's identity.
bool identity_sent;
// Holds the prefetched identity.
msg_t prefetched_id;
// Holds the prefetched message.
msg_t prefetched_msg;
// If true, more incoming message parts are expected.
bool more_in;
struct outpipe_t
{
zmq::pipe_t *pipe;
bool active;
};
// We keep a set of pipes that have not been identified yet.
std::set <pipe_t*> anonymous_pipes;
// Outbound pipes indexed by the peer IDs.
typedef std::map <blob_t, outpipe_t> outpipes_t;
outpipes_t outpipes;
// The pipe we are currently writing to.
zmq::pipe_t *current_out;
// If true, more outgoing message parts are expected.
bool more_out;
// Routing IDs are generated. It's a simple increment and wrap-over
// algorithm. This value is the next ID to use (if not used already).
uint32_t next_rid;
// If true, report EAGAIN to the caller instead of silently dropping
// the message targeting an unknown peer.
bool mandatory;
bool raw_sock;
// if true, send an empty message to every connected router peer
bool probe_router;
// If true, the router will reassign an identity upon encountering a
// name collision. The new pipe will take the identity, the old pipe
// will be terminated.
bool handover;其中prefetched_id和prefetched_msg分别用于临时存储identity和真正的数据。anonymous_pipes用来存储没有验证identity的匿名管道，outpipes存储已经验证过identity的管道。router的其他的变量都是一些状态记录变量。比如current_out是代表当前正在写入数据的管道，more_out代表一条完整的数据是否发送完。
int zmq::router_t::xrecv (msg_t *msg_)
{
if (prefetched) {
if (!identity_sent) {
int rc = msg_->move (prefetched_id);
errno_assert (rc == 0);
identity_sent = true;
}
else {
int rc = msg_->move (prefetched_msg);
errno_assert (rc == 0);
prefetched = false;
}
more_in = msg_->flags () & msg_t::more ? true : false;
return 0;
}
pipe_t *pipe = NULL;
int rc = fq.recvpipe (msg_, &pipe);
// It's possible that we receive peer's identity. That happens
// after reconnection. The current implementation assumes that
// the peer always uses the same identity.
while (rc == 0 && msg_->is_identity ())
rc = fq.recvpipe (msg_, &pipe);
if (rc != 0)
return -1;
zmq_assert (pipe != NULL);
// If we are in the middle of reading a message, just return the next part.
if (more_in)
more_in = msg_->flags () & msg_t::more ? true : false;
else {
// We are at the beginning of a message.
// Keep the message part we have in the prefetch buffer
// and return the ID of the peer instead.
rc = prefetched_msg.move (*msg_);
errno_assert (rc == 0);
prefetched = true;
blob_t identity = pipe->get_identity ();
rc = msg_->init_size (identity.size ());
errno_assert (rc == 0);
memcpy (msg_->data (), identity.data (), identity.size ());
msg_->set_flags (msg_t::more);
identity_sent = true;
}
return 0;
}当调用xrecv方法获取一条新的数据时，router会将这条数据缓存起来放入到prefetched_msg中，然后向上层返回一条identity类型的msg来标记对应的管道，下次再调用xrecv时才会返回真正的数据。如果more_in为真，表明已经向上层返回过identity和缓存的prefetched_msg，所以可以直接把该条msg返回给上层。
int zmq::router_t::xsend (msg_t *msg_)
{
// If this is the first part of the message it's the ID of the
// peer to send the message to.
if (!more_out) {
zmq_assert (!current_out);
// If we have malformed message (prefix with no subsequent message)
// then just silently ignore it.
// TODO: The connections should be killed instead.
if (msg_->flags () & msg_t::more) {
more_out = true;
// Find the pipe associated with the identity stored in the prefix.
// If there's no such pipe just silently ignore the message, unless
// router_mandatory is set.
blob_t identity ((unsigned char*) msg_->data (), msg_->size ());
outpipes_t::iterator it = outpipes.find (identity);
if (it != outpipes.end ()) {
current_out = it->second.pipe;
if (!current_out->check_write ()) {
it->second.active = false;
current_out = NULL;
if (mandatory) {
more_out = false;
errno = EAGAIN;
return -1;
}
}
}
else
if (mandatory) {
more_out = false;
errno = EHOSTUNREACH;
return -1;
}
}
int rc = msg_->close ();
errno_assert (rc == 0);
rc = msg_->init ();
errno_assert (rc == 0);
return 0;
}
// Ignore the MORE flag for raw-sock or assert?
if (options.raw_sock)
msg_->reset_flags (msg_t::more);
// Check whether this is the last part of the message.
more_out = msg_->flags () & msg_t::more ? true : false;
// Push the message into the pipe. If there's no out pipe, just drop it.
if (current_out) {
// Close the remote connection if user has asked to do so
// by sending zero length message.
// Pending messages in the pipe will be dropped (on receiving term- ack)
if (raw_sock && msg_->size() == 0) {
current_out->terminate (false);
int rc = msg_->close ();
errno_assert (rc == 0);
rc = msg_->init ();
errno_assert (rc == 0);
current_out = NULL;
return 0;
}
bool ok = current_out->write (msg_);
if (unlikely (!ok)) {
// Message failed to send - we must close it ourselves.
int rc = msg_->close ();
errno_assert (rc == 0);
current_out = NULL;
} else {
if (!more_out) {
current_out->flush ();
current_out = NULL;
}
}
}
else {
int rc = msg_->close ();
errno_assert (rc == 0);
}
// Detach the message from the data buffer.
int rc = msg_->init ();
errno_assert (rc == 0);
return 0;
}通过xsend发送的消息必须首先指定identity， 之后current_out会记录当前正在写数据的管道，之后该条数据的其他部分的msg都需要向current_out中写入。
与router相比，dealer的实现比较简单，dealer只负责消息的转发，下面是dealer的成员变量：
private:
// Messages are fair-queued from inbound pipes. And load-balanced to
// the outbound pipes.
fq_t fq;
lb_t lb;
// if true, send an empty message to every connected router peer
bool probe_router;dealer的fq负责接收数据，lb负责发送数据：
int zmq::dealer_t::xsend (msg_t *msg_)
{
return sendpipe (msg_, NULL);
}
int zmq::dealer_t::xrecv (msg_t *msg_)
{
return recvpipe (msg_, NULL);
}
int zmq::dealer_t::sendpipe (msg_t *msg_, pipe_t **pipe_)
{
return lb.sendpipe (msg_, pipe_);
}
int zmq::dealer_t::recvpipe (msg_t *msg_, pipe_t **pipe_)
{
return fq.recvpipe (msg_, pipe_);
}dealer的发送和接收数据都是调用lb和fq的对应方法。
router和dealer的使用非常广泛，下面是zeromq官网的guild上一个例子：
// 2015-02-27T11:40+08:00
// ROUTER-to-DEALER example
#include "zhelpers.h"
#include <pthread.h>
#define NBR_WORKERS 10
static void *
worker_task(void *args)
{
void *context = zmq_ctx_new();
void *worker = zmq_socket(context, ZMQ_DEALER);
#if (defined (WIN32))
s_set_id(worker, (intptr_t)args);
#else
s_set_id(worker); // Set a printable identity
#endif
zmq_connect (worker, "tcp://localhost:5671");
int total = 0;
while (1) {
// Tell the broker we're ready for work
s_sendmore(worker, "");
s_send(worker, "Hi Boss");
// Get workload from broker, until finished
free(s_recv(worker)); // Envelope delimiter
char *workload = s_recv(worker);
int finished = (strcmp(workload, "Fired!") == 0);
free(workload);
if (finished) {
printf("Completed: %d tasks\n", total);
break;
}
total++;
// Do some random work
s_sleep(randof(500) + 1);
}
zmq_close(worker);
zmq_ctx_destroy(context);
return NULL;
}
// While this example runs in a single process, that is just to make
// it easier to start and stop the example. Each thread has its own
// context and conceptually acts as a separate process.
int main(void)
{
void *context = zmq_ctx_new();
void *broker = zmq_socket(context, ZMQ_ROUTER);
zmq_bind(broker, "tcp://*:5671");
srandom((unsigned)time(NULL));
int worker_nbr;
for (worker_nbr = 0; worker_nbr < NBR_WORKERS; worker_nbr++) {
pthread_t worker;
pthread_create(&worker, NULL, worker_task, (void *)(intptr_t)worker_nbr);
}
// Run for five seconds and then tell workers to end
int64_t end_time = s_clock() + 5000;
int workers_fired = 0;
while (1) {
// Next message gives us least recently used worker
char *identity = s_recv(broker);
s_sendmore(broker, identity);
free(identity);
free(s_recv(broker)); // Envelope delimiter
free(s_recv(broker)); // Response from worker
s_sendmore(broker, "");
// Encourage workers until it's time to fire them
if (s_clock() < end_time)
s_send(broker, "Work harder");
else {
s_send(broker, "Fired!");
if (++workers_fired == NBR_WORKERS)
break;
}
}
zmq_close(broker);
zmq_ctx_destroy(context);
return 0;
}这个例子比较简单，注意其中的空msg只是为了和rep以及req模型兼容，如果只是router和dealer模型则不需要这个空的分割消息。
req & rep
req和rep也是一组比较常用的模型，他们分别继承自dealer和router同时也通常和router与dealer结合使用。req和rep是一对一的请求应答模型，即req在发送一个条消息后必须接收到应答才能发送下一条消息。rep和req的实现比较简单，就是在router和dealer上做了一些更改，req发送的消息会带一个空的头作为分隔符。req底层的dealer不做任何处理，直接将消息发送给rep。rep的底层router加上一个identity返回给rep，rep直接将identity和rep写回给底层router，而将真正的数据返回给上层。之后rep发送的数据会直接写入到router，而该数据需要的头信息在接收数据时已经写入到router了。
下面是一个req和rep的例子：
// Load-balancing broker
// Clients and workers are shown here in-process
#include "zhelpers.h"
#include <pthread.h>
#define NBR_CLIENTS 10
#define NBR_WORKERS 3
// Dequeue operation for queue implemented as array of anything
#define DEQUEUE(q) memmove (&(q)[0], &(q)[1], sizeof (q) - sizeof (q [0]))
// Basic request-reply client using REQ socket
// Because s_send and s_recv can't handle 0MQ binary identities, we
// set a printable text identity to allow routing.
//
static void *
client_task(void *args)
{
void *context = zmq_ctx_new();
void *client = zmq_socket(context, ZMQ_REQ);
#if (defined (WIN32))
s_set_id(client, (intptr_t)args);
zmq_connect(client, "tcp://localhost:5672"); // frontend
#else
s_set_id(client); // Set a printable identity
zmq_connect(client, "ipc://frontend.ipc");
#endif
// Send request, get reply
s_send(client, "HELLO");
char *reply = s_recv(client);
printf("Client: %s\n", reply);
free(reply);
zmq_close(client);
zmq_ctx_destroy(context);
return NULL;
}
// While this example runs in a single process, that is just to make
// it easier to start and stop the example. Each thread has its own
// context and conceptually acts as a separate process.
// This is the worker task, using a REQ socket to do load-balancing.
// Because s_send and s_recv can't handle 0MQ binary identities, we
// set a printable text identity to allow routing.
static void *
worker_task(void *args)
{
void *context = zmq_ctx_new();
void *worker = zmq_socket(context, ZMQ_REQ);
#if (defined (WIN32))
s_set_id(worker, (intptr_t)args);
zmq_connect(worker, "tcp://localhost:5673"); // backend
#else
s_set_id(worker);
zmq_connect(worker, "ipc://backend.ipc");
#endif
// Tell broker we're ready for work
s_send(worker, "READY");
while (1) {
// Read and save all frames until we get an empty frame
// In this example there is only 1, but there could be more
char *identity = s_recv(worker);
char *empty = s_recv(worker);
assert(*empty == 0);
free(empty);
// Get request, send reply
char *request = s_recv(worker);
printf("Worker: %s\n", request);
free(request);
s_sendmore(worker, identity);
s_sendmore(worker, "");
s_send(worker, "OK");
free(identity);
}
zmq_close(worker);
zmq_ctx_destroy(context);
return NULL;
}
// This is the main task. It starts the clients and workers, and then
// routes requests between the two layers. Workers signal READY when
// they start; after that we treat them as ready when they reply with
// a response back to a client. The load-balancing data structure is
// just a queue of next available workers.
int main(void)
{
// Prepare our context and sockets
void *context = zmq_ctx_new();
void *frontend = zmq_socket(context, ZMQ_ROUTER);
void *backend = zmq_socket(context, ZMQ_ROUTER);
#if (defined (WIN32))
zmq_bind(frontend, "tcp://*:5672"); // frontend
zmq_bind(backend, "tcp://*:5673"); // backend
#else
zmq_bind(frontend, "ipc://frontend.ipc");
zmq_bind(backend, "ipc://backend.ipc");
#endif
int client_nbr;
for (client_nbr = 0; client_nbr < NBR_CLIENTS; client_nbr++) {
pthread_t client;
pthread_create(&client, NULL, client_task, (void *)(intptr_t)client_nbr);
}
int worker_nbr;
for (worker_nbr = 0; worker_nbr < NBR_WORKERS; worker_nbr++) {
pthread_t worker;
pthread_create(&worker, NULL, worker_task, (void *)(intptr_t)worker_nbr);
}
// Here is the main loop for the least-recently-used queue. It has two
// sockets; a frontend for clients and a backend for workers. It polls
// the backend in all cases, and polls the frontend only when there are
// one or more workers ready. This is a neat way to use 0MQ's own queues
// to hold messages we're not ready to process yet. When we get a client
// reply, we pop the next available worker and send the request to it,
// including the originating client identity. When a worker replies, we
// requeue that worker and forward the reply to the original client
// using the reply envelope.
// Queue of available workers
int available_workers = 0;
char *worker_queue[10];
while (1) {
zmq_pollitem_t items[] = {
{ backend, 0, ZMQ_POLLIN, 0 },
{ frontend, 0, ZMQ_POLLIN, 0 }
};
// Poll frontend only if we have available workers
int rc = zmq_poll(items, available_workers ? 2 : 1, -1);
if (rc == -1)
break; // Interrupted
// Handle worker activity on backend
if (items[0].revents & ZMQ_POLLIN) {
// Queue worker identity for load-balancing
char *worker_id = s_recv(backend);
assert(available_workers < NBR_WORKERS);
worker_queue[available_workers++] = worker_id;
// Second frame is empty
char *empty = s_recv(backend);
assert(empty[0] == 0);
free(empty);
// Third frame is READY or else a client reply identity
char *client_id = s_recv(backend);
// If client reply, send rest back to frontend
if (strcmp(client_id, "READY") != 0) {
empty = s_recv(backend);
assert(empty[0] == 0);
free(empty);
char *reply = s_recv(backend);
s_sendmore(frontend, client_id);
s_sendmore(frontend, "");
s_send(frontend, reply);
free(reply);
if (--client_nbr == 0)
break; // Exit after N messages
}
free(client_id);
}
// Here is how we handle a client request:
if (items[1].revents & ZMQ_POLLIN) {
// Now get next client request, route to last-used worker
// Client request is [identity][empty][request]
char *client_id = s_recv(frontend);
char *empty = s_recv(frontend);
assert(empty[0] == 0);
free(empty);
char *request = s_recv(frontend);
s_sendmore(backend, worker_queue[0]);
s_sendmore(backend, "");
s_sendmore(backend, client_id);
s_sendmore(backend, "");
s_send(backend, request);
free(client_id);
free(request);
// Dequeue and drop the next worker identity
free(worker_queue[0]);
DEQUEUE(worker_queue);
available_workers--;
}
}
zmq_close(frontend);
zmq_close(backend);
zmq_ctx_destroy(context);
return 0;
}下面这幅图是这个模型的示意图：
这个例子很好的解释了数据在req，rep和router之间传输de流程：
首先worker先去挂载router，之后就是任务分发，client发送一个数据，rep将他包装成三个msg：identity，empty和数据三部分。之后选择一个woker发送出去，此时backend需要发送的数据是五段，
真正到达req的是四条消息，req本身有去掉了空的分隔符，所以worker收到的消息是三条。之后worker把这三条发送出去，backend接收到的又是五条数据。然后通过frontend把后面三条发送出去，后面三条的第一条指定了clientid，所以这条消息又回到了发出任务的req。
这个流程比较复杂，但是只要搞清楚这个流程，rep，req，router和dealer的工作机制就非常明了了。
487
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
