1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592
use atomic_option::AtomicOption;
use failure::Error;
use fnv::FnvHashMap;
use proto::{self, Client, ClientOptions, HeartbeatStatus, Reconnect};
use std::error::Error as StdError;
use std::io::prelude::*;
use std::net::TcpStream;
use std::sync::{atomic, Arc};
use proto::{Ack, Fail, Job};
const STATUS_RUNNING: usize = 0;
const STATUS_QUIET: usize = 1;
const STATUS_TERMINATING: usize = 2;
type JobRunner<E> = Fn(Job) -> Result<(), E> + Send + Sync;
type BoxedJobRunner<E> = Box<JobRunner<E>>;
/// `Consumer` is used to run a worker that processes jobs provided by Faktory.
///
/// # Building the worker
///
/// Faktory needs a decent amount of information from its workers, such as a unique worker ID, a
/// hostname for the worker, its process ID, and a set of labels used to identify the worker. In
/// order to enable setting all these, constructing a worker is a two-step process. You first use a
/// [`ConsumerBuilder`](struct.ConsumerBuilder.html) (which conveniently implements a sensible
/// `Default`) to set the worker metadata, as well as to register any job handlers. You then use
/// one of the `connect_*` methods to finalize the worker and connect to the Faktory server.
///
/// In most cases, `ConsumerBuilder::default()` will do what you want. You only need to augment it
/// with calls to [`register`](struct.ConsumerBuilder.html#method.register) to register handlers
/// for each of your job types, and then you can connect. If you have different *types* of workers,
/// you may also want to use [`labels`](struct.ConsumerBuilder.html#method.labels) to distinguish
/// them in the Faktory Web UI. To specify that some jobs should only go to some workers, use
/// different queues.
///
/// ## Handlers
///
/// For each [`Job`](struct.Job.html) that the worker receives, the handler that is registered for
/// that job's type will be called. If a job is received with a type for which no handler exists,
/// the job will be failed and returned to the Faktory server. Similarly, if a handler returns an
/// error response, the job will be failed, and the error reported back to the Faktory server.
///
/// If you are new to Rust, getting the handler types to work out can be a little tricky. If you
/// want to understand why, I highly recommend that you have a look at the chapter on [closures and
/// generic
/// parameters](https://doc.rust-lang.org/book/second-edition/ch13-01-closures.html#using-closures-with-generic-parameters-and-the-fn-traits)
/// in the Rust Book. If you just want it to work, my recommendation is to either use regular
/// functions instead of closures, and giving `&func_name` as the handler, **or** wrapping all your
/// closures in `Box::new()`.
///
/// ## Concurrency
///
/// By default, only a single thread is spun up to process the jobs given to this worker. If you
/// want to dedicate more resources to processing jobs, you have a number of options listed below.
/// As you go down the list below, efficiency increases, but fault isolation decreases. I will not
/// give further detail here, but rather recommend that if these don't mean much to you, you should
/// use the last approach and let the library handle the concurrency for you.
///
/// - You can spin up more worker processes by launching your worker program more than once.
/// - You can create more than one `Consumer`.
/// - You can call [`ConsumerBuilder::workers`](struct.ConsumerBuilder.html#method.workers) to set
/// the number of worker threads you'd like the `Consumer` to use internally.
///
/// # Connecting to Faktory
///
/// To fetch jobs, the `Consumer` must first be connected to the Faktory server. Exactly how you do
/// that depends on your setup. In most cases, you'll want to use `Consumer::connect`, and provide
/// a connection URL. If you supply a URL, it must be of the form:
///
/// ```text
/// protocol://[:password@]hostname[:port]
/// ```
///
/// Faktory suggests using the `FAKTORY_PROVIDER` and `FAKTORY_URL` environment variables (see
/// their docs for more information) with `localhost:7419` as the fallback default. If you want
/// this behavior, pass `None` as the URL.
///
/// See the [`Producer` examples](struct.Producer.html#examples) for examples of how to connect to
/// different Factory setups.
///
/// # Worker lifecycle
///
/// Okay, so you've built your worker and connected to the Faktory server. Now what?
///
/// If all this process is doing is handling jobs, reconnecting on failure, and exiting when told
/// to by the Faktory server, you should use
/// [`run_to_completion`](struct.Consumer.html#method.run_to_completion). If you want more
/// fine-grained control over the lifetime of your process, you should use
/// [`Consumer::run`](struct.Consumer.html#method.run). See the documentation for each of these
/// methods for details.
///
/// # Examples
///
/// Create a worker with all default options, register a single handler (for the `foobar` job
/// type), connect to the Faktory server, and start accepting jobs.
///
/// ```no_run
/// use faktory::ConsumerBuilder;
/// use std::io;
/// let mut c = ConsumerBuilder::default();
/// c.register("foobar", |job| -> io::Result<()> {
/// println!("{:?}", job);
/// Ok(())
/// });
/// let mut c = c.connect(None).unwrap();
/// if let Err(e) = c.run(&["default"]) {
/// println!("worker failed: {}", e);
/// }
/// ```
pub struct Consumer<S, E>
where
S: Read + Write,
{
c: Client<S>,
last_job_results: Arc<Vec<AtomicOption<Result<String, Fail>>>>,
running_jobs: Arc<Vec<AtomicOption<String>>>,
callbacks: Arc<FnvHashMap<String, BoxedJobRunner<E>>>,
terminated: bool,
}
/// Convenience wrapper for building a Faktory worker.
///
/// See the [`Consumer`](struct.Consumer.html) documentation for details.
pub struct ConsumerBuilder<E> {
opts: ClientOptions,
workers: usize,
callbacks: FnvHashMap<String, BoxedJobRunner<E>>,
}
impl<E> Default for ConsumerBuilder<E> {
/// Construct a new worker with default worker options and the url fetched from environment
/// variables.
///
/// This will construct a worker where:
///
/// - `hostname` is this machine's hostname.
/// - `wid` is a randomly generated string.
/// - `pid` is the OS PID of this process.
/// - `labels` is `["rust"]`.
///
fn default() -> Self {
ConsumerBuilder {
opts: ClientOptions::default(),
workers: 1,
callbacks: Default::default(),
}
}
}
impl<E> ConsumerBuilder<E> {
/// Set the hostname to use for this worker.
///
/// Defaults to the machine's hostname as reported by the operating system.
pub fn hostname(&mut self, hn: String) -> &mut Self {
self.opts.hostname = Some(hn);
self
}
/// Set a unique identifier for this worker.
///
/// Defaults to a randomly generated ASCII string.
pub fn wid(&mut self, wid: String) -> &mut Self {
self.opts.wid = Some(wid);
self
}
/// Set the labels to use for this worker.
///
/// Defaults to `["rust"]`.
pub fn labels(&mut self, labels: Vec<String>) -> &mut Self {
self.opts.labels = labels;
self
}
/// Set the number of workers to use for `run` and `run_to_completion_*`.
///
/// Defaults to 1.
pub fn workers(&mut self, w: usize) -> &mut Self {
self.workers = w;
self
}
/// Register a handler function for the given job type (`kind`).
///
/// Whenever a job whose type matches `kind` is fetched from the Faktory, the given handler
/// function is called with that job as its argument.
pub fn register<K, H>(&mut self, kind: K, handler: H) -> &mut Self
where
K: Into<String>,
// Annoyingly, can't just use the JobRunner<E> type alias here.
H: Fn(Job) -> Result<(), E> + Send + Sync + 'static,
{
self.callbacks.insert(kind.into(), Box::new(handler));
self
}
/// Connect to a Faktory server.
///
/// If `url` is not given, will use the standard Faktory environment variables. Specifically,
/// `FAKTORY_PROVIDER` is read to get the name of the environment variable to get the address
/// from (defaults to `FAKTORY_URL`), and then that environment variable is read to get the
/// server address. If the latter environment variable is not defined, the connection will be
/// made to
///
/// ```text
/// tcp://localhost:7419
/// ```
///
/// If `url` is given, but does not specify a port, it defaults to 7419.
pub fn connect(self, url: Option<&str>) -> Result<Consumer<TcpStream, E>, Error> {
let url = match url {
Some(url) => proto::url_parse(url),
None => proto::url_parse(&proto::get_env_url()),
}?;
let stream = TcpStream::connect(proto::host_from_url(&url))?;
Self::connect_with(self, stream, url.password().map(|p| p.to_string()))
}
/// Connect to a Faktory server with a non-standard stream.
pub fn connect_with<S: Read + Write>(
mut self,
stream: S,
pwd: Option<String>,
) -> Result<Consumer<S, E>, Error> {
self.opts.password = pwd;
Ok(Consumer::new(
Client::new(stream, self.opts)?,
self.workers,
self.callbacks,
))
}
}
enum Failed<E: StdError> {
Application(E),
BadJobType(String),
}
impl<E, S: Read + Write> Consumer<S, E> {
fn new(c: Client<S>, workers: usize, callbacks: FnvHashMap<String, BoxedJobRunner<E>>) -> Self {
Consumer {
c: c,
callbacks: Arc::new(callbacks),
running_jobs: Arc::new((0..workers).map(|_| AtomicOption::empty()).collect()),
last_job_results: Arc::new((0..workers).map(|_| AtomicOption::empty()).collect()),
terminated: false,
}
}
}
impl<E, S: Read + Write + Reconnect> Consumer<S, E> {
fn reconnect(&mut self) -> Result<(), Error> {
self.c.reconnect()
}
}
impl<S, E> Consumer<S, E>
where
S: Read + Write,
E: StdError,
{
fn run_job(&mut self, job: Job) -> Result<(), Failed<E>> {
match self.callbacks.get(&job.kind) {
Some(callback) => callback(job).map_err(Failed::Application),
None => {
// cannot execute job, since no handler exists
Err(Failed::BadJobType(job.kind))
}
}
}
/// Fetch and run a single job on the current thread, and then return.
pub fn run_one<Q>(&mut self, worker: usize, queues: &[Q]) -> Result<bool, Error>
where
Q: AsRef<str>,
{
// get a job
let job = match self.c.fetch(queues)? {
Some(job) => job,
None => return Ok(false),
};
// remember the job id
let jid = job.jid.clone();
// keep track of running job in case we're terminated during it
self.running_jobs[worker].swap(Box::new(jid.clone()), atomic::Ordering::SeqCst);
// process the job
let r = self.run_job(job);
// report back
match r {
Ok(_) => {
// job done -- acknowledge
// remember it in case we fail to notify the server (e.g., broken connection)
self.last_job_results[worker]
.swap(Box::new(Ok(jid.clone())), atomic::Ordering::SeqCst);
self.c.issue(&Ack::new(jid))?.await_ok()?;
}
Err(e) => {
// job failed -- let server know
// "unknown" is the errtype used by the go library too
let fail = match e {
Failed::BadJobType(jt) => {
Fail::new(jid, "unknown", format!("No handler for {}", jt))
}
Failed::Application(e) => {
let mut f = Fail::new(jid, "unknown", format!("{}", e));
let mut root = e.cause();
let mut backtrace = Vec::new();
while let Some(r) = root.take() {
backtrace.push(format!("{}", r));
root = r.cause();
}
f.set_backtrace(backtrace);
f
}
};
let fail2 = fail.clone();
self.last_job_results[worker].swap(Box::new(Err(fail)), atomic::Ordering::SeqCst);
self.c.issue(&fail2)?.await_ok()?;
}
}
// we won't have to tell the server again
self.last_job_results[worker].take(atomic::Ordering::SeqCst);
self.running_jobs[worker].take(atomic::Ordering::SeqCst);
Ok(true)
}
#[cfg(test)]
pub(crate) fn run_n<Q>(&mut self, n: usize, queues: &[Q]) -> Result<(), Error>
where
Q: AsRef<str>,
{
for _ in 0..n {
self.run_one(0, queues)?;
}
Ok(())
}
}
impl<S, E> Consumer<S, E>
where
S: Read + Write + Reconnect + Send + 'static,
E: StdError + 'static,
{
fn for_worker(&mut self) -> Result<Self, Error> {
Ok(Consumer {
c: self.c.connect_again()?,
callbacks: Arc::clone(&self.callbacks),
running_jobs: Arc::clone(&self.running_jobs),
last_job_results: Arc::clone(&self.last_job_results),
terminated: self.terminated,
})
}
/// Run this worker on the given `queues` until an I/O error occurs (`Err` is returned), or
/// until the server tells the worker to disengage (`Ok` is returned).
///
/// The value in an `Ok` indicates the number of workers that may still be processing jobs.
///
/// Note that if the worker fails, [`reconnect()`](struct.Consumer.html#method.reconnect)
/// should likely be called before calling `run()` again. If an error occurred while reporting
/// a job success or failure, the result will be re-reported to the server without re-executing
/// the job. If the worker was terminated (i.e., `run` returns with an `Ok` response), the
/// worker should **not** try to resume by calling `run` again. This will cause a panic.
pub fn run<Q>(&mut self, queues: &[Q]) -> Result<usize, Error>
where
Q: AsRef<str>,
{
assert!(!self.terminated, "do not re-run a terminated worker");
assert_eq!(Arc::strong_count(&self.last_job_results), 1);
// retry delivering notification about our last job result.
// we know there's no leftover thread at this point, so there's no race on the option.
for last_job_result in self.last_job_results.iter() {
if let Some(res) = last_job_result.take(atomic::Ordering::SeqCst) {
let r = match *res {
Ok(ref jid) => self.c.issue(&Ack::new(&**jid)),
Err(ref fail) => self.c.issue(fail),
};
let r = match r {
Ok(r) => r,
Err(e) => {
last_job_result.swap(res, atomic::Ordering::SeqCst);
return Err(e);
}
};
if let Err(e) = r.await_ok() {
// it could be that the server did previously get our ACK/FAIL, and that it was
// the resulting OK that failed. in that case, we would get an error response
// when re-sending the job response. this should not count as critical. other
// errors, however, should!
if e.downcast_ref::<::std::io::Error>().is_some() {
last_job_result.swap(res, atomic::Ordering::SeqCst);
return Err(e);
}
}
}
}
// keep track of the current status of each worker
let status: Vec<_> = (0..self.running_jobs.len())
.map(|_| Arc::new(atomic::AtomicUsize::new(STATUS_RUNNING)))
.collect();
// start worker threads
use std::thread;
let workers = status
.iter()
.enumerate()
.map(|(worker, status)| {
let mut w = self.for_worker()?;
let status = Arc::clone(status);
let queues: Vec<_> = queues.into_iter().map(|s| s.as_ref().to_string()).collect();
Ok(thread::spawn(move || {
while status.load(atomic::Ordering::SeqCst) == STATUS_RUNNING {
if let Err(e) = w.run_one(worker, &queues[..]) {
status.store(STATUS_TERMINATING, atomic::Ordering::SeqCst);
return Err(e);
}
}
status.store(STATUS_TERMINATING, atomic::Ordering::SeqCst);
Ok(())
}))
})
.collect::<Result<Vec<_>, Error>>()?;
// listen for heartbeats
let mut target = STATUS_RUNNING;
let exit = {
use std::time;
let mut last = time::Instant::now();
loop {
use std::thread;
thread::sleep(time::Duration::from_millis(100));
// has a worker failed?
if target == STATUS_RUNNING
&& status
.iter()
.any(|s| s.load(atomic::Ordering::SeqCst) == STATUS_TERMINATING)
{
// tell all workers to exit
// (though chances are they've all failed already)
for s in &status {
s.store(STATUS_TERMINATING, atomic::Ordering::SeqCst);
}
break Ok(false);
}
if last.elapsed().as_secs() < 5 {
// don't sent a heartbeat yet
continue;
}
match self.c.heartbeat() {
Ok(hb) => {
match hb {
HeartbeatStatus::Ok => {}
HeartbeatStatus::Quiet => {
// tell the workers to eventually terminate
for s in &status {
s.store(STATUS_QUIET, atomic::Ordering::SeqCst);
}
target = STATUS_QUIET;
}
HeartbeatStatus::Terminate => {
// tell the workers to terminate
// *and* fail the current job and immediately return
for s in &status {
s.store(STATUS_QUIET, atomic::Ordering::SeqCst);
}
break Ok(true);
}
}
}
Err(e) => {
// for this to fail, the workers have probably also failed
for s in &status {
s.store(STATUS_TERMINATING, atomic::Ordering::SeqCst);
}
break Err(e);
}
}
last = time::Instant::now();
}
};
// there are a couple of cases here:
//
// - we got TERMINATE, so we should just return, even if a worker is still running
// - we got TERMINATE and all workers has exited
// - we got an error from heartbeat()
//
self.terminated = exit.is_ok();
if let Ok(true) = exit {
// FAIL currently running jobs even though they're still running
let mut running = 0;
for running_job in self.running_jobs.iter() {
if let Some(jid) = running_job.take(atomic::Ordering::SeqCst) {
let f = Fail::new(&**jid, "unknown", "terminated");
// if this fails, we don't want to exit with Err(),
// because we *were* still terminated!
self.c.issue(&f).and_then(|r| r.await_ok()).is_ok();
running += 1;
}
}
if running != 0 {
return Ok(running);
}
}
match exit {
Ok(_) => {
// we want to expose any worker errors
workers
.into_iter()
.map(|w| w.join().unwrap())
.collect::<Result<Vec<_>, _>>()
.map(|_| 0)
}
Err(e) => {
// we want to expose worker errors, or otherwise the heartbeat error
workers
.into_iter()
.map(|w| w.join().unwrap())
.collect::<Result<Vec<_>, _>>()
.and_then(|_| Err(e))
}
}
}
/// Run this worker until the server tells us to exit or a connection cannot be re-established.
///
/// This function never returns. When the worker decides to exit, the process is terminated.
pub fn run_to_completion<Q>(mut self, queues: &[Q]) -> !
where
Q: AsRef<str>,
{
use std::process;
while self.run(queues).is_err() {
if self.reconnect().is_err() {
break;
}
}
process::exit(0);
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
// https://github.com/rust-lang/rust/pull/42219
//#[allow_fail]
#[ignore]
fn it_works() {
use producer::Producer;
use std::io;
let mut p = Producer::connect(None).unwrap();
let mut j = Job::new("foobar", vec!["z"]);
j.queue = "worker_test_1".to_string();
p.enqueue(j).unwrap();
let mut c = ConsumerBuilder::default();
c.register("foobar", |job: Job| -> Result<(), io::Error> {
assert_eq!(job.args, vec!["z"]);
Ok(())
});
let mut c = c.connect(None).unwrap();
let e = c.run_n(1, &["worker_test_1"]);
if e.is_err() {
println!("{:?}", e);
}
assert!(e.is_ok());
}
}