change worker id to be a uuid

Cargo.lock

@@ -166,7 +166,7 @@ dependencies = [
  "serde_urlencoded",
  "smallvec",
  "socket2",
- "time",
+ "time 0.3.14",
  "url",
 ]
 
@@ -351,8 +351,11 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "bfd4d1b31faaa3a89d7934dbded3111da0d2ef28e3ebccdb4f0179f5929d1ef1"
 dependencies = [
  "iana-time-zone",
+ "js-sys",
  "num-integer",
  "num-traits",
+ "time 0.1.44",
+ "wasm-bindgen",
  "winapi",
 ]
 
@@ -393,7 +396,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "94d4706de1b0fa5b132270cddffa8585166037822e260a944fe161acd137ca05"
 dependencies = [
  "percent-encoding",
- "time",
+ "time 0.3.14",
  "version_check",
 ]
 
@@ -648,7 +651,7 @@ checksum = "4eb1a864a501629691edf6c15a593b7a51eebaa1e8468e9ddc623de7c9b58ec6"
 dependencies = [
  "cfg-if",
  "libc",
- "wasi",
+ "wasi 0.11.0+wasi-snapshot-preview1",
 ]
 
 [[package]]
@@ -929,7 +932,7 @@ checksum = "57ee1c23c7c63b0c9250c339ffdc69255f110b298b901b9f6c82547b7b87caaf"
 dependencies = [
  "libc",
  "log",
- "wasi",
+ "wasi 0.11.0+wasi-snapshot-preview1",
  "windows-sys",
 ]
 
@@ -1505,6 +1508,17 @@ dependencies = [
  "syn",
 ]
 
+[[package]]
+name = "time"
+version = "0.1.44"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6db9e6914ab8b1ae1c260a4ae7a49b6c5611b40328a735b21862567685e73255"
+dependencies = [
+ "libc",
+ "wasi 0.10.0+wasi-snapshot-preview1",
+ "winapi",
+]
+
 [[package]]
 name = "time"
 version = "0.3.14"
@@ -1686,6 +1700,15 @@ dependencies = [
  "percent-encoding",
 ]
 
+[[package]]
+name = "uuid"
+version = "1.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dd6469f4314d5f1ffec476e05f17cc9a78bc7a27a6a857842170bdf8d6f98d2f"
+dependencies = [
+ "getrandom",
+]
+
 [[package]]
 name = "vcpkg"
 version = "0.2.15"
@@ -1708,6 +1731,12 @@ dependencies = [
  "try-lock",
 ]
 
+[[package]]
+name = "wasi"
+version = "0.10.0+wasi-snapshot-preview1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1a143597ca7c7793eff794def352d41792a93c481eb1042423ff7ff72ba2c31f"
+
 [[package]]
 name = "wasi"
 version = "0.11.0+wasi-snapshot-preview1"
@@ -1879,6 +1908,7 @@ version = "0.1.0"
 dependencies = [
  "clap",
  "env_logger",
+ "futures",
  "log",
  "reqwest",
  "reqwest-middleware",
@@ -1901,11 +1931,14 @@ name = "zoidberg_server"
 version = "0.1.0"
 dependencies = [
  "actix-web",
+ "chrono",
  "clap",
  "env_logger",
  "futures",
  "log",
  "serde_json",
+ "tokio",
+ "uuid",
  "zoidberg_lib",
 ]
 

zoidberg_client/Cargo.toml

@@ -20,3 +20,4 @@ tokio = { version = "1", features = ["full"] }
 clap = "3.2.22"
 env_logger = "0.9"
 log = "0.4"
+futures = "0.3.24"

zoidberg_client/src/main.rs

@@ -1,13 +1,17 @@
 use clap::{App, Arg};
 use env_logger::Env;
+use futures::future::{AbortHandle, Abortable};
 use log;
 use reqwest::{header, Client, ClientBuilder};
 use std::error::Error;
-use std::process::Command;
+use std::process::Stdio;
+use std::sync::Arc;
 use std::time::Duration;
-use std::{thread, time};
+use tokio::{process::Command, time};
 
-use zoidberg_lib::types::{FetchResponse, Job, RegisterResponse, Status, Update};
+use zoidberg_lib::types::{
+    FetchRequest, FetchResponse, Heartbeat, Job, RegisterResponse, Status, Update,
+};
 
 const VERSION: &str = env!("CARGO_PKG_VERSION");
 
@@ -28,10 +32,11 @@ fn build_client(secret: &str) -> Client {
         .expect("Could not create client")
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct Worker {
-    id: i32,
+    id: String,
     secret: String,
+    server: String,
 }
 
 impl Worker {
@@ -45,23 +50,24 @@ impl Worker {
         let r: RegisterResponse = serde_json::from_str(&body)?;
         log::info!("registered worker with id: {}", &r.id);
         Ok(Worker {
-            id: r.id,
+            id: r.id.to_string(),
             secret: secret.to_string(),
+            server: server.to_string(),
         })
     }
 
-    async fn update(self: &Self, jobs: &[Job]) -> Result<(), Box<dyn Error>> {
+    async fn update(&self, jobs: &[Job]) -> Result<(), Box<dyn Error>> {
         let updates: Vec<Update> = jobs
             .iter()
             .map(|job| Update {
-                worker: self.id,
+                worker: self.id.clone(),
                 job: job.id,
                 status: job.status.clone(),
             })
             .collect();
 
         let body = build_client(&self.secret)
-            .post("http://localhost:8080/update")
+            .post(format!("{}/update", self.server))
             .json(&updates)
             .send()
             .await?
@@ -72,9 +78,12 @@ impl Worker {
         Ok(())
     }
 
-    async fn fetch(self: &Self) -> Result<FetchResponse, Box<dyn Error>> {
+    async fn fetch(&self) -> Result<FetchResponse, Box<dyn Error>> {
         let res = build_client(&self.secret)
-            .get("http://localhost:8080/fetch")
+            .post(format!("{}/fetch", self.server))
+            .json(&FetchRequest {
+                worker_id: self.id.clone(),
+            })
             .send()
             .await?;
         let body = res.text().await?;
@@ -82,35 +91,35 @@ impl Worker {
         Ok(resp)
     }
 
-    async fn run(self: &Self, job: &Job) -> Result<(), Box<dyn Error>> {
+    async fn heartbeat(&self) {
-        let output = Command::new("bash").arg("-c").arg(&job.cmd).output()?;
+        let _ = build_client(&self.secret)
-
+            .post(format!("{}/heartbeat", self.server))
-        log::info!(
+            .json(&Heartbeat {
-            "command: {}\nstdout: {}\nstderr: {}",
+                id: self.id.clone(),
-            &job.cmd,
+            })
-            String::from_utf8_lossy(&output.stdout),
+            .send()
-            String::from_utf8_lossy(&output.stderr)
+            .await;
-        );
-        match output.status.success() {
-            true => Ok(()),
-            false => Err(Box::from("Job failed")),
-        }
     }
+}
 
-    async fn process(self: &Self, jobs: &[Job]) {
+async fn run(job: &Job) -> Result<(), Box<dyn Error>> {
-        for job in jobs {
+    let output = Command::new("bash")
-            let status = match self.run(&job).await {
+        .arg("-c")
-                Ok(()) => Status::Completed,
+        .arg(&job.cmd)
-                Err(..) => Status::Failed,
+        .stdout(Stdio::piped())
-            };
+        .stderr(Stdio::piped())
-            let n = &[Job {
+        .output();
-                status,
+    let output = output.await?;
-                ..job.clone()
+
-            }];
+    log::info!(
-            if let Err(error) = self.update(n).await {
+        "command: {}\nstdout: {}\nstderr: {}",
-                log::info!("Could not update job: {}", error);
+        &job.cmd,
-            }
+        String::from_utf8_lossy(&output.stdout),
-        }
+        String::from_utf8_lossy(&output.stderr)
+    );
+    match output.status.success() {
+        true => Ok(()),
+        false => Err(Box::from("Job failed")),
     }
 }
 
@@ -130,27 +139,72 @@ async fn main() -> Result<(), Box<dyn Error>> {
         .get_matches();
     let server = matches.value_of("server").unwrap();
     let secret = std::env::var("ZOIDBERG_SECRET").unwrap_or_else(|_| {
-        println!("Please set the $ZOIDBERG_SECRET environment variable");
+        eprintln!("Please set the $ZOIDBERG_SECRET environment variable");
         std::process::exit(1);
     });
 
-    let client = Worker::new(server, &secret)
+    let client = Arc::new(
-        .await
+        Worker::new(server, &secret)
-        .expect("Could not create client");
+            .await
+            .expect("Could not create client"),
+    );
 
     let pause = time::Duration::from_secs(1);
-    let long_pause = time::Duration::from_secs(20);
+    let long_pause = time::Duration::from_secs(40);
+    let heartbeat_pause = time::Duration::from_secs(30);
 
+    let (heartbeat_handle, abort_registration) = AbortHandle::new_pair();
+    let c = Arc::clone(&client);
+    tokio::spawn(Abortable::new(
+        async move {
+            loop {
+                time::sleep(heartbeat_pause).await;
+                c.heartbeat().await;
+            }
+        },
+        abort_registration,
+    ));
+
+    let mut fail_counter = 0;
     loop {
-        if let Ok(fetch) = client.fetch().await {
+        let jobs = if let Ok(fetch) = client.fetch().await {
+            fail_counter = 0;
             match fetch {
-                FetchResponse::Nop => thread::sleep(pause),
+                FetchResponse::Nop => {
-                FetchResponse::StopWorking => break,
+                    time::sleep(pause).await;
-                FetchResponse::Jobs(jobs) => client.process(&jobs).await,
+                    continue;
+                }
+                FetchResponse::Terminate(m) => {
+                    println!("Terminate worker: {}", m);
+                    break;
+                }
+                FetchResponse::Jobs(jobs) => jobs,
             }
         } else {
-            thread::sleep(long_pause);
+            fail_counter += 1;
+            if fail_counter == 3 {
+                log::error!("failed to fetch three times, assume that server crashed and exit");
+                std::process::exit(1);
+            }
+            log::error!("failed to fetch new jobs");
+            time::sleep(long_pause).await;
+            continue;
+        };
+
+        for job in jobs {
+            let status = match run(&job).await {
+                Ok(()) => Status::Completed,
+                Err(..) => Status::Failed,
+            };
+            let update = &[Job {
+                status,
+                ..job.clone()
+            }];
+            if let Err(error) = client.update(update).await {
+                log::info!("Could not update job: {}", error);
+            }
         }
     }
+    heartbeat_handle.abort();
     Ok(())
 }

zoidberg_lib/src/types.rs

@@ -4,7 +4,7 @@ use serde::{Deserialize, Serialize};
 
 #[derive(Serialize, Deserialize)]
 pub struct Update {
-    pub worker: i32,
+    pub worker: String,
     pub job: i32,
     pub status: Status,
 }
@@ -12,7 +12,7 @@ pub struct Update {
 #[derive(Serialize, Deserialize, Clone, Debug)]
 pub enum Status {
     Submitted,
-    Running,
+    Running(String),
     Completed,
     Failed,
 }
@@ -27,7 +27,7 @@ impl fmt::Display for Status {
     fn fmt(self: &Self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         match self {
             Status::Submitted => write!(f, "submitted"),
-            Status::Running => write!(f, "running"),
+            Status::Running(w) => write!(f, "running on worker {}", w),
             Status::Completed => write!(f, "completed"),
             Status::Failed => write!(f, "failed"),
         }
@@ -55,13 +55,18 @@ pub struct Node {
 
 #[derive(Serialize, Deserialize)]
 pub struct RegisterResponse {
-    pub id: i32,
+    pub id: String,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct FetchRequest {
+    pub worker_id: String,
 }
 
 #[derive(Serialize, Deserialize)]
 pub enum FetchResponse {
     Jobs(Vec<Job>),
-    StopWorking,
+    Terminate(String),
     Nop,
 }
 
@@ -73,5 +78,13 @@ pub struct Submit {
 #[derive(Serialize, Deserialize)]
 pub struct Worker {
     #[serde(default)]
-    pub id: i32,
+    pub id: String,
+    #[serde(default)]
+    pub last_heartbeat: Option<i64>,
+}
+
+#[derive(Serialize, Deserialize)]
+pub struct Heartbeat {
+    #[serde(default)]
+    pub id: String,
 }

zoidberg_server/Cargo.toml

@@ -16,3 +16,6 @@ clap = "3.2"
 env_logger = "0.9"
 log = "0.4"
 futures = "0.3.24"
+tokio = { version = "1", features = ["full"] }
+chrono = "0.4.22"
+uuid = { version = "1.1.2", features = ["v4"] }

zoidberg_server/src/main.rs

@@ -3,19 +3,24 @@ use actix_web::{
     dev, get, middleware::Logger, post, web, App, Error, FromRequest, HttpRequest, HttpResponse,
     HttpServer, Responder, Result,
 };
+use chrono::Utc;
 use clap;
 use env_logger::Env;
 use futures::future::{err, ok, Ready};
 use log;
 use std::sync::Mutex;
-use zoidberg_lib::types::{FetchResponse, Job, RegisterResponse, StatusRequest, Update, Worker};
+use std::time::Duration;
+use uuid::Uuid;
+use zoidberg_lib::types::{
+    FetchRequest, FetchResponse, Heartbeat, Job, RegisterResponse, Status, StatusRequest, Update,
+    Worker,
+};
 
 mod webpage;
 
 const VERSION: &str = env!("CARGO_PKG_VERSION");
 
 struct State {
-    counter_workers: Mutex<i32>,
     counter_jobs: Mutex<i32>,
     workers: Mutex<Vec<Worker>>,
     new_jobs: Mutex<Vec<Job>>,
@@ -25,7 +30,6 @@ struct State {
 impl State {
     fn new() -> Self {
         Self {
-            counter_workers: Mutex::new(0),
             counter_jobs: Mutex::new(0),
             workers: Mutex::new(Vec::new()),
             new_jobs: Mutex::new(Vec::new()),
@@ -44,7 +48,6 @@ impl FromRequest for Authorization {
         if let Some(head) = req.headers().get("cookie") {
             if let Ok(cookie) = head.to_str() {
                 if let Some(secret) = req.app_data::<String>() {
-                    println!("{} == {}", secret, cookie);
                     if secret == cookie {
                         return ok(Authorization {});
                     } else {
@@ -67,24 +70,40 @@ async fn index(data: web::Data<State>) -> impl Responder {
 
 #[get("/register")]
 async fn register(data: web::Data<State>, _: Authorization) -> Result<impl Responder> {
-    let mut counter_workers = data.counter_workers.lock().unwrap();
-    *counter_workers += 1;
-
     let mut workers = data.workers.lock().unwrap();
+    let uuid = Uuid::new_v4().to_string();
     workers.push(Worker {
-        id: *counter_workers,
+        id: uuid.clone(),
+        last_heartbeat: None,
     });
 
-    log::info!("Registered worker node with id: {}", *counter_workers);
+    log::info!("Registered worker node with id: {}", uuid);
-    Ok(web::Json(RegisterResponse {
+    Ok(web::Json(RegisterResponse { id: uuid }))
-        id: *counter_workers,
-    }))
 }
 
-#[get("/fetch")]
+#[post("/fetch")]
-async fn fetch(data: web::Data<State>, _: Authorization) -> Result<impl Responder> {
+async fn fetch(
+    data: web::Data<State>,
+    f: web::Json<FetchRequest>,
+    _: Authorization,
+) -> Result<impl Responder> {
+    let requesting_worker = f.into_inner().worker_id;
+    {
+        let workers = data.workers.lock().unwrap();
+        if workers.iter().filter(|w| w.id == requesting_worker).count() != 1 {
+            return Ok(web::Json(FetchResponse::Terminate(
+                "Worker not found".into(),
+            )));
+        }
+    }
     let mut new_jobs = data.new_jobs.lock().unwrap();
     if let Some(j) = new_jobs.pop() {
+        let mut jobs = data.jobs.lock().unwrap();
+        for cj in jobs.iter_mut() {
+            if cj.id == j.id {
+                cj.status = Status::Running(requesting_worker.clone())
+            }
+        }
         return Ok(web::Json(FetchResponse::Jobs(vec![j])));
     }
     Ok(web::Json(FetchResponse::Nop))
@@ -131,6 +150,22 @@ async fn update(
     Ok(format!("Worker updated {} job(s)", n))
 }
 
+#[post("/heartbeat")]
+async fn heartbeat(
+    heartbeat: web::Json<Heartbeat>,
+    data: web::Data<State>,
+    _: Authorization,
+) -> Result<String> {
+    log::info!("Heartbeat from worker {}", heartbeat.id);
+    let mut workers = data.workers.lock().unwrap();
+    for w in workers.iter_mut() {
+        if w.id == heartbeat.id {
+            w.last_heartbeat = Some(Utc::now().timestamp());
+        }
+    }
+    Ok(format!("Heartbeat from worker {}", heartbeat.id))
+}
+
 #[post("/submit")]
 async fn submit(
     data: web::Data<State>,
@@ -163,7 +198,7 @@ async fn main() -> std::io::Result<()> {
     env_logger::Builder::from_env(Env::default().default_filter_or("zoidberg_server=info")).init();
 
     let secret = std::env::var("ZOIDBERG_SECRET").unwrap_or_else(|_| {
-        println!("Please set the $ZOIDBERG_SECRET environment variable");
+        eprintln!("Please set the $ZOIDBERG_SECRET environment variable");
         std::process::exit(1);
     });
 
@@ -174,6 +209,18 @@ async fn main() -> std::io::Result<()> {
 
     let state = web::Data::new(State::new());
 
+    let s = state.clone();
+    tokio::spawn(async move {
+        loop {
+            tokio::time::sleep(Duration::from_secs(10)).await;
+            let mut workers = s.workers.lock().unwrap();
+            workers.retain(|w| match w.last_heartbeat {
+                None => true,
+                Some(t) => Utc::now().timestamp() - t < 60,
+            })
+        }
+    });
+
     HttpServer::new(move || {
         App::new()
             .wrap(Logger::default())
@@ -184,6 +231,7 @@ async fn main() -> std::io::Result<()> {
             .service(fetch)
             .service(status)
             .service(update)
+            .service(heartbeat)
             .service(submit)
     })
     .bind(("127.0.0.1", 8080))?
@@ -256,8 +304,8 @@ mod tests {
             FetchResponse::Nop => {
                 panic!("did not expect FetchResponse::Nop")
             }
-            FetchResponse::StopWorking => {
+            FetchResponse::Terminate(w) => {
-                panic!("did not expect FetchResponse::NotWorking")
+                panic!("did not expect FetchResponse::Terminate from worker {}", w)
             }
             FetchResponse::Jobs(new_jobs) => {
                 assert_eq!(new_jobs[0].id, 0);
@@ -280,7 +328,7 @@ mod tests {
                     jobs: Mutex::new(vec![Job {
                         id: 1,
                         cmd: cmd.clone(),
-                        status: Status::Running,
+                        status: Status::Running(0),
                     }]),
                 }))
                 .service(status),
@@ -309,7 +357,7 @@ mod tests {
             .set_json(vec![Update {
                 worker: 0,
                 job: 0,
-                status: Status::Running,
+                status: Status::Running(0),
             }])
             .uri("/update")
             .to_request();
@@ -331,7 +379,7 @@ mod tests {
             .set_json(vec![Job {
                 id: 0,
                 cmd: String::from("hi"),
-                status: Status::Running,
+                status: Status::Running(0),
             }])
             .uri("/submit")
             .to_request();

zoidberg_server/src/webpage.rs

@@ -1,3 +1,4 @@
+use chrono::Utc;
 use zoidberg_lib::types::{Job, Worker};
 
 // TODO: write nicer frontend
@@ -17,10 +18,17 @@ pub fn render(jobs: &[Job], workers: &[Worker]) -> String {
         + "</tbody></table>";
 
     let workers_html: String = String::from("<table class=\"table is-hoverable\">")
-        + "<thead><tr><th><td>ID</td></th></tr></thead><tbody>"
+        + "<thead><tr><th><td>ID</td><td>last heartbeat</td></th></tr></thead><tbody>"
         + &workers
             .iter()
-            .map(|w| format!("<tr><th></th><td>{}</td></tr>", w.id))
+            .map(|w| {
+                let ts = if let Some(ts) = w.last_heartbeat {
+                    format!("{}", Utc::now().timestamp() - ts)
+                } else {
+                    String::from("")
+                };
+                format!("<tr><th></th><td>{}</td><td>{}</td></tr>", w.id, ts)
+            })
             .collect::<Vec<String>>()
             .join("\n")
         + "</tbody></table>";
@@ -43,7 +51,8 @@ pub fn render(jobs: &[Job], workers: &[Worker]) -> String {
   <head>
     <meta charset="utf-8">
     <meta name="viewport" content="width=device-width, initial-scale=1">
-    <title>Hello Bulma!</title>
+    <title>Zoidberg</title>
+    <link rel="icon" href="data:,">
     <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/bulma@0.9.4/css/bulma.min.css">
     {}
   </head>