fail jobs if workers die

submitters/snakemake/zoidberg/grid-status.py

@@ -2,8 +2,17 @@
 
 import sys
 import requests
+from os import environ
 
-resp = requests.post("http://localhost:8080/status", json=[{"id": int(sys.argv[1])}])
+def print_and_exit(s):
+    print(s)
+    exit(0)
+
+resp = requests.post(
+    "http://localhost:8080/status",
+    json=[{"id": int(sys.argv[1])}],
+    headers={"cookie": environ["ZOIDBERG_SECRET"]},
+)
 
 translation = {
     "Submitted": "running",
@@ -11,4 +20,12 @@ translation = {
     "Failed": "failed",
 }
 
-print(translation[resp.json()[0]["status"]])
+j = resp.json()
+
+if len(j) == 0:
+    print_and_exit("failed")
+
+if "Running" in j[0]["status"]:
+    print_and_exit("running")
+
+print_and_exit(translation[resp.json()[0]["status"]])

submitters/snakemake/zoidberg/grid-submit.py

@@ -2,6 +2,7 @@
 
 import sys
 import requests
+from os import environ
 
 jobscript = sys.argv[1]
 
@@ -10,6 +11,7 @@ resp = requests.post(
     json=[
         {"cmd": jobscript},
     ],
+    headers={"cookie": environ["ZOIDBERG_SECRET"]},
 )
 assert resp.ok, "http request failed"
 

zoidberg_server/src/main.rs

@@ -40,8 +40,8 @@ impl State {
 #[get("/")]
 async fn index(data: web::Data<State>) -> impl Responder {
     let workers = data.workers.lock().unwrap();
-    let new_jobs = data.new_jobs.lock().unwrap();
-    let page = webpage::render(&*new_jobs, &*workers);
+    let jobs = data.jobs.lock().unwrap();
+    let page = webpage::render(&*jobs, &*workers);
     HttpResponse::Ok().body(page)
 }
 
@@ -133,7 +133,7 @@ async fn heartbeat(
     data: web::Data<State>,
     _: Authorization,
 ) -> Result<String> {
-    log::info!("Heartbeat from worker {}", heartbeat.id);
+    log::debug!("Heartbeat from worker {}", heartbeat.id);
     let mut workers = data.workers.lock().unwrap();
     for w in workers.iter_mut() {
         if w.id == heartbeat.id {
@@ -190,11 +190,23 @@ async fn main() -> std::io::Result<()> {
     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,
-            })
+            {
+                let mut workers = s.workers.lock().unwrap();
+                workers.retain(|w| match w.last_heartbeat {
+                    None => true,
+                    Some(t) => Utc::now().timestamp() - t < 60,
+                })
+            }
+            let workers = s.workers.lock().unwrap();
+            let mut jobs = s.jobs.lock().unwrap();
+            for job in jobs.iter_mut() {
+                if let Status::Running(w) = &job.status {
+                    let exists = workers.iter().filter(|x| &x.id == w).count() > 0;
+                    if !exists {
+                        job.status = Status::Failed;
+                    }
+                }
+            }
         }
     });
 
@@ -211,7 +223,7 @@ async fn main() -> std::io::Result<()> {
             .service(heartbeat)
             .service(submit)
     })
-    .bind(("127.0.0.1", 8080))?
+    .bind(("0.0.0.0", 8080))?
     .run()
     .await
 }

zoidberg_server/src/webpage.rs

@@ -52,7 +52,8 @@ pub fn render(jobs: &[Job], workers: &[Worker]) -> String {
     <meta charset="utf-8">
     <meta name="viewport" content="width=device-width, initial-scale=1">
     <title>Zoidberg</title>
-    <link rel="icon" href="data:,">
+    <link rel="icon" href="data:image/svg+xml,%3Csvg%20xmlns='http://www.w3.org/2000/svg'%20viewBox='0%200%2016%2016'%3E%3Ctext%20x='0'%20y='14'%3E🦀%3C/text%3E%3C/svg%3E" type="image/svg+xml" />
+
     <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/bulma@0.9.4/css/bulma.min.css">
     {}
   </head>