Hook things together

jobs-core/Cargo.toml

@@ -8,7 +8,8 @@ edition = "2018"
 failure = "0.1"
 futures = "0.1.21"
 log = "0.4"
-rkv = "0.5"
+kv = "0.6"
+lmdb = "0.8"
 serde = "1.0"
 serde_derive = "1.0"
 serde_json = "1.0"

jobs-core/src/lib.rs

@@ -5,13 +5,15 @@ extern crate log;
 #[macro_use]
 extern crate serde_derive;
 
-use std::collections::{HashMap, VecDeque};
+use std::collections::HashMap;
 
 use failure::Error;
 use futures::future::{Either, Future, IntoFuture};
 use serde::{de::DeserializeOwned, ser::Serialize};
 use serde_json::Value;
 
+pub mod storage;
+
 #[derive(Debug, Fail)]
 pub enum JobError {
     #[fail(display = "Error performing job: {}", _0)]
@@ -92,6 +94,7 @@ pub trait Processor {
             status: JobStatus::Pending,
             args: serde_json::to_value(args)?,
             retry_count: max_retries.unwrap_or(Self::max_retries()),
+            requeued_at: None,
         };
 
         Ok(job)
@@ -136,13 +139,13 @@ pub enum MaxRetries {
 }
 
 #[derive(Clone, Debug, Eq, PartialEq)]
-enum ShouldStop {
+pub enum ShouldStop {
     LimitReached,
     Requeue,
 }
 
 impl ShouldStop {
-    fn should_requeue(&self) -> bool {
+    pub fn should_requeue(&self) -> bool {
         *self == ShouldStop::Requeue
     }
 }
@@ -164,7 +167,7 @@ impl MaxRetries {
     }
 }
 
-#[derive(Deserialize, Serialize)]
+#[derive(Clone, Debug, Deserialize, Serialize)]
 pub struct JobInfo {
     /// ID of the job, None means an ID has not been set
     id: Option<usize>,
@@ -180,10 +183,13 @@ pub struct JobInfo {
 
     /// Retries left for this job, None means no limit
     retry_count: MaxRetries,
+
+    /// The time this job was re-queued
+    requeued_at: Option<usize>,
 }
 
 impl JobInfo {
-    fn decrement(&mut self) -> ShouldStop {
+    pub fn decrement(&mut self) -> ShouldStop {
         self.retry_count.decrement()
     }
 
@@ -198,35 +204,11 @@ impl JobInfo {
     }
 }
 
-#[derive(Deserialize, Serialize)]
-pub struct Jobs {
-    inner: VecDeque<JobInfo>,
-}
-
-impl Jobs {
-    fn queue(&mut self, job: JobInfo) {
-        self.inner.push_back(job);
-    }
-
-    fn pop(&mut self) -> Option<JobInfo> {
-        self.inner.pop_front()
-    }
-}
-
-impl Default for Jobs {
-    fn default() -> Self {
-        Jobs {
-            inner: Default::default(),
-        }
-    }
-}
-
 pub type ProcessFn =
     Box<dyn Fn(Value) -> Box<dyn Future<Item = (), Error = JobError> + Send> + Send>;
 
 pub struct Processors {
     inner: HashMap<String, ProcessFn>,
-    jobs: Jobs,
 }
 
 impl Processors {
@@ -244,67 +226,42 @@ impl Processors {
         );
     }
 
-    pub fn queue(&mut self, job: JobInfo) {
-        self.jobs.queue(job);
-    }
+    pub fn process_job(&self, job: JobInfo) -> impl Future<Item = JobInfo, Error = ()> {
+        let opt = self
+            .inner
+            .get(&job.processor)
+            .map(|processor| process(processor, job.clone()));
 
-    pub fn turn(mut self) -> impl Future<Item = Self, Error = ()> {
-        match self.jobs.pop() {
-            Some(job) => Either::A(self.process_job(job)),
-            None => Either::B(Ok(self).into_future()),
+        if let Some(fut) = opt {
+            Either::A(fut)
+        } else {
+            error!("Processor {} not present", job.processor);
+            Either::B(Ok(job).into_future())
         }
     }
-
-    fn process_job(mut self, job: JobInfo) -> impl Future<Item = Self, Error = ()> {
-        let processor = self.inner.remove(&job.processor);
-
-        processor
-            .ok_or_else(|| {
-                error!("No processor");
-                ()
-            })
-            .into_future()
-            .and_then(move |processor| process(self, processor, job))
-    }
 }
 
 impl Default for Processors {
     fn default() -> Self {
         Processors {
             inner: Default::default(),
-            jobs: Default::default(),
         }
     }
 }
 
-fn process(
-    mut processors: Processors,
-    process_fn: ProcessFn,
-    mut job: JobInfo,
-) -> impl Future<Item = Processors, Error = ()> {
+fn process(process_fn: &ProcessFn, job: JobInfo) -> impl Future<Item = JobInfo, Error = ()> {
     let args = job.args.clone();
 
     let processor = job.processor.clone();
 
-    let fut = process_fn(args).then(move |res| match res {
-        Ok(_) => Ok(info!("Job completed, {}", processor)),
+    process_fn(args).then(move |res| match res {
+        Ok(_) => {
+            info!("Job completed, {}", processor);
+            Ok(job)
+        }
         Err(e) => {
             error!("Job errored, {}, {}", processor, e);
-            Err(e)
+            Ok(job)
         }
-    });
-
-    processors.inner.insert(job.processor.clone(), process_fn);
-
-    fut.then(|res| {
-        if let Err(e) = res {
-            if job.decrement().should_requeue() {
-                processors.jobs.queue(job);
-            } else {
-                error!("Job failed permanently, {}, {}", &job.processor, e);
-            }
-        }
-
-        Ok(processors)
     })
 }

jobs-core/src/storage.rs

@@ -1,10 +1,10 @@
 use std::{
     path::PathBuf,
-    sync::{Arc, RwLock},
+    sync::{Arc, RwLock, RwLockWriteGuard},
 };
 
-use jobs_core::JobInfo;
-use kv::{bincode::Bincode, Config, Error, Manager, Serde, Store, ValueBuf};
+use crate::JobInfo;
+use kv::{bincode::Bincode, Bucket, Config, CursorOp, Error, Manager, Serde, Store, Txn, ValueBuf};
 use lmdb::Error as LmdbError;
 
 /// Set the status of a job when storing it
@@ -17,8 +17,12 @@ pub enum JobStatus {
 
     /// Job has failed
     Failed,
+
+    /// Job has finished
+    Finished,
 }
 
+#[derive(Clone)]
 pub struct Storage {
     runner_id: usize,
     store: Arc<RwLock<Store>>,
@@ -48,41 +52,7 @@ impl Storage {
 
         let bucket = store.bucket::<&str, ValueBuf<Bincode<usize>>>(Some(Storage::id_store()))?;
 
-        let mut txn = store.write_txn()?;
-
-        let mut other_runner_id = 0;
-
-        loop {
-            let lock_value = Bincode::to_value_buf(self.runner_id)?;
-            match txn.set_no_overwrite(&bucket, "id-lock", lock_value) {
-                Ok(_) => break,
-                Err(e) => {
-                    match txn.get(&bucket, "id-lock") {
-                        Ok(other_id) => {
-                            let other_id = other_id.inner()?.to_serde();
-
-                            if other_runner_id != other_id {
-                                other_runner_id = other_id;
-                                info!("Id lock held by runner {}", other_id);
-                            }
-                        }
-                        Err(e) => match e {
-                            Error::NotFound => continue,
-                            e => return Err(e),
-                        },
-                    }
-
-                    match e {
-                        Error::LMDB(lmdb) => match lmdb {
-                            LmdbError::KeyExist => continue,
-                            e => return Err(Error::LMDB(e)),
-                        },
-                        e => return Err(e),
-                    }
-                }
-            }
-        }
-
+        let new_id = self.with_lock(&store, &bucket, "id-lock", |txn| {
             let id = match txn.get(&bucket, "current-id") {
                 Ok(id) => id.inner()?.to_serde(),
                 Err(e) => match e {
@@ -96,12 +66,59 @@ impl Storage {
             let new_id_value = Bincode::to_value_buf(new_id)?;
             txn.set(&bucket, "current-id", new_id_value)?;
 
-        txn.del(&bucket, "id-lock")?;
-        txn.commit()?;
+            Ok(new_id)
+        })?;
 
         Ok(new_id)
     }
 
+    pub fn dequeue_job(&self) -> Result<Option<JobInfo>, Error> {
+        let store = self.store.write()?;
+
+        let job_bucket =
+            store.bucket::<&[u8], ValueBuf<Bincode<JobInfo>>>(Some(Storage::job_queue()))?;
+
+        let txn = store.read_txn()?;
+
+        let maybe_job = {
+            let cursor = txn.read_cursor(&job_bucket)?;
+            match cursor.get(None, CursorOp::Current) {
+                Ok((k, v)) => {
+                    let v = v.inner()?.to_serde();
+                    Some((k, v))
+                }
+                Err(e) => match e {
+                    Error::NotFound => None,
+                    e => return Err(e),
+                },
+            }
+        };
+
+        let lock_bucket =
+            store.bucket::<&str, ValueBuf<Bincode<usize>>>(Some(Storage::job_lock()))?;
+
+        let result = if let Some((maybe_key, job)) = maybe_job {
+            if let Some(key) = maybe_key {
+                let job_key = std::str::from_utf8(&key).unwrap();
+
+                self.with_lock(&store, &lock_bucket, job_key, move |_| {
+                    self.run_job(job_key.parse().unwrap())
+                })
+                .map(|_| Some(job))
+            } else {
+                warn!("Didn't get key from cursor");
+                Ok(None)
+            }
+        } else {
+            info!("No jobs queued");
+            Ok(None)
+        };
+
+        txn.commit()?;
+
+        result
+    }
+
     pub fn store_job(&self, mut job: JobInfo, status: JobStatus) -> Result<(), Error> {
         let job_id = match job.id() {
             Some(id) => id,
@@ -128,6 +145,7 @@ impl Storage {
             JobStatus::Pending => self.queue_job(job_id)?,
             JobStatus::Running => self.run_job(job_id)?,
             JobStatus::Failed => self.fail_job(job_id)?,
+            JobStatus::Finished => self.finish_job(job_id)?,
         }
 
         Ok(())
@@ -137,6 +155,7 @@ impl Storage {
         self.add_job_to(id, Storage::job_queue())?;
         self.delete_job_from(id, Storage::job_failed())?;
         self.delete_job_from(id, Storage::job_running())?;
+        self.delete_job_from(id, Storage::job_finished())?;
 
         Ok(())
     }
@@ -145,6 +164,7 @@ impl Storage {
         self.add_job_to(id, Storage::job_failed())?;
         self.delete_job_from(id, Storage::job_queue())?;
         self.delete_job_from(id, Storage::job_running())?;
+        self.delete_job_from(id, Storage::job_finished())?;
 
         Ok(())
     }
@@ -153,6 +173,16 @@ impl Storage {
         self.add_job_to(id, Storage::job_running())?;
         self.delete_job_from(id, Storage::job_queue())?;
         self.delete_job_from(id, Storage::job_failed())?;
+        self.delete_job_from(id, Storage::job_finished())?;
+
+        Ok(())
+    }
+
+    fn finish_job(&self, id: usize) -> Result<(), Error> {
+        self.add_job_to(id, Storage::job_finished())?;
+        self.delete_job_from(id, Storage::job_running())?;
+        self.delete_job_from(id, Storage::job_queue())?;
+        self.delete_job_from(id, Storage::job_failed())?;
 
         Ok(())
     }
@@ -191,13 +221,80 @@ impl Storage {
         Ok(())
     }
 
-    fn buckets() -> [&'static str; 5] {
+    // In all likelihood, this function is not necessary
+    //
+    // But in the event of multiple processes running on the same machine, it is good to have some
+    // way to make sure they don't step on eachother's toes
+    fn with_lock<F, T>(
+        &self,
+        store: &RwLockWriteGuard<Store>,
+        bucket: &Bucket<&str, ValueBuf<Bincode<usize>>>,
+        lock_key: &str,
+        callback: F,
+    ) -> Result<T, Error>
+    where
+        F: Fn(&mut Txn) -> Result<T, Error>,
+    {
+        let mut txn = store.write_txn()?;
+        let mut other_runner_id = 0;
+
+        loop {
+            let lock_value = Bincode::to_value_buf(self.runner_id)?;
+
+            let mut inner_txn = txn.txn()?;
+            let res = inner_txn.set_no_overwrite(bucket, lock_key, lock_value);
+            inner_txn.commit()?;
+
+            match res {
+                Ok(_) => break,
+                Err(e) => {
+                    let inner_txn = txn.txn()?;
+                    let res = inner_txn.get(bucket, lock_key);
+                    inner_txn.commit()?;
+
+                    match res {
+                        Ok(other_id) => {
+                            let other_id = other_id.inner()?.to_serde();
+
+                            if other_runner_id != other_id {
+                                other_runner_id = other_id;
+                                info!("Lock held by runner {}", other_id);
+                            }
+                        }
+                        Err(e) => match e {
+                            Error::NotFound => continue,
+                            e => return Err(e),
+                        },
+                    }
+
+                    match e {
+                        Error::LMDB(lmdb) => match lmdb {
+                            LmdbError::KeyExist => continue,
+                            e => return Err(Error::LMDB(e)),
+                        },
+                        e => return Err(e),
+                    }
+                }
+            }
+        }
+
+        let item = callback(&mut txn)?;
+
+        txn.del(bucket, lock_key)?;
+        txn.commit()?;
+
+        Ok(item)
+    }
+
+    fn buckets() -> [&'static str; 7] {
         [
             Storage::id_store(),
             Storage::job_store(),
             Storage::job_queue(),
             Storage::job_failed(),
             Storage::job_running(),
+            Storage::job_lock(),
+            Storage::job_finished(),
         ]
     }
 
@@ -220,4 +317,12 @@ impl Storage {
     fn job_running() -> &'static str {
         "job-running"
     }
+
+    fn job_finished() -> &'static str {
+        "job-finished"
+    }
+
+    fn job_lock() -> &'static str {
+        "job-lock"
+    }
 }

jobs-tokio/Cargo.toml

@@ -6,7 +6,6 @@ edition = "2018"
 
 [dependencies]
 futures = "0.1"
-lmdb = "0.8"
 log = "0.4"
 tokio = "0.1"
 tokio-threadpool = "0.1"

jobs-tokio/src/lib.rs

@@ -1,20 +1,23 @@
 #[macro_use]
 extern crate log;
 
-mod storage;
-
-use std::path::{Path, PathBuf};
+use std::{
+    path::{Path, PathBuf},
+    time::{Duration, Instant},
+};
 
 use futures::{
-    future::poll_fn,
+    future::{poll_fn, Either, IntoFuture},
     sync::mpsc::{channel, Receiver, SendError, Sender},
     Future, Sink, Stream,
 };
-use jobs_core::{JobInfo, Processor, Processors};
+use jobs_core::{
+    storage::{JobStatus, Storage},
+    JobInfo, Processor, Processors,
+};
+use tokio::timer::Interval;
 use tokio_threadpool::blocking;
 
-use crate::storage::Storage;
-
 pub struct ProcessorHandle {
     spawner: Sender<JobInfo>,
 }
@@ -28,7 +31,7 @@ impl ProcessorHandle {
     }
 }
 
-fn setup_kv(db_path: PathBuf) -> impl Future<Item = Storage, Error = ()> {
+fn setup_kv(runner_id: usize, db_path: PathBuf) -> impl Future<Item = Storage, Error = ()> {
     tokio::fs::create_dir_all(db_path.clone())
         .map_err(|e| error!("Failed to create db directory: {}", e))
         .and_then(move |_| {
@@ -36,7 +39,8 @@ fn setup_kv(db_path: PathBuf) -> impl Future<Item = Storage, Error = ()> {
                 let path = db_path.clone();
 
                 blocking(move || {
-                    Storage::init(0, path).map_err(|e| error!("Error initializing db, {}", e))
+                    Storage::init(runner_id, path)
+                        .map_err(|e| error!("Error initializing db, {}", e))
                 })
                 .map_err(|e| error!("Error in blocking, {}", e))
             })
@@ -44,15 +48,117 @@ fn setup_kv(db_path: PathBuf) -> impl Future<Item = Storage, Error = ()> {
         .and_then(|res| res)
 }
 
+enum ProcessorMessage {
+    Job(JobInfo),
+    Time(Instant),
+    Stop,
+}
+
+fn return_job(
+    storage: Storage,
+    processor_count: usize,
+    processors: Processors,
+    job: JobInfo,
+) -> impl Future<Item = (Processors, usize), Error = ()> {
+    poll_fn(move || {
+        let storage = storage.clone();
+        let job = job.clone();
+
+        blocking(move || {
+            storage
+                .store_job(job, JobStatus::Finished)
+                .map_err(|e| error!("Error finishing job, {}", e))
+        })
+        .map_err(|e| error!("Error blocking, {}", e))
+    })
+    .and_then(|res| res)
+    .map(move |_| (processors, processor_count + 1))
+}
+
+fn try_process_job(
+    storage: Storage,
+    processor_count: usize,
+    processors: Processors,
+    tx: Sender<ProcessorMessage>,
+) -> impl Future<Item = (Processors, usize), Error = ()> {
+    if processor_count > 0 {
+        let fut = poll_fn(move || {
+            let storage = storage.clone();
+
+            blocking(move || {
+                storage
+                    .dequeue_job()
+                    .map_err(|e| error!("Error dequeuing job, {}", e))
+            })
+            .map_err(|e| error!("Error blocking, {}", e))
+        })
+        .and_then(|res| res)
+        .then(move |res| match res {
+            Ok(maybe_job) => {
+                if let Some(job) = maybe_job {
+                    // TODO: return JobInfo to DB with job status
+                    tokio::spawn(processors.process_job(job).and_then(move |job| {
+                        tx.send(ProcessorMessage::Job(job))
+                            .map(|_| ())
+                            .map_err(|e| error!("Error returning job, {}", e))
+                    }));
+                    Ok((processors, processor_count - 1))
+                } else {
+                    Ok((processors, processor_count))
+                }
+            }
+            Err(_) => Ok((processors, processor_count)),
+        });
+
+        Either::A(fut)
+    } else {
+        Either::B(Ok((processors, processor_count)).into_future())
+    }
+}
+
+fn process_jobs(
+    storage: Storage,
+    num_processors: usize,
+    processors: Processors,
+    tx: Sender<ProcessorMessage>,
+    rx: Receiver<ProcessorMessage>,
+) -> impl Future<Item = (), Error = ()> {
+    Interval::new(Instant::now(), Duration::from_millis(500))
+        .map(ProcessorMessage::Time)
+        .map_err(|e| error!("Error in timer, {}", e))
+        .select(rx)
+        .fold(
+            (processors, num_processors),
+            move |(processors, processor_count), msg| match msg {
+                ProcessorMessage::Job(job) => Either::A(return_job(
+                    storage.clone(),
+                    processor_count,
+                    processors,
+                    job,
+                )),
+                ProcessorMessage::Time(_) => Either::B(Either::A(try_process_job(
+                    storage.clone(),
+                    processor_count,
+                    processors,
+                    tx.clone(),
+                ))),
+                ProcessorMessage::Stop => Either::B(Either::B(Err(()).into_future())),
+            },
+        )
+        .map(|_| ())
+}
+
 pub struct JobRunner {
     processors: Processors,
     receiver: Receiver<JobInfo>,
     sender: Sender<JobInfo>,
     db_path: PathBuf,
+    num_processors: usize,
+    runner_id: usize,
 }
 
 impl JobRunner {
-    pub fn new<P: AsRef<Path>>(db_path: P) -> Self {
+    pub fn new<P: AsRef<Path>>(runner_id: usize, num_processors: usize, db_path: P) -> Self {
         let (tx, rx) = channel::<JobInfo>(100);
 
         JobRunner {
@@ -60,6 +166,8 @@ impl JobRunner {
             receiver: rx,
             sender: tx,
             db_path: db_path.as_ref().to_owned(),
+            num_processors,
+            runner_id,
         }
     }
 
@@ -74,22 +182,45 @@ impl JobRunner {
         let JobRunner {
             processors,
             receiver,
-            sender,
+            sender: _,
             db_path,
+            num_processors,
+            runner_id,
         } = self;
 
-        let _ = sender;
-        let _ = db_path;
+        let (tx, rx) = channel::<ProcessorMessage>(100);
+        let tx2 = tx.clone();
 
-        // tokio::spawn(setup_kv(db_path));
+        setup_kv(runner_id, db_path)
+            .and_then(move |storage| {
+                tokio::spawn(process_jobs(
+                    storage.clone(),
+                    num_processors,
+                    processors,
+                    tx,
+                    rx,
+                ));
 
-        receiver
-            .fold(processors, |mut processors, job| {
-                processors.queue(job);
-
-                Box::new(processors.turn())
+                receiver.fold(storage, |storage, job| {
+                    poll_fn(move || {
+                        let job = job.clone();
+                        let storage = storage.clone();
+                        blocking(|| {
+                            storage
+                                .store_job(job, JobStatus::Pending)
+                                .map_err(|e| error!("Error storing job, {}", e))
+                                .map(|_| storage)
                         })
+                        .map_err(|e| error!("Error blocking, {}", e))
+                    })
+                    .and_then(|res| res)
+                })
+            })
+            .and_then(|_| {
+                tx2.send(ProcessorMessage::Stop)
                     .map(|_| ())
+                    .map_err(|e| error!("Error shutting down processor, {}", e))
+            })
     }
 
     pub fn spawn(self) -> ProcessorHandle {