mirrors/pict-rs
1
0
Fork 0
mirror of https://git.asonix.dog/asonix/pict-rs.git synced 2024-12-29 05:40:32 +00:00
Code Issues Projects Releases Packages Wiki Activity

Finish queue implementation update for sled

Browse source
This commit is contained in:
asonix 2023-08-13 19:47:20 -05:00
parent c48b7bcdfe
commit 5d4486fdf4
13 changed files with 154 additions and 156 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.cargo/config
View file

@ -1,2 +1,2 @@
[build]
rustflags = ["--cfg", "tokio_unstable"]
rustflags = ["--cfg", "tokio_unstable", "--cfg", "uuid_unstable"]

7
Cargo.lock generated
View file

@ -320,6 +320,12 @@ dependencies = [
"syn 2.0.28",
]
[[package]]
name = "atomic"
version = "0.5.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c59bdb34bc650a32731b31bd8f0829cc15d24a708ee31559e0bb34f2bc320cba"
[[package]]
name = "autocfg"
version = "1.1.0"
@ -2969,6 +2975,7 @@ version = "1.4.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "79daa5ed5740825c40b389c5e50312b9c86df53fccd33f281df655642b43869d"
dependencies = [
"atomic",
"getrandom",
"serde",
]

2
Cargo.toml
View file

@ -74,7 +74,7 @@ tracing-subscriber = { version = "0.3.0", features = [
"tracing-log",
] }
url = { version = "2.2", features = ["serde"] }
uuid = { version = "1", features = ["v4", "serde"] }
uuid = { version = "1", features = ["serde", "std", "v4", "v7"] }
[dependencies.tracing-actix-web]
version = "0.7.5"

1
defaults.toml
View file

@ -1,6 +1,5 @@
[server]
address = "0.0.0.0:8080"
worker_id = "pict-rs-1"
read_only = false
max_file_count = 1

1
dev.toml
View file

@ -1,6 +1,5 @@
[server]
address = '0.0.0.0:8080'
worker_id = 'pict-rs-1'
api_key = 'api-key'
max_file_count = 10

8
pict-rs.toml
View file

@ -5,14 +5,6 @@
# default: 0.0.0.0:8080
address = '0.0.0.0:8080'
## Optional: pict-rs worker id
# environment variable PICTRS__SERVER__WORKER_ID
# default: pict-rs-1
#
# This is used for the internal job queue. It will have more meaning once a shared metadata
# repository (like postgres) can be defined.
worker_id = 'pict-rs-1'
## Optional: shared secret for internal endpoints
# environment variable: PICTRS__SERVER__API_KEY
# default: empty

8
src/config/commandline.rs
View file

@ -47,7 +47,6 @@ impl Args {
Command::Run(Run {
address,
api_key,
worker_id,
client_pool_size,
client_timeout,
metrics_prometheus_address,
@ -100,7 +99,6 @@ impl Args {
let server = Server {
address,
api_key,
worker_id,
read_only,
max_file_count,
};
@ -386,8 +384,6 @@ struct Server {
#[serde(skip_serializing_if = "Option::is_none")]
address: Option<SocketAddr>,
#[serde(skip_serializing_if = "Option::is_none")]
worker_id: Option<String>,
#[serde(skip_serializing_if = "Option::is_none")]
api_key: Option<String>,
#[serde(skip_serializing_if = "std::ops::Not::not")]
read_only: bool,
@ -763,10 +759,6 @@ struct Run {
#[arg(long)]
api_key: Option<String>,
/// ID of this pict-rs node. Doesn't do much yet
#[arg(long)]
worker_id: Option<String>,
/// Number of connections the internel HTTP client should maintain in its pool
///
/// This number defaults to 100, and the total number is multiplied by the number of cores

2
src/config/defaults.rs
View file

@ -21,7 +21,6 @@ pub(crate) struct Defaults {
#[serde(rename_all = "snake_case")]
struct ServerDefaults {
address: SocketAddr,
worker_id: String,
read_only: bool,
max_file_count: u32,
}
@ -183,7 +182,6 @@ impl Default for ServerDefaults {
fn default() -> Self {
ServerDefaults {
address: "0.0.0.0:8080".parse().expect("Valid address string"),
worker_id: String::from("pict-rs-1"),
read_only: false,
max_file_count: 1,
}

2
src/config/file.rs
View file

@ -95,8 +95,6 @@ pub(crate) enum Repo {
pub(crate) struct Server {
pub(crate) address: SocketAddr,
pub(crate) worker_id: String,
#[serde(skip_serializing_if = "Option::is_none")]
pub(crate) api_key: Option<String>,

24
src/lib.rs
View file

@ -48,7 +48,6 @@ use std::{
future::ready,
path::Path,
path::PathBuf,
sync::atomic::{AtomicU64, Ordering},
time::{Duration, SystemTime},
};
use tokio::sync::Semaphore;
@ -1495,14 +1494,6 @@ fn build_client(config: &Configuration) -> Result<ClientWithMiddleware, Error> {
.build())
}
fn next_worker_id(config: &Configuration) -> String {
static WORKER_ID: AtomicU64 = AtomicU64::new(0);
let next_id = WORKER_ID.fetch_add(1, Ordering::Relaxed);
format!("{}-{}", config.server.worker_id, next_id)
}
fn configure_endpoints<
R: FullRepo + 'static,
S: Store + 'static,
@ -1638,26 +1629,15 @@ where
R: FullRepo + 'static,
S: Store + 'static,
{
let worker_id_1 = next_worker_id(&config);
let worker_id_2 = next_worker_id(&config);
tracing::trace_span!(parent: None, "Spawn task").in_scope(|| {
actix_rt::spawn(queue::process_cleanup(
repo.clone(),
store.clone(),
config.clone(),
worker_id_1,
))
});
tracing::trace_span!(parent: None, "Spawn task").in_scope(|| {
actix_rt::spawn(queue::process_images(
repo,
store,
process_map,
config,
worker_id_2,
))
});
tracing::trace_span!(parent: None, "Spawn task")
.in_scope(|| actix_rt::spawn(queue::process_images(repo, store, process_map, config)));
}
async fn launch_file_store<R: FullRepo + 'static, F: Fn(&mut web::ServiceConfig) + Send + Clone>(

107
src/queue.rs
View file

@ -187,17 +187,8 @@ pub(crate) async fn process_cleanup<R: FullRepo, S: Store>(
repo: R,
store: S,
config: Configuration,
worker_id: String,
) {
process_jobs(
&repo,
&store,
&config,
worker_id,
CLEANUP_QUEUE,
cleanup::perform,
)
.await
process_jobs(&repo, &store, &config, CLEANUP_QUEUE, cleanup::perform).await
}
pub(crate) async fn process_images<R: FullRepo + 'static, S: Store + 'static>(
@ -205,14 +196,12 @@ pub(crate) async fn process_images<R: FullRepo + 'static, S: Store + 'static>(
store: S,
process_map: ProcessMap,
config: Configuration,
worker_id: String,
) {
process_image_jobs(
&repo,
&store,
&process_map,
&config,
worker_id,
PROCESS_QUEUE,
process::perform,
)
@ -225,7 +214,6 @@ async fn process_jobs<R, S, F>(
repo: &R,
store: &S,
config: &Configuration,
worker_id: String,
queue: &'static str,
callback: F,
) where
@ -235,8 +223,10 @@ async fn process_jobs<R, S, F>(
for<'a> F: Fn(&'a R, &'a S, &'a Configuration, &'a [u8]) -> LocalBoxFuture<'a, Result<(), Error>>
+ Copy,
{
let worker_id = uuid::Uuid::new_v4();
loop {
let res = job_loop(repo, store, config, worker_id.clone(), queue, callback).await;
let res = job_loop(repo, store, config, worker_id, queue, callback).await;
if let Err(e) = res {
tracing::warn!("Error processing jobs: {}", format!("{e}"));
@ -249,15 +239,15 @@ async fn process_jobs<R, S, F>(
}
struct MetricsGuard {
worker_id: String,
worker_id: uuid::Uuid,
queue: &'static str,
start: Instant,
armed: bool,
}
impl MetricsGuard {
fn guard(worker_id: String, queue: &'static str) -> Self {
metrics::increment_counter!("pict-rs.job.start", "queue" => queue, "worker-id" => worker_id.clone());
fn guard(worker_id: uuid::Uuid, queue: &'static str) -> Self {
metrics::increment_counter!("pict-rs.job.start", "queue" => queue, "worker-id" => worker_id.to_string());
Self {
worker_id,
@ -274,8 +264,8 @@ impl MetricsGuard {
impl Drop for MetricsGuard {
fn drop(&mut self) {
metrics::histogram!("pict-rs.job.duration", self.start.elapsed().as_secs_f64(), "queue" => self.queue, "worker-id" => self.worker_id.clone(), "completed" => (!self.armed).to_string());
metrics::increment_counter!("pict-rs.job.end", "queue" => self.queue, "worker-id" => self.worker_id.clone(), "completed" => (!self.armed).to_string());
metrics::histogram!("pict-rs.job.duration", self.start.elapsed().as_secs_f64(), "queue" => self.queue, "worker-id" => self.worker_id.to_string(), "completed" => (!self.armed).to_string());
metrics::increment_counter!("pict-rs.job.end", "queue" => self.queue, "worker-id" => self.worker_id.to_string(), "completed" => (!self.armed).to_string());
}
}
@ -283,7 +273,7 @@ async fn job_loop<R, S, F>(
repo: &R,
store: &S,
config: &Configuration,
worker_id: String,
worker_id: uuid::Uuid,
queue: &'static str,
callback: F,
) -> Result<(), Error>
@ -295,21 +285,27 @@ where
+ Copy,
{
loop {
let (job_id, bytes) = repo.pop(queue, worker_id.as_bytes().to_vec()).await?;
let (job_id, bytes) = repo.pop(queue).await?;
let span = tracing::info_span!("Running Job", worker_id = ?worker_id);
let guard = MetricsGuard::guard(worker_id.clone(), queue);
let guard = MetricsGuard::guard(worker_id, queue);
span.in_scope(|| {
heartbeat(
repo,
job_id,
(callback)(repo, store, config, bytes.as_ref()),
)
})
.instrument(span)
.await?;
let res = span
.in_scope(|| {
heartbeat(
repo,
queue,
job_id,
(callback)(repo, store, config, bytes.as_ref()),
)
})
.instrument(span)
.await;
repo.complete_job(queue, job_id).await?;
res?;
guard.disarm();
}
@ -320,7 +316,6 @@ async fn process_image_jobs<R, S, F>(
store: &S,
process_map: &ProcessMap,
config: &Configuration,
worker_id: String,
queue: &'static str,
callback: F,
) where
@ -336,17 +331,11 @@ async fn process_image_jobs<R, S, F>(
) -> LocalBoxFuture<'a, Result<(), Error>>
+ Copy,
{
let worker_id = uuid::Uuid::new_v4();
loop {
let res = image_job_loop(
repo,
store,
process_map,
config,
worker_id.clone(),
queue,
callback,
)
.await;
let res =
image_job_loop(repo, store, process_map, config, worker_id, queue, callback).await;
if let Err(e) = res {
tracing::warn!("Error processing jobs: {}", format!("{e}"));
@ -363,7 +352,7 @@ async fn image_job_loop<R, S, F>(
store: &S,
process_map: &ProcessMap,
config: &Configuration,
worker_id: String,
worker_id: uuid::Uuid,
queue: &'static str,
callback: F,
) -> Result<(), Error>
@ -381,27 +370,33 @@ where
+ Copy,
{
loop {
let (job_id, bytes) = repo.pop(queue, worker_id.as_bytes().to_vec()).await?;
let (job_id, bytes) = repo.pop(queue).await?;
let span = tracing::info_span!("Running Job", worker_id = ?worker_id);
let guard = MetricsGuard::guard(worker_id.clone(), queue);
let guard = MetricsGuard::guard(worker_id, queue);
span.in_scope(|| {
heartbeat(
repo,
job_id,
(callback)(repo, store, process_map, config, bytes.as_ref()),
)
})
.instrument(span)
.await?;
let res = span
.in_scope(|| {
heartbeat(
repo,
queue,
job_id,
(callback)(repo, store, process_map, config, bytes.as_ref()),
)
})
.instrument(span)
.await;
repo.complete_job(queue, job_id).await?;
res?;
guard.disarm();
}
}
async fn heartbeat<R, Fut>(repo: &R, job_id: JobId, fut: Fut) -> Fut::Output
async fn heartbeat<R, Fut>(repo: &R, queue: &'static str, job_id: JobId, fut: Fut) -> Fut::Output
where
R: QueueRepo,
Fut: std::future::Future,
@ -419,7 +414,7 @@ where
}
_ = interval.tick() => {
if hb.is_none() {
hb = Some(repo.heartbeat(job_id));
hb = Some(repo.heartbeat(queue, job_id));
}
}
opt = poll_opt(hb.as_mut()), if hb.is_some() => {

34
src/repo.rs
View file

@ -289,7 +289,7 @@ pub(crate) struct JobId(Uuid);
impl JobId {
pub(crate) fn gen() -> Self {
Self(Uuid::new_v4())
Self(Uuid::now_v7())
}
pub(crate) const fn as_bytes(&self) -> &[u8; 16] {
@ -303,19 +303,13 @@ impl JobId {
#[async_trait::async_trait(?Send)]
pub(crate) trait QueueRepo: BaseRepo {
async fn requeue_timed_out(&self, worker_prefix: Vec<u8>) -> Result<(), RepoError>;
async fn push(&self, queue: &'static str, job: Self::Bytes) -> Result<(), RepoError>;
async fn pop(
&self,
queue: &'static str,
worker_id: Vec<u8>,
) -> Result<(JobId, Self::Bytes), RepoError>;
async fn pop(&self, queue: &'static str) -> Result<(JobId, Self::Bytes), RepoError>;
async fn heartbeat(&self, job_id: JobId) -> Result<(), RepoError>;
async fn heartbeat(&self, queue: &'static str, job_id: JobId) -> Result<(), RepoError>;
async fn complete_job(&self, job_id: JobId) -> Result<(), RepoError>;
async fn complete_job(&self, queue: &'static str, job_id: JobId) -> Result<(), RepoError>;
}
#[async_trait::async_trait(?Send)]
@ -323,28 +317,20 @@ impl<T> QueueRepo for actix_web::web::Data<T>
where
T: QueueRepo,
{
async fn requeue_timed_out(&self, worker_prefix: Vec<u8>) -> Result<(), RepoError> {
T::requeue_timed_out(self, worker_prefix).await
}
async fn push(&self, queue: &'static str, job: Self::Bytes) -> Result<(), RepoError> {
T::push(self, queue, job).await
}
async fn pop(
&self,
queue: &'static str,
worker_id: Vec<u8>,
) -> Result<(JobId, Self::Bytes), RepoError> {
T::pop(self, queue, worker_id).await
async fn pop(&self, queue: &'static str) -> Result<(JobId, Self::Bytes), RepoError> {
T::pop(self, queue).await
}
async fn heartbeat(&self, job_id: JobId) -> Result<(), RepoError> {
T::heartbeat(self, job_id).await
async fn heartbeat(&self, queue: &'static str, job_id: JobId) -> Result<(), RepoError> {
T::heartbeat(self, queue, job_id).await
}
async fn complete_job(&self, job_id: JobId) -> Result<(), RepoError> {
T::complete_job(self, job_id).await
async fn complete_job(&self, queue: &'static str, job_id: JobId) -> Result<(), RepoError> {
T::complete_job(self, queue, job_id).await
}
}

112
src/repo/sled.rs
View file

@ -57,6 +57,9 @@ pub(crate) enum SledError {
#[error("Operation panicked")]
Panic,
#[error("Another process updated this value before us")]
Conflict,
}
#[derive(Clone)]
@ -129,7 +132,7 @@ impl SledRepo {
}
fn open(mut path: PathBuf, cache_capacity: u64) -> Result<Db, SledError> {
path.push("v0.4.0-alpha.1");
path.push("v0.5.0");
let db = ::sled::Config::new()
.cache_capacity(cache_capacity)
@ -626,7 +629,7 @@ impl UploadRepo for SledRepo {
enum JobState {
Pending,
Running(Vec<u8>),
Running([u8; 8]),
}
impl JobState {
@ -637,35 +640,34 @@ impl JobState {
fn running() -> Self {
Self::Running(
time::OffsetDateTime::now_utc()
.format(&time::format_description::well_known::Rfc3339)
.expect("Can format")
.into_bytes(),
.unix_timestamp()
.to_be_bytes(),
)
}
fn as_bytes(&self) -> &[u8] {
match self {
Self::Pending => b"pending",
Self::Pending => b"pend",
Self::Running(ref bytes) => bytes,
}
}
}
fn job_key(queue: &'static str, job_id: JobId) -> Arc<[u8]> {
let mut key = queue.as_bytes().to_vec();
key.extend(job_id.as_bytes());
Arc::from(key)
}
#[async_trait::async_trait(?Send)]
impl QueueRepo for SledRepo {
#[tracing::instrument(skip_all, fields(worker_id = %String::from_utf8_lossy(&worker_prefix)))]
async fn requeue_timed_out(&self, worker_prefix: Vec<u8>) -> Result<(), RepoError> {
todo!()
}
#[tracing::instrument(skip(self, job), fields(job = %String::from_utf8_lossy(&job)))]
async fn push(&self, queue_name: &'static str, job: Self::Bytes) -> Result<(), RepoError> {
let metrics_guard = PushMetricsGuard::guard(queue_name);
let id = JobId::gen();
let mut key = queue_name.as_bytes().to_vec();
key.push(0);
key.extend(id.as_bytes());
let key = job_key(queue_name, id);
let queue = self.queue.clone();
let job_state = self.job_state.clone();
@ -674,8 +676,8 @@ impl QueueRepo for SledRepo {
(&queue, &job_state).transaction(|(queue, job_state)| {
let state = JobState::pending();
queue.insert(key.as_slice(), &job)?;
job_state.insert(key.as_slice(), state.as_bytes())?;
queue.insert(&key[..], &job)?;
job_state.insert(&key[..], state.as_bytes())?;
Ok(())
})
@ -705,25 +707,35 @@ impl QueueRepo for SledRepo {
Ok(())
}
#[tracing::instrument(skip(self, worker_id), fields(worker_id = %String::from_utf8_lossy(&worker_id)))]
async fn pop(
&self,
queue_name: &'static str,
worker_id: Vec<u8>,
) -> Result<(JobId, Self::Bytes), RepoError> {
#[tracing::instrument(skip(self))]
async fn pop(&self, queue_name: &'static str) -> Result<(JobId, Self::Bytes), RepoError> {
let metrics_guard = PopMetricsGuard::guard(queue_name);
let now = time::OffsetDateTime::now_utc();
loop {
let queue = self.queue.clone();
let job_state = self.job_state.clone();
let opt = actix_rt::task::spawn_blocking(move || {
// Job IDs are generated with Uuid version 7 - defining their first bits as a
// timestamp. Scanning a prefix should give us jobs in the order they were queued.
for res in job_state.scan_prefix(queue_name) {
let (key, value) = res?;
if value != "pending" {
// TODO: requeue dead jobs
continue;
if value.len() == 8 {
let unix_timestamp =
i64::from_be_bytes(value[0..8].try_into().expect("Verified length"));
let timestamp = time::OffsetDateTime::from_unix_timestamp(unix_timestamp)
.expect("Valid timestamp");
// heartbeats should update every 5 seconds, so 30 seconds without an
// update is 6 missed beats
if timestamp.saturating_add(time::Duration::seconds(30)) > now {
// job hasn't expired
continue;
}
}
let state = JobState::running();
@ -738,7 +750,7 @@ impl QueueRepo for SledRepo {
}
}
let id_bytes = &key[queue_name.len() + 1..];
let id_bytes = &key[queue_name.len()..];
let id_bytes: [u8; 16] = id_bytes.try_into().expect("Key length");
@ -780,12 +792,52 @@ impl QueueRepo for SledRepo {
}
}
async fn heartbeat(&self, job_id: JobId) -> Result<(), RepoError> {
todo!()
#[tracing::instrument(skip(self))]
async fn heartbeat(&self, queue_name: &'static str, job_id: JobId) -> Result<(), RepoError> {
let key = job_key(queue_name, job_id);
let job_state = self.job_state.clone();
actix_rt::task::spawn_blocking(move || {
if let Some(state) = job_state.get(&key)? {
let new_state = JobState::running();
match job_state.compare_and_swap(&key, Some(state), Some(new_state.as_bytes()))? {
Ok(_) => Ok(()),
Err(_) => Err(SledError::Conflict),
}
} else {
Ok(())
}
})
.await
.map_err(|_| RepoError::Canceled)??;
Ok(())
}
async fn complete_job(&self, job_id: JobId) -> Result<(), RepoError> {
todo!()
#[tracing::instrument(skip(self))]
async fn complete_job(&self, queue_name: &'static str, job_id: JobId) -> Result<(), RepoError> {
let key = job_key(queue_name, job_id);
let queue = self.queue.clone();
let job_state = self.job_state.clone();
let res = actix_rt::task::spawn_blocking(move || {
(&queue, &job_state).transaction(|(queue, job_state)| {
queue.remove(&key[..])?;
job_state.remove(&key[..])?;
Ok(())
})
})
.await
.map_err(|_| RepoError::Canceled)?;
if let Err(TransactionError::Abort(e) | TransactionError::Storage(e)) = res {
return Err(RepoError::from(SledError::from(e)));
}
Ok(())
}
}