Constellation, Spacedust, Slingshot, UFOs: atproto crates and services for microcosm
handle shutdown in all tasks
author
phil
date
(Aug 1, 2025, 5:05 PM -0400)
commit
efc93039
efc93039d5beb58259ff5650fa73b7f77713b824
parent
80421fe6
80421fe66d9dc289ca8cb4083e8623da24f41c1f
+47
-12
+2
slingshot/src/error.rs
+2
slingshot/src/error.rs
+5
-1
slingshot/src/firehose_cache.rs
+5
-1
slingshot/src/firehose_cache.rs
···
10
10
.memory(64 * 2_usize.pow(20))
11
11
.with_weighter(|k: &String, v| k.len() + std::mem::size_of_val(v))
12
12
.storage(Engine::large())
13
-
.with_device_options(DirectFsDeviceOptions::new(cache_dir))
13
+
.with_device_options(
14
+
DirectFsDeviceOptions::new(cache_dir)
15
+
.with_capacity(2_usize.pow(30)) // TODO: configurable (1GB to have something)
16
+
.with_file_size(16 * 2_usize.pow(20)), // note: this does limit the max cached item size, warning jumbo records
17
+
)
14
18
.build()
15
19
.await
16
20
.map_err(|e| format!("foyer setup error: {e:?}"))?;
+17
-3
slingshot/src/identity.rs
+17
-3
slingshot/src/identity.rs
···
13
13
/// 3. DID -> handle resolution: for bidirectional handle validation and in case we want to offer this
14
14
use std::time::Duration;
15
15
use tokio::sync::Mutex;
16
+
use tokio_util::sync::CancellationToken;
16
17
17
18
use crate::error::IdentityError;
18
19
use atrium_api::{
···
175
176
.with_name("identity")
176
177
.memory(16 * 2_usize.pow(20))
177
178
.with_weighter(|k, v| std::mem::size_of_val(k) + std::mem::size_of_val(v))
178
-
.storage(Engine::large())
179
-
.with_device_options(DirectFsDeviceOptions::new(cache_dir))
179
+
.storage(Engine::small())
180
+
.with_device_options(
181
+
DirectFsDeviceOptions::new(cache_dir)
182
+
.with_capacity(2_usize.pow(30)) // TODO: configurable (1GB to have something)
183
+
.with_file_size(2_usize.pow(20)), // note: this does limit the max cached item size, warning jumbo records
184
+
)
180
185
.build()
181
186
.await?;
182
187
···
403
408
}
404
409
405
410
/// run the refresh queue consumer
406
-
pub async fn run_refresher(&self) -> Result<(), IdentityError> {
411
+
pub async fn run_refresher(&self, shutdown: CancellationToken) -> Result<(), IdentityError> {
407
412
let _guard = self
408
413
.refresher
409
414
.try_lock()
410
415
.expect("there to only be one refresher running");
411
416
loop {
417
+
if shutdown.is_cancelled() {
418
+
log::info!("identity refresher: exiting for shutdown: closing cache...");
419
+
if let Err(e) = self.cache.close().await {
420
+
log::error!("cache close errored: {e}");
421
+
} else {
422
+
log::info!("identity cache closed.")
423
+
}
424
+
return Ok(());
425
+
}
412
426
let Some(task_key) = self.peek_refresh().await else {
413
427
tokio::time::sleep(tokio::time::Duration::from_millis(100)).await;
414
428
continue;
+12
-3
slingshot/src/main.rs
+12
-3
slingshot/src/main.rs
···
89
89
.map_err(|e| format!("identity setup failed: {e:?}"))?;
90
90
log::info!("identity service ready.");
91
91
let identity_refresher = identity.clone();
92
+
let identity_shutdown = shutdown.clone();
92
93
tasks.spawn(async move {
93
-
identity_refresher.run_refresher().await?;
94
+
identity_refresher.run_refresher(identity_shutdown).await?;
94
95
Ok(())
95
96
});
96
97
···
113
114
});
114
115
115
116
let consumer_shutdown = shutdown.clone();
117
+
let consumer_cache = cache.clone();
116
118
tasks.spawn(async move {
117
119
consume(
118
120
args.jetstream,
119
121
None,
120
122
args.jetstream_no_zstd,
121
123
consumer_shutdown,
122
-
cache,
124
+
consumer_cache,
123
125
)
124
126
.await?;
125
127
Ok(())
···
133
135
}
134
136
}
135
137
138
+
tasks.spawn(async move {
139
+
cache
140
+
.close()
141
+
.await
142
+
.map_err(MainTaskError::FirehoseCacheCloseError)
143
+
});
144
+
136
145
tokio::select! {
137
146
_ = async {
138
147
while let Some(completed) = tasks.join_next().await {
139
148
log::info!("shutdown: task completed: {completed:?}");
140
149
}
141
150
} => {},
142
-
_ = tokio::time::sleep(std::time::Duration::from_secs(3)) => {
151
+
_ = tokio::time::sleep(std::time::Duration::from_secs(30)) => {
143
152
log::info!("shutdown: not all tasks completed on time. aborting...");
144
153
tasks.shutdown().await;
145
154
},
+11
-5
slingshot/src/server.rs
+11
-5
slingshot/src/server.rs
···
410
410
host: Option<String>,
411
411
acme_contact: Option<String>,
412
412
certs: Option<PathBuf>,
413
-
_shutdown: CancellationToken,
413
+
shutdown: CancellationToken,
414
414
) -> Result<(), ServerError> {
415
415
let repo = Arc::new(repo);
416
416
let api_service = OpenApiService::new(
···
452
452
}
453
453
let auto_cert = auto_cert.build().map_err(ServerError::AcmeBuildError)?;
454
454
455
-
run(TcpListener::bind("0.0.0.0:443").acme(auto_cert), app).await
455
+
run(
456
+
TcpListener::bind("0.0.0.0:443").acme(auto_cert),
457
+
app,
458
+
shutdown,
459
+
)
460
+
.await
456
461
} else {
457
-
run(TcpListener::bind("127.0.0.1:3000"), app).await
462
+
run(TcpListener::bind("127.0.0.1:3000"), app, shutdown).await
458
463
}
459
464
}
460
465
461
-
async fn run<L>(listener: L, app: Route) -> Result<(), ServerError>
466
+
async fn run<L>(listener: L, app: Route, shutdown: CancellationToken) -> Result<(), ServerError>
462
467
where
463
468
L: Listener + 'static,
464
469
{
···
472
477
.with(Tracing);
473
478
Server::new(listener)
474
479
.name("slingshot")
475
-
.run(app)
480
+
.run_with_graceful_shutdown(app, shutdown.cancelled(), None)
476
481
.await
477
482
.map_err(ServerError::ServerExited)
483
+
.inspect(|()| log::info!("server ended. goodbye."))
478
484
}