spool: add optional deadline for store operations

Under high concurrency, a backlog of store operations can build
up, especially if rocksdb isn't tuned for the workload.

In order to work together with the data_processing_timeout of the smtp
server, we'd like to be able to cancel store operations that are past
the deadline, but the spool operations are not guaranteed to be cancel
safe because both the rocksdb and the local disk implementations can
ultimately call into spawn_blocking to perform the critical portion of
the work, and that is not cancel safe.

This means that we cannot simply apply timeout_at to the store
operation: while it may appear to the caller to have been cancelled at
the appropriate time, the actual work storing the spool may be in
progress and continue to store the message to the spool, which could
lead to us thinking that we are accountable for the message when we next
restart, even though we're presumably about to tell the sender that we
did not take the message.

To deal with this situation more safely, we pass the deadline through
to the underlying operation and have it internally decide whether
it should continue.

In practice, this means that the rocksdb store operation will apply
timeout_at to the semaphore operation it does if
`limit_concurrent_stores` has been configured fr that spool.  Otherwise
the deadline is ignored.

The implementation of this is imperfect in a number of ways. A more
robust solution would be to employ something like
<https://docs.rs/cancel-safe-futures/0.1.5/cancel_safe_futures/coop_cancel/index.html>,
but that is a more invasive change and one that is a bit difficult to
wrangle across the spool trait definition in its current form.

If `limit_concurrent_stores` is not in use, this commit doesn't
change the behavior of the spool store operation.
This commit is contained in:
Wez Furlong
2025-03-04 16:23:13 -07:00
parent 2bd7410ca6
commit 591e9f5380
8 changed files with 90 additions and 28 deletions
+2 -2
View File
@@ -561,7 +561,7 @@ async fn process_recipient<'a>(
request.trace_headers.apply_supplemental(&message)?;
if !request.deferred_spool {
message.save().await?;
message.save(None).await?;
}
log_disposition(LogDisposition {
kind: RecordType::Reception,
@@ -617,7 +617,7 @@ async fn queue_deferred(
message.set_meta("received_from", peer_address.to_string())?;
message.set_meta("queue", GENERATOR_QUEUE_NAME)?;
if !request.deferred_spool {
message.save().await?;
message.save(None).await?;
}
log_disposition(LogDisposition {
kind: RecordType::Reception,
+1 -1
View File
@@ -175,7 +175,7 @@ impl LogHookState {
if enqueue {
let queue_name = msg.get_queue_name()?;
if !deferred_spool {
msg.save().await?;
msg.save(None).await?;
}
QueueManager::insert(&queue_name, msg, InsertReason::Received.into()).await?;
}
+14 -9
View File
@@ -45,6 +45,7 @@ use thiserror::Error;
use throttle::{ThrottleResult, ThrottleSpec};
use timeq::{PopResult, TimeQ, TimerEntryWithDelay, TimerError};
use tokio::sync::{Notify, Semaphore};
use tokio::time::timeout_at;
use tracing::instrument;
static MANAGER: LazyLock<QueueManager> = LazyLock::new(|| QueueManager::new());
@@ -1391,7 +1392,7 @@ impl Queue {
}
if msg.needs_save() {
if let Err(err) = msg.save().await {
if let Err(err) = msg.save(None).await {
tracing::error!("failed to save msg after rebind: {err:#}");
}
}
@@ -1707,7 +1708,7 @@ impl Queue {
) -> anyhow::Result<()> {
tracing::trace!("save_if_needed {}", msg.id());
if msg.needs_save() {
msg.save().await?;
msg.save(None).await?;
}
match queue_config {
@@ -2725,15 +2726,19 @@ impl QueueManager {
deadline: Option<Instant>,
) -> anyhow::Result<()> {
tracing::trace!("QueueManager::insert {context:?}");
let timer = RESOLVE_LATENCY.start_timer();
let entry = Self::resolve(name).await?;
timer.stop_and_record();
if let Some(deadline) = deadline {
if deadline <= Instant::now() {
anyhow::bail!("data processing deadline exceeded, stopping short of insertion");
let timer = RESOLVE_LATENCY.start_timer();
let entry = if let Some(deadline) = deadline {
match timeout_at(deadline.into(), Self::resolve(name)).await {
Err(_) => {
anyhow::bail!("data processing deadline exceeded, stopping short of insertion")
}
Ok(result) => result?,
}
}
} else {
Self::resolve(name).await?
};
timer.stop_and_record();
let _timer = INSERT_LATENCY.start_timer();
entry.insert(msg, context).await
+41 -2
View File
@@ -1741,7 +1741,7 @@ impl SmtpServer {
Err(_) => {
self.write_response(
451,
"4.4.5 data_processing_timeout exceeded",
"4.4.5 data_processing_timeout exceeded (rx)",
Some("DATA".into()),
)
.await?;
@@ -1773,6 +1773,29 @@ impl SmtpServer {
let mut was_arf_or_oob = false;
let mut black_holed = false;
// pre-resolve any queues; there can be DNS and other async components
// to resolution that can cause this to take a non-trivial amount of time,
// so let's get that out of the way before we start writing to spool,
// to make it less complex to unwind if we exceed the allowed time.
for message in &accepted_messages {
let queue_name = message.get_queue_name()?;
match timeout_at(deadline.into(), QueueManager::resolve(&queue_name)).await {
Err(_) => {
self.write_response(
451,
"4.4.5 data_processing_timeout exceeded (resolve)",
Some("DATA".into()),
)
.await?;
return Ok(());
}
Ok(Err(err)) => {
anyhow::bail!("QueueManager::resolve({queue_name}) failed: {err:#}");
}
Ok(Ok(_handle)) => {}
}
}
for message in accepted_messages {
self.params.trace_headers.apply_supplemental(&message)?;
@@ -1807,7 +1830,23 @@ impl SmtpServer {
if queue_name != "null" {
if relay_disposition.relay && !self.params.deferred_spool {
message.save().await?;
match message.save(Some(deadline)).await {
Err(err) => {
// FIXME: unwind rest of batch
if err.root_cause().is::<tokio::time::error::Elapsed>() {
self.write_response(
451,
"4.4.5 data_processing_timeout exceeded (spool)",
Some("DATA".into()),
)
.await?;
return Ok(());
}
return Err(err);
}
Ok(_) => {}
}
}
}
+17 -10
View File
@@ -25,7 +25,7 @@ use serde::{Deserialize, Serialize};
use spool::{get_data_spool, get_meta_spool, Spool, SpoolId};
use std::hash::Hash;
use std::sync::{Arc, LazyLock, Mutex, Weak};
use std::time::Duration;
use std::time::{Duration, Instant};
use timeq::TimerEntryWithDelay;
bitflags::bitflags! {
@@ -477,15 +477,17 @@ impl Message {
|| inner.flags.contains(MessageFlags::DATA_DIRTY)
}
pub async fn save(&self) -> anyhow::Result<()> {
pub async fn save(&self, deadline: Option<Instant>) -> anyhow::Result<()> {
let _timer = SAVE_HIST.start_timer();
self.save_to(&**get_meta_spool(), &**get_data_spool()).await
self.save_to(&**get_meta_spool(), &**get_data_spool(), deadline)
.await
}
pub async fn save_to(
&self,
meta_spool: &(dyn Spool + Send + Sync),
data_spool: &(dyn Spool + Send + Sync),
deadline: Option<Instant>,
) -> anyhow::Result<()> {
let force_sync = self
.msg_and_id
@@ -498,7 +500,7 @@ impl Message {
let data_fut = if let Some(data) = self.get_data_if_dirty() {
anyhow::ensure!(!data.is_empty(), "message data must not be empty");
data_spool
.store(self.msg_and_id.id, data, force_sync)
.store(self.msg_and_id.id, data, force_sync, deadline)
.map(|_| true)
.boxed()
} else {
@@ -507,13 +509,18 @@ impl Message {
let meta_fut = if let Some(meta) = self.get_meta_if_dirty() {
let meta = Arc::new(serde_json::to_vec(&meta)?.into_boxed_slice());
meta_spool
.store(self.msg_and_id.id, meta, force_sync)
.store(self.msg_and_id.id, meta, force_sync, deadline)
.map(|_| true)
.boxed()
} else {
futures::future::ready(false).boxed()
};
// NOTE: if we have a deadline, it is tempting to want to use
// timeout_at here to enforce it, but the underlying spool
// futures are not guaranteed to be fully cancel safe, which
// is why we pass the deadline down to the save method to allow
// them to handle timeouts internally.
let (data_res, meta_res) = tokio::join!(data_fut, meta_fut);
if data_res {
@@ -541,13 +548,13 @@ impl Message {
/// Save the data+meta if needed, then release both
pub async fn save_and_shrink(&self) -> anyhow::Result<bool> {
self.save().await?;
self.save(None).await?;
self.shrink()
}
/// Save the data+meta if needed, then release just the data
pub async fn save_and_shrink_data(&self) -> anyhow::Result<bool> {
self.save().await?;
self.save(None).await?;
self.shrink_data()
}
@@ -1256,14 +1263,14 @@ impl UserData for Message {
methods.add_async_method("shrink", |_, this, _: ()| async move {
if this.needs_save() {
this.save().await.map_err(any_err)?;
this.save(None).await.map_err(any_err)?;
}
this.shrink().map_err(any_err)
});
methods.add_async_method("shrink_data", |_, this, _: ()| async move {
if this.needs_save() {
this.save().await.map_err(any_err)?;
this.save(None).await.map_err(any_err)?;
}
this.shrink_data().map_err(any_err)
});
@@ -1392,7 +1399,7 @@ impl UserData for Message {
});
methods.add_async_method("save", |_, this, ()| async move {
this.save().await.map_err(any_err)
this.save(None).await.map_err(any_err)
});
methods.add_method("set_force_sync", move |_, this, force: bool| {
+2
View File
@@ -2,6 +2,7 @@ use async_trait::async_trait;
use chrono::{DateTime, Utc};
use flume::Sender;
use std::sync::{Arc, OnceLock};
use std::time::Instant;
pub mod local_disk;
#[cfg(feature = "rocksdb")]
@@ -30,6 +31,7 @@ pub trait Spool: Send + Sync {
id: SpoolId,
data: Arc<Box<[u8]>>,
force_sync: bool,
deadline: Option<Instant>,
) -> anyhow::Result<()>;
/// Scan the contents of the spool, and emit a SpoolEntry for each item
+3
View File
@@ -8,6 +8,7 @@ use std::io::Write;
use std::os::fd::AsRawFd;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use std::time::Instant;
use tempfile::NamedTempFile;
use tokio::runtime::Handle;
@@ -98,6 +99,7 @@ impl Spool for LocalDiskSpool {
id: SpoolId,
data: Arc<Box<[u8]>>,
force_sync: bool,
_deadline: Option<Instant>,
) -> anyhow::Result<()> {
let path = self.compute_path(id);
let new_dir = self.path.join("new");
@@ -300,6 +302,7 @@ mod test {
id,
Arc::new(format!("I am {i}").as_bytes().to_vec().into_boxed_slice()),
false,
None,
)
.await?;
ids.push(id);
+10 -4
View File
@@ -10,9 +10,10 @@ use rocksdb::{
use serde::{Deserialize, Serialize};
use std::path::Path;
use std::sync::{Arc, LazyLock, Weak};
use std::time::Duration;
use std::time::{Duration, Instant};
use tokio::runtime::Handle;
use tokio::sync::Semaphore;
use tokio::time::timeout_at;
#[derive(Serialize, Deserialize, Debug)]
pub struct RocksSpoolParams {
@@ -260,6 +261,7 @@ impl Spool for RocksSpool {
id: SpoolId,
data: Arc<Box<[u8]>>,
force_sync: bool,
deadline: Option<Instant>,
) -> anyhow::Result<()> {
let mut opts = WriteOptions::default();
opts.set_sync(force_sync);
@@ -270,9 +272,12 @@ impl Spool for RocksSpool {
match self.db.write_opt(batch, &opts) {
Ok(()) => Ok(()),
Err(err) if err.kind() == ErrorKind::Incomplete => {
let permit = match self.limit_concurrent_stores.clone() {
Some(s) => Some(s.acquire_owned().await?),
None => None,
let permit = match (self.limit_concurrent_stores.clone(), deadline) {
(Some(s), Some(deadline)) => {
Some(timeout_at(deadline.into(), s.acquire_owned()).await??)
}
(Some(s), None) => Some(s.acquire_owned().await?),
(None, _) => None,
};
let db = self.db.clone();
tokio::task::Builder::new()
@@ -441,6 +446,7 @@ mod test {
id,
Arc::new(format!("I am {i}").as_bytes().to_vec().into_boxed_slice()),
false,
None,
)
.await?;
ids.push(id);