Author SHA1 Message Date
jcoffey-dev 639a415a4f Search: find addresses by local part, domain or name on PostgreSQL and MySQL
ci / fork-checks (pull_request) Successful in 43s
ci / build (pull_request) Successful in 4m20s
A 3-node PostgreSQL rehearsal found IMAP SEARCH FROM "noreply" matched
0-2 messages where RocksDB matched 23 of 930. The message indexer hands
each address and display name of From/To/Cc/Bcc to the search store as
keyword text (Language::None). The built-in index splits keyword text
into lowercase runs of alphanumerics, so an address is found by its full
form, its local part, its domain or a display-name word. The SQL
backends didn't:

- PostgreSQL's text parser keeps "[email protected]" as one email
  token (host names and URLs likewise), so neither "noreply" nor
  "amazon.com" ever matched it. Keyword text is now split the same way
  as the built-in index (SpaceTokenizer) before to_tsvector on insert
  and before plainto_tsquery/phraseto_tsquery on search, still under
  the 'simple' configuration, so the GIN index keeps serving the query.
  The sort columns keep the raw text.
- MySQL's FULLTEXT parser already splits on punctuation, but InnoDB
  never indexes its stopwords ("com", "de", "www", ...) or words under
  innodb_ft_min_token_size (3), and a required +word it hasn't indexed
  matches no row. So "amazon.com", "[email protected]" or "jane doe" found
  nothing. Those words are now matched with a word-boundary REGEXP on
  the rows the indexed words select. In language text (bodies,
  subjects) they are dropped when other words remain, and only checked
  when nothing else is left, so "the invoice" no longer finds nothing
  either.

Existing PostgreSQL search indexes hold the old single-token vectors and
need a reindex (the reindexAccounts task) before address searches find
old messages. MySQL needs none: only the query changed.

store::search_tests gains test_address_search: five messages, 28
FROM/TO/CC/BCC searches by full address, local part, domain, domain
labels, display name and hyphenated local part, plus a TEXT-style OR,
with the same expected ids on every backend. It passes on RocksDB,
SQLite, PostgreSQL and MySQL; on main it fails on PostgreSQL (From
"noreply") and MySQL (From "[email protected]").
2026-09-24 11:25:25 -07:00
jcoffey-dev 9311c1a38b Merge pull request 'Coordinator: join the cluster when NATS comes up, report the connection' (#36) from fix/coordinator-retry-and-health into main
ci / build (push) Failing after 3h0m50s
ci / fork-checks (push) Successful in 40s
2026-09-24 15:51:06 +00:00
jcoffey-dev 24be4a1b85 Merge pull request 'Publish amd64 first, then arm64, on a builder that keeps its cache' (#34) from ci/faster-publish into main
ci / fork-checks (push) Successful in 53s
ci / build (push) Canceled after 5m39s
Reviewed-on: #34
2026-09-24 15:45:26 +00:00
jcoffey-dev 95f0445d83 Coordinator: join the cluster when NATS comes up, report the connection
ci / fork-checks (pull_request) Successful in 46s
ci / build (pull_request) Successful in 11m8s
A node that started while NATS was down never got a coordinator. The
connect failed at boot, bootstrap recorded a build error and the node ran
with Coordinator::None until restarted. It had no broadcast subscriber
or publisher, so cross-node push and cache invalidation to it stayed
broken, and its healthcheck said nothing about it. Losing NATS after
startup was silent too.

- The NATS client now connects in the background
  (retry_on_initial_connect): startup never waits on NATS or fails over
  it, the node gets its coordinator, subscriber and publisher at once,
  and the client keeps trying (async-nats's backoff, at most 4 s apart)
  until NATS answers. Subscriptions made meanwhile start delivering when
  it does. A configured maxReconnects still ends the attempts.
- Three new events report the connection: cluster.coordinator-connected
  (info), cluster.coordinator-disconnected (warn: lost, closed, gave up,
  or not connected within the connection timeout at startup) and
  cluster.coordinator-error (warn: a failed attempt, reported once per
  outage rather than every retry, and server errors, slow consumers and
  lame duck mode). They are in the packaged schema, ids 644 to 646.
- GET /healthz/cluster reports the coordinator: 200
  {"coordinator":"connected"}, 503 {"coordinator":"disconnected"}, or
  200 with "none" (no coordinator) or "unknown" (a backend that doesn't
  track its connection). /healthz/live and /healthz/ready are unchanged
  on purpose: a node without its coordinator still serves mail, and
  failing those would have orchestrators restart, or pull out of
  service, every node at once whenever NATS is down.

Only NATS connects lazily; the other coordinator backends still fail at
boot as before.

cluster::coordinator::coordinator_reconnect_tests starts a node against a
NATS port with nothing behind it, checks it boots with a coordinator and
reports it disconnected, subscribes, then starts NATS on that port: the
node connects on its own and the subscription receives a message from a
second client. Stopping and restarting NATS shows disconnected, then
connected, and the same subscription keeps working.
2026-09-24 08:38:59 -07:00
jcoffey-dev c974a0918e Merge pull request 'Task manager: release task locks on stop, recheck claims held elsewhere' (#35) from fix/task-lock-recovery into main
ci / build (push) Canceled after 6m47s
ci / fork-checks (push) Successful in 49s
2026-09-24 15:38:39 +00:00
jcoffey-dev 7c80a12d75 Task manager: release task locks on stop, recheck claims held elsewhere
ci / build (pull_request) Successful in 17m2s
ci / fork-checks (pull_request) Successful in 18s
A cluster rehearsal (PostgreSQL + NATS) left index tasks pending well
past the one-hour task lock after the node that claimed them was stopped
or killed. The exact cause there isn't confirmed; this closes every path
found in the task manager that stretches a takeover past the lock, or
keeps a task claimed without running it:

- A graceful stop never released the locks it held, so every task the
  node had claimed stayed blocked for an hour. The server now tracks the
  locks it holds (common::ipc::TaskLocks) and, once the shutdown signal
  arrives, stops claiming and releases them before exiting.
- A node that failed to claim a task (another node held it) set its own
  local hold for a full lock lifetime from that scan. If the holder
  claimed it just after the scan began, or ran on a clock ahead, that
  hold ran out a moment before the lock did and was set for another
  hour: two hours in all. Such claims are now tried again every five
  minutes (a twelfth of the lock lifetime), and the task manager wakes
  up for them: before, a node without a coordinator could sleep up to
  five minutes past the recheck, or until something else woke it.
- A worker that panicked took its task type down on that node for good,
  while the scan kept claiming that type's tasks and failing to hand them
  over, re-taking each lock as it expired and so starving every other
  node of them. Each batch now runs on a task of its own; a panic is
  logged, the batch's locks are released and the worker carries on. A
  failed hand-over releases the lock too.
- A claimed task the worker couldn't read, or found gone, kept its lock
  for the hour. It is released.
- An IndexDocument task for a file (not indexed) returned no result,
  which shifted every later result in the batch onto the wrong task in
  update_tasks. It returns Ignored. Nothing queues such a task today.

The lock lifetime stays one hour; it now lives per server so the tests
can shorten it.

store::task_locks::task_lock_tests plays a second node by writing its
locks straight into the in-memory store: tasks it claimed and abandoned
run here once its locks expire, including locks that outlive this node's
view of them, and a graceful stop hands this node's locks back at once
and claims nothing more. It passes on RocksDB, SQLite and PostgreSQL.
With the old recheck it fails.
2026-09-24 08:19:05 -07:00
jcoffey-dev 22d8ad8572 Publish amd64 first, then arm64, on a builder that keeps its cache
ci / fork-checks (pull_request) Successful in 49s
ci / build (pull_request) Successful in 4m30s
Two release builds side by side on one machine each take twice as long,
and production only needs amd64. publish-amd64 now pushes :<version> as
soon as the amd64 build is done; publish-arm64 builds arm64 afterwards,
then replaces :<version> with the two-platform index and moves :latest.

Both jobs use one named BuildKit builder whose container outlives the
job, so the dependency layer (cargo chef cook) is reused until the
dependencies change. The release is created after amd64; the binaries
are attached once arm64 is in.
2026-09-24 08:04:53 -07:00
14 changed files with 870 additions and 182 deletions
+69 -15
View File
@@ -3,11 +3,28 @@
# whether a person pushed it or weekly-release.yml created it through the
# releases API.
#
# The image is multi-arch (linux/amd64, linux/arm64) as before, but built in
# one buildx run on host1 instead of one native runner per architecture: the
# Dockerfile's builder stage runs on the build platform and cross-compiles
# with an aarch64 linker, so only the small final stage (apt, setcap) goes
# through QEMU for arm64. No digest-joining job is needed.
# The image is multi-arch (linux/amd64, linux/arm64), built by two jobs on
# the image-build runner rather than one buildx run for both. The Dockerfile's
# builder stage runs on the build platform and cross-compiles with an aarch64
# linker, so only the small final stage (apt, setcap) goes through QEMU for
# arm64 -- but two release builds (LTO, one codegen unit) side by side on one
# machine each take twice as long. Production runs amd64, so amd64 goes first
# and on its own:
# * publish-amd64 pushes :<version>-amd64 and :<version>, a plain amd64
# image, as soon as its build is done. A deploy can start from it.
# * publish-arm64 then builds arm64, pushes :<version>-arm64, and replaces
# :<version> with the two-platform index. :latest moves only here, so it
# never names an image without arm64.
#
# Both jobs use one BuildKit builder, `gitea-builder`, whose container
# (buildx_buildkit_gitea-builder0) and state volume stay on the runner's host
# between jobs: a job container's `buildx create` finds the existing container
# and reuses it and its cache. The dependency build (`cargo chef cook`) is
# keyed on the recipe, which only a dependency change alters, so a release
# normally compiles just the workspace. Removing that container or its volume
# costs the next release a cold build, nothing more. The planner and dependency
# layers for the build platform are shared, so arm64 also reuses what amd64
# just did where it can.
#
# Two guards before anything is pushed:
# * the tag must be v<brand_version!>. The version is a string in
@@ -62,7 +79,7 @@ jobs:
echo "version=$V" >> "$GITHUB_OUTPUT"
echo "version $V"
publish:
publish-amd64:
needs: [version]
runs-on: docker
container:
@@ -81,16 +98,15 @@ jobs:
test -n "$REGISTRY" && test -n "$VERSION"
test -n "$PACKAGE_TOKEN" || { echo "PACKAGE_TOKEN secret is not set on this repository" >&2; exit 1; }
echo "$PACKAGE_TOKEN" | docker login -u jcoffey-dev --password-stdin "$REGISTRY"
docker run --privileged --rm tonistiigi/binfmt --install arm64
docker buildx create --use --name gitea-builder --driver docker-container || docker buildx use gitea-builder
# Attestations off, as before: they add manifests of their own to the
# index, and the index should hold the two images and nothing else.
# Attestations off, as before: they add manifests of their own, and the
# index should hold the two images and nothing else.
- run: |
docker buildx build \
--platform linux/amd64,linux/arm64 \
--platform linux/amd64 \
--provenance=false --sbom=false \
--tag "$IMAGE:$VERSION-amd64" \
--tag "$IMAGE:$VERSION" \
--tag "$IMAGE:latest" \
--push .
docker buildx imagetools inspect "$IMAGE:$VERSION"
# Gitea keeps a container package on its owner; linking it shows it on
@@ -103,11 +119,47 @@ jobs:
- if: always()
run: docker logout "$REGISTRY" || true
publish-arm64:
needs: [version, publish-amd64]
runs-on: docker
container:
image: docker:28-cli@sha256:625d9431a9f54c5a2bc90f24f0e1c3d55b1349fd857dd85035f98c2c9acbdd4d # 28-cli
volumes:
- /var/run/docker.sock:/var/run/docker.sock
env:
DOCKER_BUILDKIT: "1"
REGISTRY: ${{ vars.REGISTRY }}
IMAGE: ${{ vars.REGISTRY }}/${{ github.repository }}
VERSION: ${{ needs.version.outputs.version }}
PACKAGE_TOKEN: ${{ secrets.PACKAGE_TOKEN }}
steps:
- uses: coffey-labs/actions/checkout@fab0c4d45e0162963965f1555df27b7bed5e20ec
- run: |
echo "$PACKAGE_TOKEN" | docker login -u jcoffey-dev --password-stdin "$REGISTRY"
docker run --privileged --rm tonistiigi/binfmt --install arm64
docker buildx create --use --name gitea-builder --driver docker-container || docker buildx use gitea-builder
# The index is built from the two per-architecture tags rather than from
# :<version>, which by now is the amd64 image and would be read as such.
- run: |
docker buildx build \
--platform linux/arm64 \
--provenance=false --sbom=false \
--tag "$IMAGE:$VERSION-arm64" \
--push .
docker buildx imagetools create \
--tag "$IMAGE:$VERSION" \
--tag "$IMAGE:latest" \
"$IMAGE:$VERSION-amd64" "$IMAGE:$VERSION-arm64"
docker buildx imagetools inspect "$IMAGE:$VERSION"
- if: always()
run: docker logout "$REGISTRY" || true
# The weekly release creates its Release (and so the tag) first; a tag
# pushed by hand has none. Either way the tag ends up with exactly one
# Release, created after the image exists so its pull instructions work.
# Release, created once the amd64 image exists so its pull instructions
# work; arm64 and the binaries follow.
release:
needs: [version, publish]
needs: [version, publish-amd64]
runs-on: light
container:
image: python:3.13-slim@sha256:8d9d0b8bcf6506481eae4907c18f5e3e7902e629f5f6d684f9e7c32e85e3ddf0 # 3.13-slim
@@ -131,7 +183,9 @@ jobs:
except urllib.error.HTTPError as e:
if e.code != 404: raise
image = f"{os.environ['REGISTRY']}/{os.environ['REPO']}:{version}"
body = (f"Container image: `{image}` (linux/amd64, linux/arm64); also `:latest`.\n\n"
body = (f"Container image: `{image}` (linux/amd64, linux/arm64); also `:latest`. "
"amd64 is published first; arm64 is added to the same tag when its build "
"finishes, and `:latest` moves then.\n\n"
"Binaries for a host install are attached: `inbuxa-linux-amd64.tar.gz` and "
"`inbuxa-linux-arm64.tar.gz`, with `SHA256SUMS`. Each is the binary out of this "
"release's image for that architecture, so it is the same build. The image "
@@ -154,7 +208,7 @@ jobs:
# `docker create` does not start anything, so pulling an arm64 image on an
# amd64 runner and copying a file out of it needs no emulation.
binaries:
needs: [version, publish, release]
needs: [version, publish-arm64, release]
runs-on: docker
container:
image: docker:28-cli@sha256:625d9431a9f54c5a2bc90f24f0e1c3d55b1349fd857dd85035f98c2c9acbdd4d # 28-cli
+54
View File
@@ -335,3 +335,57 @@ impl EmailPush {
}
}
}
/// inbuxa: the task locks this node holds, so a graceful stop can hand them
/// back instead of leaving the tasks blocked until the locks expire.
pub struct TaskLocks {
held: parking_lot::Mutex<ahash::AHashSet<u64>>,
stopping: AtomicBool,
expiry: std::sync::atomic::AtomicU64,
}
impl TaskLocks {
/// How long a task lock lasts, in seconds, unless it is released first.
pub const DEFAULT_EXPIRY: u64 = 60 * 60;
pub fn is_stopping(&self) -> bool {
self.stopping.load(Ordering::Acquire)
}
/// Stops new claims and returns the ids of every lock still held.
pub fn stop(&self) -> Vec<u64> {
self.stopping.store(true, Ordering::Release);
self.held.lock().drain().collect()
}
pub fn insert(&self, id: u64) {
self.held.lock().insert(id);
}
pub fn remove(&self, id: u64) {
self.held.lock().remove(&id);
}
pub fn held(&self) -> usize {
self.held.lock().len()
}
pub fn expiry(&self) -> u64 {
self.expiry.load(Ordering::Relaxed)
}
/// Changes the lock lifetime; the tests shorten it.
pub fn set_expiry(&self, seconds: u64) {
self.expiry.store(seconds.max(1), Ordering::Relaxed);
}
}
impl Default for TaskLocks {
fn default() -> Self {
Self {
held: Default::default(),
stopping: AtomicBool::new(false),
expiry: std::sync::atomic::AtomicU64::new(Self::DEFAULT_EXPIRY),
}
}
}
+2
View File
@@ -279,6 +279,8 @@ pub struct HttpAuthCache {
pub struct Ipc {
pub push_tx: mpsc::Sender<PushEvent>,
pub task_tx: Arc<Notify>,
// inbuxa: task locks held by this node, released on a graceful stop
pub task_locks: Arc<crate::ipc::TaskLocks>,
pub queue_tx: mpsc::Sender<QueueEvent>,
pub report_tx: mpsc::Sender<ReportingEvent>,
pub broadcast_tx: Option<mpsc::Sender<BroadcastEvent>>,
+1
View File
@@ -297,6 +297,7 @@ pub fn build_ipc(has_pubsub: bool) -> (Ipc, IpcReceivers) {
report_tx,
broadcast_tx: has_pubsub.then_some(broadcast_tx),
task_tx: Arc::new(Notify::new()),
task_locks: Arc::new(crate::ipc::TaskLocks::default()),
train_task_controller: Arc::new(TrainTaskController::default()),
},
IpcReceivers {
+4
View File
@@ -109,6 +109,10 @@ async fn main() -> std::io::Result<()> {
// Wait for shutdown signal
wait_for_shutdown().await;
// inbuxa: hand back the task locks this node holds, so other nodes can
// run those tasks now rather than when the locks expire
services::task_manager::lock::release_task_locks(&inner.build_server()).await;
// Shutdown collector
Collector::shutdown();
+9 -1
View File
@@ -91,7 +91,15 @@ impl SearchIndexTask for Server {
build_contact_document(self, account_id, document_id).await
}
IndexDocumentType::File => {
// File indexing not implemented yet
// File indexing not implemented yet. inbuxa: still
// one result per task: update_tasks pairs them by
// position, and a missing one shifts every result
// after it onto the wrong task
results.push(IndexTaskResult {
task_type: TaskType::Insert,
index: task.document_type,
result: TaskResult::Ignored,
});
continue;
}
};
+34 -2
View File
@@ -2,6 +2,8 @@
* SPDX-FileCopyrightText: 2020 Stalwart Labs LLC <[email protected]>
*
* SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-SEL
*
* Modified by Coffey Labs in 2026 for INBUXA.
*/
use crate::task_manager::*;
@@ -13,13 +15,21 @@ pub trait TaskLockManager: Sync + Send {
impl TaskLockManager for Server {
async fn try_lock_task(&self, id: u64) -> bool {
// inbuxa: a node that is stopping claims nothing new
let locks = &self.inner.ipc.task_locks;
if locks.is_stopping() {
return false;
}
match self
.in_memory_store()
.try_lock(KV_LOCK_TASK, &id.to_be_bytes(), DEFAULT_LOCK_EXPIRY)
.try_lock(KV_LOCK_TASK, &id.to_be_bytes(), locks.expiry())
.await
{
Ok(result) => {
if !result {
if result {
locks.insert(id);
} else {
trc::event!(
TaskManager(TaskManagerEvent::TaskLocked),
Id = id,
@@ -48,5 +58,27 @@ impl TaskLockManager for Server {
.caused_by(trc::location!())
);
}
self.inner.ipc.task_locks.remove(id);
}
}
/// inbuxa: on a graceful stop, stops claiming tasks and releases every task
/// lock this node holds, so the rest of the cluster can pick the tasks up at
/// once instead of after the lock expires. Returns how many were released.
pub async fn release_task_locks(server: &Server) -> usize {
let ids = server.inner.ipc.task_locks.stop();
for id in &ids {
if let Err(err) = server
.in_memory_store()
.remove_lock(KV_LOCK_TASK, &id.to_be_bytes())
.await
{
trc::error!(
err.details("Failed to release task lock on shutdown")
.ctx(trc::Key::Id, *id)
.caused_by(trc::location!())
);
}
}
ids.len()
}
+197 -140
View File
@@ -2,6 +2,8 @@
* SPDX-FileCopyrightText: 2020 Stalwart Labs LLC <[email protected]>
*
* SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-SEL
*
* Modified by Coffey Labs in 2026 for INBUXA.
*/
use crate::task_manager::acme::AcmeTask;
@@ -18,7 +20,7 @@ use crate::task_manager::report::{self, SubmitReportTask};
use crate::task_manager::restore_item::RestoreItemTask;
use crate::task_manager::spam_classifier::SpamFilterMaintenanceTask;
use crate::task_manager::{
DEFAULT_LOCK_EXPIRY, Locked, QUEUE_REFRESH_INTERVAL, TaskDetails, TaskFailureType, TaskInfo,
CLAIM_RECHECK_INTERVAL, Locked, QUEUE_REFRESH_INTERVAL, TaskDetails, TaskFailureType, TaskInfo,
TaskJob, TaskManagerIpc, TaskResult,
};
use common::BuildServer;
@@ -124,72 +126,47 @@ pub fn spawn_task_manager(inner: Arc<Inner>) {
let server = inner.build_server();
let batch_size = server.core.email.index_batch_size;
let mut batch = Vec::with_capacity(batch_size);
match server
.store()
.get_value::<Task>(ValueKey::from(ValueClass::TaskQueue(
TaskQueueClass::Task { id: job.id },
)))
.await
{
Ok(Some(task)) => {
batch.push(TaskDetails { task, info: job });
}
Ok(None) => {
trc::event!(
TaskManager(TaskManagerEvent::TaskIgnored),
Id = job.id,
Reason = "Task not found in store, likely already processed.",
);
}
Err(err) => {
trc::error!(
err.id(job.id)
.details("Failed to retrieve task details.")
.caused_by(trc::location!())
);
}
if let Some(task) = fetch_task(&server, job).await {
batch.push(task);
}
while batch.len() < batch_size {
match rx.try_recv() {
Ok(job) => {
match server
.store()
.get_value::<Task>(ValueKey::from(ValueClass::TaskQueue(
TaskQueueClass::Task { id: job.id },
)))
.await
{
Ok(Some(task)) => {
batch.push(TaskDetails { task, info: job });
}
Ok(None) => {
trc::event!(
TaskManager(TaskManagerEvent::TaskIgnored),
Id = job.id,
Reason = "Task not found in store, likely already processed.",
);
}
Err(err) => {
trc::error!(
err.id(job.id)
.details("Failed to retrieve task details.")
.caused_by(trc::location!())
);
}
if let Some(task) = fetch_task(&server, job).await {
batch.push(task);
}
}
Err(_) => break,
}
}
// Dispatch
// Dispatch. inbuxa: on a task of its own, so a panic
// releases the batch's locks and leaves this worker
// running; a dead worker would keep claiming tasks it
// can never run
let mut refresh_queue = false;
let results = server.index(&batch).await.into_iter().map(|r| {
refresh_queue |= r.result.is_retry();
r.result
});
update_tasks(&server, &mut batch, results).await;
let ids = batch.iter().map(|task| task.info.id).collect::<Vec<_>>();
let run = {
let server = server.clone();
tokio::spawn(async move {
let results = server.index(&batch).await;
(batch, results)
})
};
match run.await {
Ok((mut batch, results)) => {
let results = results.into_iter().map(|r| {
refresh_queue |= r.result.is_retry();
r.result
});
update_tasks(&server, &mut batch, results).await;
}
Err(err) => {
worker_failed(&server, &ids, err).await;
refresh_queue = true;
}
}
if refresh_queue || rx.is_empty() {
server.notify_task_queue();
@@ -203,83 +180,31 @@ pub fn spawn_task_manager(inner: Arc<Inner>) {
let server = inner.build_server();
let mut refresh_queue = false;
match server
.store()
.get_value::<Task>(ValueKey::from(ValueClass::TaskQueue(
TaskQueueClass::Task { id: job.id },
)))
.await
{
Ok(Some(task)) => {
let result = match &task {
Task::CalendarAlarmEmail(task) => {
server.send_email_alarm(task, server_instance.clone()).await
}
Task::CalendarAlarmNotification(task) => {
server.send_display_alarm(task).await
}
Task::CalendarItipMessage(task) => {
server.send_imip(task, server_instance.clone()).await
}
Task::MergeThreads(task) => server.merge_threads(task).await,
Task::DmarcReport(task) => {
server
.submit_report(report::ReportId::Dmarc(task.report_id.id()))
.await
}
Task::TlsReport(task) => {
server
.submit_report(report::ReportId::Tls(task.report_id.id()))
.await
}
Task::RestoreArchivedItem(task) => server.restore_item(task).await,
Task::DestroyAccount(task) => server.destroy_account(task).await,
Task::AccountMaintenance(task) => {
server.account_maintenance(task).await
}
Task::TenantMaintenance(task) => {
server.tenant_maintenance(task).await
}
Task::StoreMaintenance(task) => {
server.store_maintenance(task).await
}
Task::SpamFilterMaintenance(task) => {
Box::pin(server.spam_filter_maintenance(task)).await
}
Task::AcmeRenewal(task) => server.acme_management(task).await,
Task::DkimManagement(task_dkim_rotation) => {
server.dkim_management(task_dkim_rotation).await
}
Task::DnsManagement(task_dns_management) => {
server.dns_management(task_dns_management).await
}
Task::IndexDocument(_)
| Task::UnindexDocument(_)
| Task::IndexTrace(_) => unreachable!(),
};
if let Some(TaskDetails { task, info }) = fetch_task(&server, job).await {
// inbuxa: on a task of its own, as above
let run = {
let server = server.clone();
let server_instance = server_instance.clone();
tokio::spawn(async move {
let result = run_task(&server, &task, server_instance).await;
(task, result)
})
};
match run.await {
Ok((task, result)) => {
refresh_queue = result.is_retry();
refresh_queue = result.is_retry();
update_tasks(
&server,
&mut [TaskDetails { task, info: job }],
vec![result],
)
.await;
}
Ok(None) => {
trc::event!(
TaskManager(TaskManagerEvent::TaskIgnored),
Id = job.id,
Reason = "Task not found in store, likely already processed.",
);
}
Err(err) => {
trc::error!(
err.id(job.id)
.details("Failed to retrieve task details.")
.caused_by(trc::location!())
);
update_tasks(
&server,
&mut [TaskDetails { task, info }],
vec![result],
)
.await;
}
Err(err) => {
worker_failed(&server, &[info.id], err).await;
refresh_queue = true;
}
}
}
@@ -318,6 +243,13 @@ pub(crate) trait TaskQueueManager: Sync + Send {
impl TaskQueueManager for Server {
async fn process_tasks(&self, ipc: &mut TaskManagerIpc) -> Duration {
// inbuxa: a node that is stopping has released its locks and claims
// nothing new
let task_locks = &self.inner.ipc.task_locks;
if task_locks.is_stopping() {
return Duration::from_secs(QUEUE_REFRESH_INTERVAL);
}
let lock_expiry = task_locks.expiry();
let now_timestamp = now();
let from_key = ValueKey::<ValueClass> {
account_id: 0,
@@ -393,9 +325,7 @@ impl TaskQueueManager for Server {
let locked = entry.get_mut();
if locked.expires <= now || locked.due < task_due {
locked.expires = Instant::now()
+ std::time::Duration::from_secs(
DEFAULT_LOCK_EXPIRY + 1,
);
+ std::time::Duration::from_secs(lock_expiry + 1);
locked.due = task_due;
tasks.push((
TaskJob {
@@ -411,9 +341,7 @@ impl TaskQueueManager for Server {
Entry::Vacant(entry) => {
entry.insert(Locked {
expires: Instant::now()
+ std::time::Duration::from_secs(
DEFAULT_LOCK_EXPIRY + 1,
),
+ std::time::Duration::from_secs(lock_expiry + 1),
due: task_due,
revision: ipc.revision,
});
@@ -464,12 +392,26 @@ impl TaskQueueManager for Server {
let tx = &ipc.txs[task_type_idx as usize];
if tx.capacity() > 0 {
if self.try_lock_task(task_job.id).await && tx.send(task_job).await.is_err() {
let id = task_job.id;
if !self.try_lock_task(id).await {
// inbuxa: another node holds the task. Look again after a
// short while rather than a full lock lifetime from now:
// the holder may have claimed it after this scan began,
// or run on a clock ahead of this one, and waiting the
// whole lifetime again would leave the task stuck for
// another hour past its lock if that holder died
if let Some(locked) = ipc.locked.get_mut(&id) {
locked.expires =
Instant::now() + Duration::from_secs(claim_recheck_interval(lock_expiry));
}
} else if tx.send(task_job).await.is_err() {
trc::event!(
Server(trc::ServerEvent::ThreadError),
Details = "Error sending task.",
CausedBy = trc::location!()
);
// inbuxa: nothing will run it here, so don't hold it
self.remove_index_lock(id).await;
}
} else {
// If the channel is full, release the lock so it can be picked up in the next iteration
@@ -481,9 +423,117 @@ impl TaskQueueManager for Server {
let now = Instant::now();
ipc.locked
.retain(|_, locked| locked.expires > now && locked.revision == ipc.revision);
Duration::from_secs(next_event.map_or(QUEUE_REFRESH_INTERVAL, |timestamp| {
let sleep_for = Duration::from_secs(next_event.map_or(QUEUE_REFRESH_INTERVAL, |timestamp| {
timestamp.saturating_sub(store::write::now())
}))
}));
// inbuxa: wake up when a claim held elsewhere is due to be tried
// again, rather than only on the next task or refresh
ipc.locked
.values()
.map(|locked| locked.expires.saturating_duration_since(now))
.min()
.map_or(sleep_for, |recheck| sleep_for.min(recheck.max(Duration::from_secs(1))))
}
}
async fn run_task(
server: &Server,
task: &Task,
server_instance: Arc<ServerInstance>,
) -> TaskResult {
match task {
Task::CalendarAlarmEmail(task) => {
server.send_email_alarm(task, server_instance.clone()).await
}
Task::CalendarAlarmNotification(task) => {
server.send_display_alarm(task).await
}
Task::CalendarItipMessage(task) => {
server.send_imip(task, server_instance.clone()).await
}
Task::MergeThreads(task) => server.merge_threads(task).await,
Task::DmarcReport(task) => {
server
.submit_report(report::ReportId::Dmarc(task.report_id.id()))
.await
}
Task::TlsReport(task) => {
server
.submit_report(report::ReportId::Tls(task.report_id.id()))
.await
}
Task::RestoreArchivedItem(task) => server.restore_item(task).await,
Task::DestroyAccount(task) => server.destroy_account(task).await,
Task::AccountMaintenance(task) => {
server.account_maintenance(task).await
}
Task::TenantMaintenance(task) => {
server.tenant_maintenance(task).await
}
Task::StoreMaintenance(task) => {
server.store_maintenance(task).await
}
Task::SpamFilterMaintenance(task) => {
Box::pin(server.spam_filter_maintenance(task)).await
}
Task::AcmeRenewal(task) => server.acme_management(task).await,
Task::DkimManagement(task_dkim_rotation) => {
server.dkim_management(task_dkim_rotation).await
}
Task::DnsManagement(task_dns_management) => {
server.dns_management(task_dns_management).await
}
Task::IndexDocument(_)
| Task::UnindexDocument(_)
| Task::IndexTrace(_) => unreachable!(),
}
}
/// Reads a claimed task. When it is gone or can't be read, the claim is
/// released: inbuxa: holding it would block the task, everywhere, until
/// the lock expired.
async fn fetch_task(server: &Server, job: TaskJob) -> Option<TaskDetails> {
match server
.store()
.get_value::<Task>(ValueKey::from(ValueClass::TaskQueue(TaskQueueClass::Task {
id: job.id,
})))
.await
{
Ok(Some(task)) => Some(TaskDetails { task, info: job }),
Ok(None) => {
trc::event!(
TaskManager(TaskManagerEvent::TaskIgnored),
Id = job.id,
Reason = "Task not found in store, likely already processed.",
);
server.remove_index_lock(job.id).await;
None
}
Err(err) => {
trc::error!(
err.id(job.id)
.details("Failed to retrieve task details.")
.caused_by(trc::location!())
);
server.remove_index_lock(job.id).await;
None
}
}
}
/// inbuxa: a task panicked: its locks are released so it runs again, here or
/// on another node, and the worker carries on.
async fn worker_failed(server: &Server, ids: &[u64], err: tokio::task::JoinError) {
trc::event!(
Server(trc::ServerEvent::ThreadError),
Details = "Task worker failed",
Reason = err.to_string(),
CausedBy = trc::location!()
);
for id in ids {
server.remove_index_lock(*id).await;
}
}
@@ -614,6 +664,13 @@ async fn update_tasks(
}
}
/// inbuxa: how long to wait before trying again to claim a task another node
/// holds: a twelfth of the lock lifetime, so five minutes for the one-hour
/// lock, never more than that and never under a second.
pub(crate) fn claim_recheck_interval(lock_expiry: u64) -> u64 {
(lock_expiry / 12).clamp(1, CLAIM_RECHECK_INTERVAL)
}
pub fn perpetual_retry_time(typ: TaskType, attempt: u64) -> Option<u64> {
matches!(
typ,
+3 -1
View File
@@ -35,7 +35,9 @@ pub mod scheduler;
pub mod spam_classifier;
const QUEUE_REFRESH_INTERVAL: u64 = 60 * 5; // 5 minutes
const DEFAULT_LOCK_EXPIRY: u64 = 60 * 60; // 1 hour
// inbuxa: the lock lifetime (one hour) lives in common::ipc::TaskLocks, per
// server, so a graceful stop can release the locks and the tests can shorten it
const CLAIM_RECHECK_INTERVAL: u64 = 60 * 5; // 5 minutes
pub(crate) struct TaskManagerIpc {
txs: [mpsc::Sender<TaskJob>; TaskType::COUNT],
+76 -13
View File
@@ -2,6 +2,8 @@
* SPDX-FileCopyrightText: 2020 Stalwart Labs LLC <[email protected]>
*
* SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-SEL
*
* Modified by Coffey Labs in 2026 for INBUXA.
*/
use crate::{
@@ -19,7 +21,7 @@ use crate::{
write::SearchIndex,
};
use mysql_async::{IsolationLevel, TxOpts, Value, prelude::Queryable};
use nlp::tokenizers::word::WordTokenizer;
use nlp::{language::Language, tokenizers::word::WordTokenizer};
use std::fmt::Write;
impl MysqlStore {
@@ -146,6 +148,20 @@ impl MysqlStore {
}
}
// inbuxa: InnoDB's default full-text stopword list
// (INFORMATION_SCHEMA.INNODB_FT_DEFAULT_STOPWORD) and innodb_ft_min_token_size
// default; words outside these are not in a FULLTEXT index.
const FT_STOPWORDS: &[&str] = &[
"a", "about", "an", "are", "as", "at", "be", "by", "com", "de", "en", "for", "from", "how",
"i", "in", "is", "it", "la", "of", "on", "or", "that", "the", "this", "to", "was", "what",
"when", "where", "who", "will", "with", "und", "www",
];
const FT_MIN_TOKEN_SIZE: usize = 3;
fn is_ft_indexed(word: &str) -> bool {
word.chars().count() >= FT_MIN_TOKEN_SIZE && !FT_STOPWORDS.contains(&word)
}
fn build_filter(query: &mut String, filters: &[SearchFilter]) -> Vec<Value> {
if filters.is_empty() {
return Vec::new();
@@ -171,30 +187,77 @@ fn build_filter(query: &mut String, filters: &[SearchFilter]) -> Vec<Value> {
if field.is_text() && matches!(op, SearchOperator::Equal | SearchOperator::Contains)
{
let (value, mode) = match (value, op) {
(SearchValue::Text { value, .. }, SearchOperator::Equal) => {
(Value::Bytes(format!("{value:?}").into_bytes()), "BOOLEAN")
}
(SearchValue::Text { value, .. }, ..) => {
let (value, mode, unindexed) = match (value, op) {
(SearchValue::Text { value, .. }, SearchOperator::Equal) => (
Value::Bytes(format!("{value:?}").into_bytes()),
"BOOLEAN",
Vec::new(),
),
(SearchValue::Text { value, language }, ..) => {
let mut text_query = String::with_capacity(value.len() + 1);
let mut unindexed = Vec::new();
for item in WordTokenizer::new(value, MAX_TOKEN_LENGTH) {
if !text_query.is_empty() {
text_query.push(' ');
// inbuxa: InnoDB never indexes stopwords ("com",
// "de", "www", ...) or words under
// innodb_ft_min_token_size, and a required
// (+word) term it has not indexed matches no row,
// so "example.com" or "[email protected]" found
// nothing. Such words are matched with a
// word-boundary REGEXP instead.
if is_ft_indexed(&item.word) {
if !text_query.is_empty() {
text_query.push(' ');
}
text_query.push('+');
text_query.push_str(&item.word);
} else {
unindexed.push(item.word);
}
text_query.push('+');
text_query.push_str(&item.word);
}
(Value::Bytes(text_query.into_bytes()), "BOOLEAN")
// For language text (bodies, subjects) the unindexed
// words are noise words and only checked when nothing
// else is left to match; keyword text (addresses,
// contact fields) checks every word, as the other
// backends do.
if !text_query.is_empty() && !matches!(language, Language::None) {
unindexed.clear();
}
(Value::Bytes(text_query.into_bytes()), "BOOLEAN", unindexed)
}
_ => {
debug_assert!(false, "Invalid search value for text field");
continue;
}
};
let _ = write!(query, "MATCH({}) AGAINST(? IN {mode} MODE)", field.column());
values.push(value);
if unindexed.is_empty() {
let _ =
write!(query, "MATCH({}) AGAINST(? IN {mode} MODE)", field.column());
values.push(value);
} else {
query.push('(');
let is_empty = matches!(&value, Value::Bytes(v) if v.is_empty());
if !is_empty {
let _ = write!(
query,
"MATCH({}) AGAINST(? IN {mode} MODE) AND ",
field.column()
);
values.push(value);
}
for (i, word) in unindexed.iter().enumerate() {
if i > 0 {
query.push_str(" AND ");
}
let _ = write!(query, "{} REGEXP ?", field.column());
values.push(Value::Bytes(
format!("(^|[^[:alnum:]]){word}([^[:alnum:]]|$)").into_bytes(),
));
}
query.push(')');
}
} else if let SearchValue::KeyValues(kv) = value {
let (key, value) = kv.iter().next().unwrap();
+79 -10
View File
@@ -2,12 +2,17 @@
* SPDX-FileCopyrightText: 2020 Stalwart Labs LLC <[email protected]>
*
* SPDX-License-Identifier: AGPL-3.0-only OR LicenseRef-SEL
*
* Modified by Coffey Labs in 2026 for INBUXA.
*/
use crate::{
backend::postgres::{
DELETE_CHUNK_SIZE, MIN_DELETE_CHUNK_SIZE, PostgresStore, PsqlSearchField, into_error,
into_pool_error, is_timeout_error,
backend::{
MAX_TOKEN_LENGTH,
postgres::{
DELETE_CHUNK_SIZE, MIN_DELETE_CHUNK_SIZE, PostgresStore, PsqlSearchField, into_error,
into_pool_error, is_timeout_error,
},
},
search::{
IndexDocument, SearchComparator, SearchDocumentId, SearchFilter, SearchOperator,
@@ -15,7 +20,7 @@ use crate::{
},
write::SearchIndex,
};
use nlp::language::Language;
use nlp::{language::Language, tokenizers::space::SpaceTokenizer};
use std::fmt::Write;
use tokio_postgres::{
IsolationLevel,
@@ -43,6 +48,19 @@ impl PostgresStore {
let primary_keys = index.primary_keys();
let all_fields = index.all_fields();
let fields = document.fields;
// inbuxa: keyword text (addresses, contact fields, ...) is split into
// words before it reaches the text parser, see keyword_terms().
let keywords = primary_keys
.iter()
.chain(all_fields)
.map(|field| match fields.get(field) {
Some(SearchValue::Text {
value,
language: Language::None,
}) if field.is_text() => Some(keyword_terms(value)),
_ => None,
})
.collect::<Vec<_>>();
let mut values = Vec::with_capacity(fields.len() + 2);
let mut query = format!("INSERT INTO {} (", index.psql_table());
@@ -74,7 +92,20 @@ impl PostgresStore {
(0, PG_UNSTEMMED_LANG)
};
if field.is_text() {
if let Some(keywords) = &keywords[i] {
let _ = write!(&mut query, "to_tsvector('{language}',{value_ref})");
values.push(keywords as &(dyn ToSql + Sync));
if field.sort_column().is_some() {
let value_ref = format!("${}", values.len() + 1);
if text_len > 255 {
let _ = write!(&mut query, ",left({value_ref},255)");
} else {
let _ = write!(&mut query, ",{value_ref}");
}
values.push(value as &(dyn ToSql + Sync));
}
continue;
} else if field.is_text() {
let _ = write!(&mut query, "to_tsvector('{language}',{value_ref})");
} else if text_len > 512 {
query.push_str("left(");
@@ -134,6 +165,7 @@ impl PostgresStore {
) -> trc::Result<Vec<R>> {
let mut query = format!("SELECT {} FROM {}", R::field().column(), index.psql_table());
let params = self.build_filter(&mut query, filters);
let params = params.iter().map(SqlParam::as_sql).collect::<Vec<_>>();
if !sort.is_empty() {
build_sort(&mut query, sort);
}
@@ -155,6 +187,7 @@ impl PostgresStore {
let table = filter.index.psql_table();
let mut where_clause = String::new();
let params = self.build_filter(&mut where_clause, &filter.filters);
let params = params.iter().map(SqlParam::as_sql).collect::<Vec<_>>();
let conn = self.conn_pool.get().await.map_err(into_pool_error)?;
let s = conn
.prepare_cached(&format!("DELETE FROM {table}{where_clause}"))
@@ -196,7 +229,7 @@ impl PostgresStore {
&self,
query: &mut String,
filters: &'x [SearchFilter],
) -> Vec<&'x (dyn ToSql + Sync)> {
) -> Vec<SqlParam<'x>> {
if filters.is_empty() {
return Vec::new();
}
@@ -237,6 +270,10 @@ impl PostgresStore {
if matches!(language, Language::None) {
let _ = write!(query, "@@ {method}('{config}', ${value_pos})");
if let SearchValue::Text { value, .. } = value {
values.push(SqlParam::Owned(keyword_terms(value)));
continue;
}
} else {
let _ = write!(query, "@@ ({method}('{config}', ${value_pos})");
for fallback in [PG_FALLBACK_LANG, PG_UNSTEMMED_LANG] {
@@ -247,18 +284,18 @@ impl PostgresStore {
}
query.push(')');
}
values.push(value as &(dyn ToSql + Sync));
values.push(SqlParam::Ref(value));
} else if let SearchValue::KeyValues(kv) = value {
query.push_str(field.column());
query.push(' ');
let (key, value) = kv.iter().next().unwrap();
values.push(key as &(dyn ToSql + Sync));
values.push(SqlParam::Ref(key));
if !value.is_empty() {
let _ = write!(query, "->> ${value_pos} ");
op.write_pqsql(query, values.len() + 1);
values.push(value as &(dyn ToSql + Sync));
values.push(SqlParam::Ref(value));
} else {
let _ = write!(query, " ? ${value_pos}");
}
@@ -267,7 +304,7 @@ impl PostgresStore {
query.push(' ');
op.write_pqsql(query, value_pos);
values.push(value as &(dyn ToSql + Sync));
values.push(SqlParam::Ref(value));
}
}
SearchFilter::And | SearchFilter::Or => {
@@ -321,6 +358,38 @@ impl PostgresStore {
}
}
// inbuxa: PostgreSQL's text parser keeps "[email protected]" (and host names,
// URLs, file paths, ...) as a single token, so a search for "user" or
// "example.com" never matched an address. Keyword text is split into words the
// same way the built-in index splits it (SpaceTokenizer: lowercase runs of
// alphanumerics) on both the indexing and the query side, so a full address,
// its local part, its domain and the display-name words all match, as they do
// on the other backends.
pub(crate) fn keyword_terms(value: &str) -> String {
let mut terms = String::with_capacity(value.len());
for token in SpaceTokenizer::new(value, MAX_TOKEN_LENGTH) {
if !terms.is_empty() {
terms.push(' ');
}
terms.push_str(&token);
}
terms
}
pub(super) enum SqlParam<'x> {
Ref(&'x (dyn ToSql + Sync)),
Owned(String),
}
impl SqlParam<'_> {
fn as_sql(&self) -> &(dyn ToSql + Sync) {
match self {
SqlParam::Ref(value) => *value,
SqlParam::Owned(value) => value,
}
}
}
fn build_sort(query: &mut String, sort: &[SearchComparator]) {
query.push_str(" ORDER BY ");
for (i, comparator) in sort.iter().enumerate() {
+1
View File
@@ -21,6 +21,7 @@ pub mod replica_cluster; // inbuxa: read replicas across nodes
pub mod scaleout; // inbuxa: scale-out storage
#[cfg(any(feature = "postgres", feature = "mysql"))]
pub mod sql_timeout;
pub mod task_locks; // inbuxa: task locks across nodes
use crate::utils::server::TestServerBuilder;
use std::io::Read;
+157
View File
@@ -128,6 +128,11 @@ pub async fn test(test: &TestServer) {
println!("Running trace document tests...");
test_trace_documents(store.clone()).await;
// inbuxa: address fields match by full address, local part, domain and
// display name on every backend
println!("Running address search tests...");
test_address_search(store.clone()).await;
// Large document insert test
println!("Running large document insert tests...");
let mut large_text = String::with_capacity(20 * 1024 * 1024);
@@ -972,3 +977,155 @@ async fn test_trace_documents(store: SearchStore) {
.unwrap();
}
}
// inbuxa: the message indexer passes each display name and each address of
// From/To/Cc/Bcc as keyword text (Language::None). The built-in index splits
// that text into words, so an address is found by its full form, its local
// part, its domain or a display-name word; PostgreSQL kept the whole address
// as one token and MySQL dropped stopwords such as "com" and words under three
// characters. The expected results below are the built-in (RocksDB/SQLite)
// results and must be the same on every backend.
async fn test_address_search(store: SearchStore) {
const ACCOUNT_ID: u32 = 7;
let messages: [[&[(&str, &str)]; 4]; 5] = [
// From, To, Cc, Bcc
[
&[("Amazon.com", "[email protected]")],
&[("Jane Doe", "[email protected]")],
&[],
&[],
],
[
&[("", "[email protected]")],
&[("", "[email protected]")],
&[("Jane Doe", "[email protected]")],
&[],
],
[
&[("GitHub", "[email protected]")],
&[("Jo Li", "[email protected]")],
&[],
&[("Audit", "[email protected]")],
],
[
&[("Jane Doe", "[email protected]")],
&[("Amazon Web Services", "[email protected]")],
&[("Bob", "[email protected]")],
&[("", "[email protected]")],
],
[
&[("Newsletter", "[email protected]")],
&[("", "[email protected]")],
&[],
&[],
],
];
let fields = [
EmailSearchField::From,
EmailSearchField::To,
EmailSearchField::Cc,
EmailSearchField::Bcc,
];
let mut documents = Vec::new();
let mut mask = RoaringBitmap::new();
for (document_id, message) in messages.iter().enumerate() {
let mut document = IndexDocument::new(SearchIndex::Email)
.with_account_id(ACCOUNT_ID)
.with_document_id(document_id as u32);
for (field, addresses) in fields.iter().zip(message.iter()) {
for (name, address) in addresses.iter() {
if !name.is_empty() {
document.index_text(field.clone(), name, Language::None);
}
document.index_text(field.clone(), address, Language::None);
}
}
document.index_unsigned(EmailSearchField::ReceivedAt, document_id as u64);
documents.push(document);
mask.insert(document_id as u32);
}
store.index(documents).await.unwrap();
if let SearchStore::ElasticSearch(store) = &store {
store.refresh_index(SearchIndex::Email).await.unwrap();
}
for (field, text, expected) in [
// full address
(EmailSearchField::From, "[email protected]", vec![0u32]),
(EmailSearchField::To, "[email protected]", vec![0]),
(EmailSearchField::Cc, "[email protected]", vec![1]),
(EmailSearchField::Bcc, "[email protected]", vec![3]),
(EmailSearchField::To, "[email protected]", vec![1, 2]),
// local part
(EmailSearchField::From, "noreply", vec![0, 2]),
(EmailSearchField::To, "jo", vec![1, 2]),
(EmailSearchField::Cc, "bob", vec![3]),
(EmailSearchField::Bcc, "audit", vec![2]),
// domain
(EmailSearchField::From, "amazon.com", vec![0, 1]),
(EmailSearchField::From, "amazon", vec![0, 1]),
(EmailSearchField::To, "example.org", vec![0, 4]),
(EmailSearchField::To, "io.de", vec![1, 2]),
(EmailSearchField::Cc, "example.net", vec![3]),
(EmailSearchField::Bcc, "example.org", vec![2]),
(EmailSearchField::From, "www.example.com", vec![4]),
(EmailSearchField::From, "com", vec![0, 1, 2, 4]),
// display name
(EmailSearchField::From, "Jane", vec![3]),
(EmailSearchField::From, "jane doe", vec![3]),
(EmailSearchField::To, "Web Services", vec![3]),
(EmailSearchField::To, "Li", vec![2]),
(EmailSearchField::Cc, "Doe", vec![1]),
(EmailSearchField::Bcc, "Audit", vec![2]),
// hyphenated local part
(EmailSearchField::From, "shipment-tracking", vec![1]),
(EmailSearchField::From, "tracking", vec![1]),
// no match
(EmailSearchField::From, "amazon.org", vec![]),
(EmailSearchField::To, "noreply", vec![]),
(EmailSearchField::Bcc, "jane", vec![]),
] {
let ids = store
.query_account(
SearchQuery::new(SearchIndex::Email)
.with_filters(vec![
SearchFilter::eq(SearchField::AccountId, ACCOUNT_ID),
SearchFilter::has_keyword(field.clone(), text),
])
.with_comparator(SearchComparator::ascending(EmailSearchField::ReceivedAt))
.with_mask(mask.clone()),
)
.await
.unwrap();
assert_eq!(ids, expected, "{field:?} {text:?}");
}
// TEXT-style search across all address fields
let ids = store
.query_account(
SearchQuery::new(SearchIndex::Email)
.with_filters(vec![
SearchFilter::eq(SearchField::AccountId, ACCOUNT_ID),
SearchFilter::Or,
SearchFilter::has_keyword(EmailSearchField::From, "example.org"),
SearchFilter::has_keyword(EmailSearchField::To, "example.org"),
SearchFilter::has_keyword(EmailSearchField::Cc, "example.org"),
SearchFilter::has_keyword(EmailSearchField::Bcc, "example.org"),
SearchFilter::End,
])
.with_comparator(SearchComparator::ascending(EmailSearchField::ReceivedAt))
.with_mask(mask.clone()),
)
.await
.unwrap();
assert_eq!(ids, vec![0, 1, 2, 3, 4]);
store
.unindex(
SearchQuery::new(SearchIndex::Email)
.with_filter(SearchFilter::eq(SearchField::AccountId, ACCOUNT_ID)),
)
.await
.unwrap();
}
+184
View File
@@ -0,0 +1,184 @@
/*
* SPDX-FileCopyrightText: 2026 Coffey Labs
*
* SPDX-License-Identifier: AGPL-3.0-only
*/
//! Task locks across nodes: tasks claimed by a node that then disappears
//! run elsewhere once its locks expire, and a node that stops gracefully
//! hands its locks back at once. The other node is played by writing its
//! locks straight into the shared in-memory store, as a node that claimed
//! the tasks and died leaves them.
use crate::utils::server::TestServerBuilder;
use common::{KV_LOCK_TASK, Server};
use registry::schema::{
enums::IndexDocumentType,
structs::{Task, TaskIndexDocument, TaskStatus},
};
use services::task_manager::lock::{TaskLockManager, release_task_locks};
use std::time::{Duration, Instant};
use store::{
ValueKey,
write::{BatchBuilder, TaskQueueClass, ValueClass},
};
use utils::snowflake::SnowflakeIdGenerator;
// Short enough for a test, long enough that the recheck interval (a twelfth
// of it) is well below it
const LOCK_EXPIRY: u64 = 12;
#[tokio::test(flavor = "multi_thread")]
pub async fn task_lock_tests() {
let test = TestServerBuilder::new("task_lock_tests")
.await
.build()
.await;
let server = test.server.clone();
println!(
"Running task lock tests on {}...",
std::env::var("STORE").unwrap_or_default()
);
server.inner.ipc.task_locks.set_expiry(LOCK_EXPIRY);
// 1. Another node claimed the tasks and died. Its locks block them until
// they expire; then this node runs them, without waiting for anything
// else to wake it
let ids = new_task_ids(4);
for id in &ids {
assert!(foreign_lock(&server, *id, LOCK_EXPIRY).await);
}
schedule(&server, &ids).await;
let started = Instant::now();
server.notify_task_queue();
tokio::time::sleep(Duration::from_secs(3)).await;
assert_eq!(pending(&server, &ids).await, ids.len(), "held by the other node");
wait_until_done(&server, &ids, Duration::from_secs(LOCK_EXPIRY + 10)).await;
let elapsed = started.elapsed();
assert!(
elapsed >= Duration::from_secs(LOCK_EXPIRY - 2),
"ran before the other node's locks expired: {elapsed:?}"
);
// 2. The other node's locks outlive what this node expects: claimed just
// after this node looked, or by a node whose clock runs ahead. This node
// keeps checking at the recheck interval, so the tasks run soon after
// those locks expire, not a whole lock lifetime later
let held_for = LOCK_EXPIRY + LOCK_EXPIRY / 2;
let ids = new_task_ids(4);
for id in &ids {
assert!(foreign_lock(&server, *id, held_for).await);
}
schedule(&server, &ids).await;
let started = Instant::now();
server.notify_task_queue();
wait_until_done(&server, &ids, Duration::from_secs(held_for + 8)).await;
let elapsed = started.elapsed();
assert!(
elapsed >= Duration::from_secs(held_for - 2),
"ran before the other node's locks expired: {elapsed:?}"
);
// 3. A graceful stop releases the locks this node holds: another node
// can claim those tasks at once, and this one claims nothing more
let ids = new_task_ids(3);
for id in &ids {
assert!(server.try_lock_task(*id).await, "claim {id}");
}
assert_eq!(server.inner.ipc.task_locks.held(), ids.len());
for id in &ids {
assert!(
!foreign_lock(&server, *id, LOCK_EXPIRY).await,
"held while this node runs"
);
}
assert_eq!(release_task_locks(&server).await, ids.len());
assert_eq!(server.inner.ipc.task_locks.held(), 0);
for id in &ids {
assert!(
foreign_lock(&server, *id, LOCK_EXPIRY).await,
"released on stop: {id}"
);
}
let [id] = new_task_ids(1)[..] else {
unreachable!()
};
assert!(!server.try_lock_task(id).await, "a stopping node claims nothing");
for id in ids {
let _ = server
.in_memory_store()
.remove_lock(KV_LOCK_TASK, &id.to_be_bytes())
.await;
}
if test.is_reset() {
test.temp_dir.delete();
}
}
fn new_task_ids(count: usize) -> Vec<u64> {
(0..count)
.map(|_| SnowflakeIdGenerator::global_id().unwrap())
.collect()
}
/// The other node's claim on a task, as its task manager takes it.
async fn foreign_lock(server: &Server, id: u64, seconds: u64) -> bool {
server
.in_memory_store()
.try_lock(KV_LOCK_TASK, &id.to_be_bytes(), seconds)
.await
.unwrap()
}
/// Unindex tasks for files that don't exist: files aren't search-indexed and
/// there is no undelete note, so running one only drops it from the queue.
async fn schedule(server: &Server, ids: &[u64]) {
let mut batch = BatchBuilder::new();
for (n, id) in ids.iter().enumerate() {
batch.schedule_task_with_id(
*id,
Task::UnindexDocument(TaskIndexDocument {
account_id: 0u32.into(),
document_id: (u32::MAX - n as u32).into(),
document_type: IndexDocumentType::File,
status: TaskStatus::now(),
}),
);
}
server.store().write(batch.build_all()).await.unwrap();
}
async fn pending(server: &Server, ids: &[u64]) -> usize {
let mut count = 0;
for id in ids {
if server
.store()
.get_value::<Task>(ValueKey::from(ValueClass::TaskQueue(
TaskQueueClass::Task { id: *id },
)))
.await
.unwrap()
.is_some()
{
count += 1;
}
}
count
}
async fn wait_until_done(server: &Server, ids: &[u64], within: Duration) {
let started = Instant::now();
loop {
let left = pending(server, ids).await;
if left == 0 {
return;
}
assert!(
started.elapsed() < within,
"{left} task(s) still pending after {:?}",
started.elapsed()
);
tokio::time::sleep(Duration::from_millis(250)).await;
}
}