Separate orchestration by protocol action

This commit is contained in:
ms2k 2026-07-19 14:27:59 +09:00
commit 6f0f20f5cc
40 changed files with 6974 additions and 6839 deletions

File diff suppressed because it is too large Load diff

View file

@ -0,0 +1,404 @@
use super::*;
pub(super) async fn create_source_credential(
state: web::Data<PublicApiState>,
http: HttpRequest,
request: web::Json<SourceCredentialCreateDto>,
) -> Result<HttpResponse, PublicApiError> {
validate_request("SourceCredentialCreate", &*request)?;
let principal = state.principal(&http).await?;
let key = idempotency_key(&http)?;
let request = request.into_inner();
let request_fingerprint =
fingerprint(&request).map_err(|_| PublicApiError::invalid_request())?;
let create = create_credential(request)?;
let credentials = state.credentials.clone();
let owner = principal.id.clone();
let scope = MutationScope {
subject: &principal.id,
operation: "credential.create",
key,
};
if let Some(outcome) = observe_mutation(
&state.pool,
scope,
"create_source_credential",
&[],
request_fingerprint,
)
.await?
{
return credential_mutation_response(outcome, StatusCode::CREATED);
}
let prepared = credentials
.prepare_create(&owner, create)
.await
.map_err(PublicApiError::from_credential)?;
let outcome = execute_mutation(
&state.pool,
scope,
"create_source_credential",
&[],
request_fingerprint,
move |tx| {
Box::pin(async move {
let credential = CredentialRegistry::commit_prepared_create_in_tx(tx, prepared)
.await
.map_err(|error| DurableError::Effect(error.to_string()))?;
let response = serde_json::to_value(
credential_dto(credential)
.map_err(|error| DurableError::Effect(error.to_string()))?,
)
.map_err(|error| DurableError::Effect(error.to_string()))?;
Ok((MutationStatus::Accepted, response))
})
},
)
.await?;
credential_mutation_response(outcome, StatusCode::CREATED)
}
pub(super) async fn list_source_credentials(
state: web::Data<PublicApiState>,
http: HttpRequest,
query: web::Query<PageQuery>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let caller = credential_caller(&principal);
let filter = CursorCodec::filter_fingerprint(&json!({
"resource": "credentials",
"authorization": authorization_snapshot(&principal)
}))?;
let cursor = query
.cursor
.as_deref()
.map(|token| state.cursors.bind(token, principal.id.as_str(), filter))
.transpose()
.map_err(PublicApiError::from_cursor)?;
let registry_cursor = cursor
.as_ref()
.and_then(|value| value.after.as_ref())
.map(|value| value.id.as_str());
let page = state
.credentials
.list(&caller, page_limit(query.limit)?, registry_cursor)
.await
.map_err(PublicApiError::from_credential)?;
let next_cursor = page
.next_cursor
.as_deref()
.map(|raw| {
let next = KeysetPosition {
created_at: OffsetDateTime::UNIX_EPOCH,
id: raw.to_owned(),
};
let bound = match cursor.as_ref() {
Some(current) => next_page_cursor(current, next),
None => first_page_cursor(
principal.id.as_str(),
filter,
KeysetPosition {
created_at: OffsetDateTime::UNIX_EPOCH,
id: String::new(),
},
next,
),
};
state.cursors.encode(&bound)
})
.transpose()?;
let dto = SourceCredentialPageDto {
items: page
.items
.into_iter()
.map(credential_dto)
.collect::<Result<_, _>>()?,
next_cursor: NullableDto(next_cursor),
};
Ok(HttpResponse::Ok().json(dto))
}
pub(super) async fn get_source_credential(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let credential_id = parse_id(path.into_inner())?;
let credential = state
.credentials
.get(&credential_caller(&principal), &credential_id)
.await
.map_err(PublicApiError::from_credential)?
.ok_or_else(PublicApiError::not_found)?;
Ok(HttpResponse::Ok().json(credential_dto(credential)?))
}
pub(super) async fn rotate_source_credential(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
request: web::Json<SourceCredentialRotateDto>,
) -> Result<HttpResponse, PublicApiError> {
validate_request("SourceCredentialRotate", &*request)?;
let principal = state.principal(&http).await?;
let credential_id = parse_id(path.into_inner())?;
let key = idempotency_key(&http)?;
let request = request.into_inner();
let request_fingerprint = fingerprint(&json!({
"credential_id": credential_id,
"request": request
}))
.map_err(|_| PublicApiError::invalid_request())?;
let credentials = state.credentials.clone();
let caller = credential_caller(&principal);
let scope = MutationScope {
subject: &principal.id,
operation: "credential.rotate",
key,
};
if let Some(outcome) = observe_mutation(
&state.pool,
scope,
"rotate_source_credential",
&[],
request_fingerprint,
)
.await?
{
return credential_mutation_response(outcome, StatusCode::CREATED);
}
let existing = credentials
.get(&caller, &credential_id)
.await
.map_err(PublicApiError::from_credential)?
.ok_or_else(PublicApiError::not_found)?;
let material = credential_material_for_kind(request.material, existing.kind);
let prepared = credentials
.prepare_rotate(&caller, &credential_id, material)
.await
.map_err(PublicApiError::from_credential)?;
let outcome = execute_mutation(
&state.pool,
scope,
"rotate_source_credential",
&[],
request_fingerprint,
move |tx| {
Box::pin(async move {
match CredentialRegistry::commit_prepared_rotate_in_tx(tx, prepared).await? {
RotateTxOutcome::Applied(credential) => {
let response = serde_json::to_value(
credential_dto(credential)
.map_err(|error| DurableError::Effect(error.to_string()))?,
)
.map_err(|error| DurableError::Effect(error.to_string()))?;
Ok((MutationStatus::Accepted, response))
}
RotateTxOutcome::Revoked => Ok((
MutationStatus::Rejected,
public_error_body(
ApiErrorCodeDto::PreconditionFailed,
"precondition failed",
),
)),
RotateTxOutcome::IntentClaimed => Err(DurableError::Effect(
"credential rotation intent is unavailable".into(),
)),
}
})
},
)
.await?;
credential_mutation_response(outcome, StatusCode::CREATED)
}
pub(super) async fn revoke_source_credential(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let credential_id = parse_id(path.into_inner())?;
let key = idempotency_key(&http)?;
let request_fingerprint = fingerprint(&json!({ "credential_id": credential_id }))
.map_err(|_| PublicApiError::internal())?;
let caller = credential_caller(&principal);
let scope = MutationScope {
subject: &principal.id,
operation: "credential.revoke",
key,
};
if let Some(outcome) = observe_mutation(
&state.pool,
scope,
"revoke_source_credential",
&[],
request_fingerprint,
)
.await?
{
return credential_mutation_response(outcome, StatusCode::OK);
}
state
.credentials
.get(&caller, &credential_id)
.await
.map_err(PublicApiError::from_credential)?
.ok_or_else(PublicApiError::not_found)?;
let credential_for_effect = credential_id.clone();
let outcome = execute_mutation(
&state.pool,
scope,
"revoke_source_credential",
&[],
request_fingerprint,
move |tx| {
Box::pin(async move {
let credential =
CredentialRegistry::revoke_in_tx(tx, &caller, &credential_for_effect)
.await
.map_err(|error| DurableError::Effect(error.to_string()))?;
let response = serde_json::to_value(
credential_dto(credential)
.map_err(|error| DurableError::Effect(error.to_string()))?,
)
.map_err(|error| DurableError::Effect(error.to_string()))?;
Ok((MutationStatus::Accepted, response))
})
},
)
.await?;
credential_mutation_response(outcome, StatusCode::OK)
}
pub(super) fn create_credential(
request: SourceCredentialCreateDto,
) -> Result<CreateCredential, PublicApiError> {
let (label, material) = match request {
SourceCredentialCreateDto::BasicAuth(value) => (
value.label,
SecretMaterial::BasicAuth {
username: value.material.username,
password: value.material.password,
},
),
SourceCredentialCreateDto::BearerToken(value) => (
value.label,
SecretMaterial::BearerToken {
token: value.material.token,
},
),
SourceCredentialCreateDto::HttpHeaders(value) => (
value.label,
SecretMaterial::HttpHeaders {
headers: value.material.headers,
},
),
SourceCredentialCreateDto::ObjectStore(value) => (
value.label,
SecretMaterial::ObjectStore {
access_key_id: value.material.access_key_id,
secret_access_key: value.material.secret_access_key,
session_token: value.material.session_token,
},
),
SourceCredentialCreateDto::HuggingFace(value) => (
value.label,
SecretMaterial::HuggingFace {
token: value.material.token,
},
),
};
Ok(CreateCredential { label, material })
}
pub(super) fn credential_material(material: CredentialMaterialDto) -> SecretMaterial {
match material {
CredentialMaterialDto::BasicAuth(value) => SecretMaterial::BasicAuth {
username: value.username,
password: value.password,
},
CredentialMaterialDto::Token(value) => SecretMaterial::BearerToken { token: value.token },
CredentialMaterialDto::HttpHeaders(value) => SecretMaterial::HttpHeaders {
headers: value.headers,
},
CredentialMaterialDto::ObjectStore(value) => SecretMaterial::ObjectStore {
access_key_id: value.access_key_id,
secret_access_key: value.secret_access_key,
session_token: value.session_token,
},
}
}
pub(super) fn credential_material_for_kind(
material: CredentialMaterialDto,
kind: SourceCredentialKind,
) -> SecretMaterial {
match (material, kind) {
(CredentialMaterialDto::Token(value), SourceCredentialKind::HuggingFace) => {
SecretMaterial::HuggingFace { token: value.token }
}
(material, _) => credential_material(material),
}
}
pub(super) fn credential_mutation_response(
outcome: remora_durable::MutationOutcome,
accepted_status: StatusCode,
) -> Result<HttpResponse, PublicApiError> {
if outcome.disposition == MutationDisposition::Conflict {
return Err(PublicApiError::idempotency_conflict());
}
let response = if outcome.status == MutationStatus::Rejected {
HttpResponse::PreconditionFailed()
.content_type("application/problem+json")
.json(outcome.response)
} else {
HttpResponse::build(accepted_status).json(outcome.response)
};
Ok(with_idempotency_expiry(response))
}
pub(super) fn credential_dto(
value: SourceCredential,
) -> Result<SourceCredentialDto, PublicApiError> {
let kind = match value.kind {
SourceCredentialKind::BasicAuth => SourceCredentialKindDto::BasicAuth,
SourceCredentialKind::BearerToken => SourceCredentialKindDto::BearerToken,
SourceCredentialKind::HttpHeaders => SourceCredentialKindDto::HttpHeaders,
SourceCredentialKind::ObjectStore => SourceCredentialKindDto::ObjectStore,
SourceCredentialKind::HuggingFace => SourceCredentialKindDto::HuggingFace,
};
let status = match value.status {
SourceCredentialStatus::Active => SourceCredentialStatusDto::Active,
SourceCredentialStatus::Revoked => SourceCredentialStatusDto::Revoked,
};
Ok(SourceCredentialDto {
id: value.id,
owner: value.owner,
label: value.label,
kind,
status,
active_version: NullableDto(
value
.active_version
.map(|version| {
Ok::<_, PublicApiError>(SourceCredentialVersionDto {
id: version.id,
created_at: datetime(version.created_at)?,
})
})
.transpose()?,
),
created_at: datetime(value.created_at)?,
updated_at: datetime(value.updated_at)?,
})
}
pub(super) fn credential_caller(principal: &Principal) -> CredentialCaller {
CredentialCaller {
principal_id: principal.id.clone(),
administrator: principal.administrator,
}
}

View file

@ -0,0 +1,100 @@
use super::*;
pub(super) async fn soft_delete_job(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
lifecycle_mutation(state, http, path, LifecycleAction::SoftDelete).await
}
pub(super) async fn restore_job(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
lifecycle_mutation(state, http, path, LifecycleAction::Restore).await
}
pub(super) async fn lifecycle_mutation(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
action: LifecycleAction,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let job_id = parse_id(path.into_inner())?;
let key = idempotency_key(&http)?.to_owned();
let (operation, management) = match action {
LifecycleAction::SoftDelete => ("job.soft_delete", ManagementOperation::SoftDelete),
LifecycleAction::Restore => ("job.restore", ManagementOperation::Restore),
};
authorize_job(&state, &principal, &job_id, management).await?;
let action = match action {
LifecycleAction::SoftDelete => JobMutationAction::SoftDelete,
LifecycleAction::Restore => JobMutationAction::Restore,
};
let job = idempotent_job_mutation(&state, &principal, &job_id, &key, operation, action).await?;
Ok(with_idempotency_expiry(HttpResponse::Ok().json(job)))
}
pub(super) async fn hard_delete_job(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let job_id = parse_id(path.into_inner())?;
authorize_job(&state, &principal, &job_id, ManagementOperation::HardDelete).await?;
let key = idempotency_key(&http)?.to_owned();
let request_fingerprint =
fingerprint(&json!({ "job_id": job_id })).map_err(|_| PublicApiError::internal())?;
let principal_for_effect = principal.clone();
let job_for_effect = job_id.clone();
let outcome = execute_mutation(
&state.pool,
MutationScope {
subject: &principal.id,
operation: "job.hard_delete",
key: &key,
},
"hard_delete_job",
&[],
request_fingerprint,
move |tx| {
Box::pin(async move {
let outcome =
DeletionService::hard_delete_in_tx(tx, &principal_for_effect, &job_for_effect)
.await
.map_err(DurableError::effect)?;
Ok((
MutationStatus::Accepted,
deletion_projection(&outcome.operation)
.map_err(|error| DurableError::Effect(error.to_string()))?,
))
})
},
)
.await?;
if outcome.disposition == MutationDisposition::Conflict {
return Err(PublicApiError::idempotency_conflict());
}
Ok(with_idempotency_expiry(
HttpResponse::Accepted().json(outcome.response),
))
}
pub(super) async fn get_deletion_operation(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let operation_id = parse_id(path.into_inner())?;
let operation = state
.deletion
.operation(&principal, &operation_id)
.await
.map_err(PublicApiError::from_deletion)?;
Ok(HttpResponse::Ok().json(deletion_projection(&operation)?))
}

View file

@ -21,6 +21,21 @@ use super::{
current_progress, encode_event_cursor, event_cursor, load_job_row, parse_id, request_id, current_progress, encode_event_cursor, event_cursor, load_job_row, parse_id, request_id,
}; };
mod connection;
mod frames;
mod progress;
mod projections;
mod tickets;
#[cfg(test)]
use connection::decode_event_cursor;
pub(super) use connection::{connect, prune_history};
use frames::*;
use progress::*;
use projections::*;
pub(super) use tickets::issue_ticket;
use tickets::redeem_ticket;
const EVENT_TOPICS: [&str; 2] = ["job.updated", "deletion.updated"]; const EVENT_TOPICS: [&str; 2] = ["job.updated", "deletion.updated"];
const EVENT_BATCH: i64 = 100; const EVENT_BATCH: i64 = 100;
@ -46,366 +61,11 @@ struct TicketScope {
prefix: String, prefix: String,
} }
pub(super) async fn issue_ticket(
state: web::Data<PublicApiState>,
request: HttpRequest,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&request).await?;
let origin = allowed_origin(&state, &request)?;
let ticket = random_ticket();
let digest = ticket_digest(ticket.as_str());
let mut tx = state.pool.begin().await?;
sqlx::query("DELETE FROM live_event_tickets WHERE expires_at <= now()")
.execute(tx.as_mut())
.await?;
let expires_at: OffsetDateTime = sqlx::query_scalar(
"INSERT INTO live_event_tickets (
ticket_digest, principal, origin, created_at, expires_at
) VALUES (
$1, $2, $3, now(),
now() + $4 * interval '1 second'
)
RETURNING expires_at",
)
.bind(digest.as_slice())
.bind(Json(ticket_principal_projection(&principal)))
.bind(&origin)
.bind(state.event_ticket_ttl.as_secs_f64())
.fetch_one(tx.as_mut())
.await?;
tx.commit().await?;
Ok(HttpResponse::Created()
.insert_header((header::CACHE_CONTROL, "no-store"))
.insert_header((header::PRAGMA, "no-cache"))
.json(EventTicketDto {
ticket,
expires_at: event_time(expires_at)?,
}))
}
pub(super) async fn connect(
state: web::Data<PublicApiState>,
request: HttpRequest,
body: web::Payload,
query: web::Query<EventQuery>,
) -> Result<HttpResponse, PublicApiError> {
let bearer = request.headers().contains_key(header::AUTHORIZATION);
let principal = match (bearer, query.ticket.as_deref()) {
(true, None) => state.principal(&request).await?,
(false, Some(ticket)) => redeem_ticket(&state, &request, ticket).await?,
(false, None) => return Err(PublicApiError::unauthenticated()),
(true, Some(_)) => return Err(PublicApiError::invalid_request()),
};
let now = OffsetDateTime::now_utc();
let retention =
time::Duration::try_from(state.event_retention).map_err(|_| PublicApiError::internal())?;
prune_history(&state.pool, now - retention).await?;
let start = classify_cursor(&state, &principal, query.cursor.as_deref(), now).await?;
let (response, session, messages) =
actix_ws::handle(&request, body).map_err(|_| PublicApiError::invalid_request())?;
let state = state.get_ref().clone();
actix_web::rt::spawn(async move {
match start {
EventStart::Replay(after_id) => {
run_connection(state, principal, after_id, session, messages).await;
}
EventStart::Resync(reason) => {
send_resync(state, principal, reason, session).await;
}
}
});
Ok(response)
}
fn allowed_origin(state: &PublicApiState, request: &HttpRequest) -> Result<String, PublicApiError> {
let mut origins = request.headers().get_all(header::ORIGIN);
let origin = origins
.next()
.and_then(|value| value.to_str().ok())
.ok_or_else(PublicApiError::forbidden)?;
if origins.next().is_some() || !state.event_allowed_origins.contains(origin) {
return Err(PublicApiError::forbidden());
}
Ok(origin.to_owned())
}
async fn redeem_ticket(
state: &PublicApiState,
request: &HttpRequest,
ticket: &str,
) -> Result<Principal, PublicApiError> {
let origin = allowed_origin(state, request)?;
let digest = ticket_digest(ticket);
let projection: Option<Json<TicketPrincipal>> = sqlx::query_scalar(
"DELETE FROM live_event_tickets
WHERE ticket_digest = $1 AND origin = $2 AND expires_at > now()
RETURNING principal",
)
.bind(digest.as_slice())
.bind(origin)
.fetch_optional(&state.pool)
.await?;
projection
.ok_or_else(PublicApiError::unauthenticated)
.and_then(|projection| ticket_principal(projection.0))
}
fn ticket_principal_projection(principal: &Principal) -> TicketPrincipal {
TicketPrincipal {
id: principal.id.as_str().to_owned(),
issuer: principal.issuer.clone(),
subject: principal.subject.clone(),
administrator: principal.administrator,
scopes: principal
.scopes
.iter()
.map(|scope| TicketScope {
store: scope.store.as_str().to_owned(),
bucket: scope.bucket.as_str().to_owned(),
prefix: scope.prefix.canonical(),
})
.collect(),
}
}
fn ticket_principal(projection: TicketPrincipal) -> Result<Principal, PublicApiError> {
let scopes = projection
.scopes
.into_iter()
.map(|scope| {
Ok(StorageScope {
store: TargetStoreId::parse(scope.store).map_err(|_| PublicApiError::internal())?,
bucket: BucketName::parse(scope.bucket).map_err(|_| PublicApiError::internal())?,
prefix: TargetPrefix::parse(&scope.prefix, std::iter::empty())
.map_err(|_| PublicApiError::internal())?,
})
})
.collect::<Result<Vec<_>, PublicApiError>>()?;
Ok(Principal {
id: OpaqueId::parse(projection.id).map_err(|_| PublicApiError::internal())?,
issuer: projection.issuer,
subject: projection.subject,
administrator: projection.administrator,
scopes,
})
}
fn random_ticket() -> OpaqueId {
let mut bytes = [0_u8; 32];
getrandom::fill(&mut bytes).expect("operating system randomness is required");
let mut value = String::from("t");
for byte in bytes {
use std::fmt::Write as _;
write!(value, "{byte:02x}").expect("writing to String cannot fail");
}
OpaqueId::parse(value).expect("random event ticket is valid")
}
fn ticket_digest(ticket: &str) -> [u8; 32] {
Sha256::digest(ticket.as_bytes()).into()
}
pub(super) async fn prune_history(
pool: &PgPool,
before: OffsetDateTime,
) -> Result<u64, DurableError> {
let mut tx = pool.begin().await?;
let pruned: Vec<i64> = sqlx::query_scalar(
"DELETE FROM outbox_messages
WHERE topic = ANY($1) AND created_at < $2
RETURNING id",
)
.bind(&EVENT_TOPICS[..])
.bind(before)
.fetch_all(tx.as_mut())
.await?;
if let Some(floor) = pruned.iter().max() {
sqlx::query(
"UPDATE live_event_history_state
SET floor_outbox_id = GREATEST(floor_outbox_id, $1)
WHERE singleton",
)
.bind(floor)
.execute(tx.as_mut())
.await?;
}
tx.commit().await?;
Ok(pruned.len() as u64)
}
enum EventStart { enum EventStart {
Replay(i64), Replay(i64),
Resync(ResyncReasonDto), Resync(ResyncReasonDto),
} }
async fn classify_cursor(
state: &PublicApiState,
principal: &Principal,
token: Option<&str>,
now: OffsetDateTime,
) -> Result<EventStart, PublicApiError> {
let Some(token) = token else {
return Ok(EventStart::Resync(ResyncReasonDto::CursorAbsent));
};
let after_id =
match decode_event_cursor(&state.cursors, state.event_retention, principal, token, now)? {
Ok(after_id) => after_id,
Err(reason) => return Ok(EventStart::Resync(reason)),
};
let (floor, maximum): (i64, i64) = sqlx::query_as(
"SELECT history.floor_outbox_id,
GREATEST(
COALESCE((SELECT MAX(id) FROM outbox_messages), 0),
history.floor_outbox_id
)
FROM live_event_history_state history
WHERE history.singleton",
)
.fetch_one(&state.pool)
.await?;
if after_id < floor || after_id > maximum {
return Ok(EventStart::Resync(ResyncReasonDto::HistoryUnavailable));
}
Ok(EventStart::Replay(after_id))
}
fn decode_event_cursor(
cursors: &remora_durable::CursorCodec,
retention: std::time::Duration,
principal: &Principal,
token: &str,
now: OffsetDateTime,
) -> Result<Result<i64, ResyncReasonDto>, PublicApiError> {
let cursor = match cursors.decode(token) {
Ok(cursor) => cursor,
Err(_) => return Ok(Err(ResyncReasonDto::CursorInvalid)),
};
let Some(after_id) = event_position(&cursor) else {
return Ok(Err(ResyncReasonDto::CursorInvalid));
};
let expected = remora_durable::CursorCodec::filter_fingerprint(&serde_json::json!({
"resource": "events",
"authorization": authorization_snapshot(principal)
}))?;
if cursor.subject != principal.id.as_str() || cursor.filter_fingerprint != expected {
return Ok(Err(ResyncReasonDto::AuthorizationChanged));
}
let retention = time::Duration::try_from(retention).map_err(|_| PublicApiError::internal())?;
if cursor.high_water.created_at + retention < now {
return Ok(Err(ResyncReasonDto::CursorExpired));
}
Ok(Ok(after_id))
}
fn event_position(cursor: &KeysetCursor) -> Option<i64> {
let high_water = cursor
.high_water
.id
.strip_prefix("event-")?
.parse::<i64>()
.ok()?;
let after = cursor.after.as_ref()?;
let after_id = after.id.strip_prefix("event-")?.parse::<i64>().ok()?;
(high_water == after_id && after.created_at == cursor.high_water.created_at).then_some(after_id)
}
async fn send_resync(
state: PublicApiState,
principal: Principal,
reason: ResyncReasonDto,
mut session: actix_ws::Session,
) {
let Ok(cursor) = event_cursor(&state, &principal).await else {
let _ = session.close(None).await;
return;
};
let Ok(frame) = resync_frame(reason, 1, &cursor, OffsetDateTime::now_utc()) else {
let _ = session.close(None).await;
return;
};
let _ = send_frame(&mut session, &frame).await;
let _ = session.close(None).await;
}
async fn run_connection(
state: PublicApiState,
principal: Principal,
mut after_id: i64,
mut session: actix_ws::Session,
mut messages: actix_ws::MessageStream,
) {
let mut sequence = 0_u64;
let mut poll = tokio::time::interval(state.event_poll_interval);
let mut heartbeat = tokio::time::interval(state.event_heartbeat_interval);
poll.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
heartbeat.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
heartbeat.tick().await;
loop {
tokio::select! {
message = messages.next() => {
match message {
Some(Ok(actix_ws::Message::Ping(bytes))) => {
if session.pong(&bytes).await.is_err() {
break;
}
}
Some(Ok(actix_ws::Message::Pong(_))) => {}
Some(Ok(actix_ws::Message::Close(reason))) => {
let _ = session.close(reason).await;
return;
}
Some(Err(_)) | None => break,
Some(Ok(_)) => break,
}
}
_ = poll.tick() => {
if poll_progress(&state).await.is_err() {
break;
}
match replay_events(
&state,
&principal,
&mut after_id,
&mut sequence,
&mut session,
).await {
Ok(ReplayStatus::Current) => {}
Ok(ReplayStatus::HistoryUnavailable) => {
let now = OffsetDateTime::now_utc();
let Ok(cursor) = event_cursor(&state, &principal).await else {
break;
};
let Ok(frame) = resync_frame(
ResyncReasonDto::HistoryUnavailable,
next_sequence(&mut sequence),
&cursor,
now,
) else {
break;
};
let _ = send_frame(&mut session, &frame).await;
break;
}
Err(()) => break,
}
}
_ = heartbeat.tick() => {
let now = OffsetDateTime::now_utc();
let Ok(cursor) = encode_event_cursor(&state, &principal, (after_id, now)) else {
break;
};
let Ok(frame) = heartbeat_frame(next_sequence(&mut sequence), &cursor, now) else {
break;
};
if send_frame(&mut session, &frame).await.is_err() {
break;
}
}
}
}
let _ = session.close(None).await;
}
#[derive(FromRow)] #[derive(FromRow)]
struct ReplayRow { struct ReplayRow {
id: i64, id: i64,
@ -428,134 +88,6 @@ enum ReplayStatus {
HistoryUnavailable, HistoryUnavailable,
} }
async fn replay_events(
state: &PublicApiState,
principal: &Principal,
after_id: &mut i64,
sequence: &mut u64,
session: &mut actix_ws::Session,
) -> Result<ReplayStatus, ()> {
for _ in 0..10 {
let batch = sqlx::query_as::<_, ReplayBatchRow>(
"SELECT history.floor_outbox_id, events.id, events.topic,
events.payload, events.created_at
FROM live_event_history_state history
LEFT JOIN LATERAL (
SELECT id, topic, payload, created_at
FROM outbox_messages
WHERE id > $1 AND topic = ANY($2)
ORDER BY id
LIMIT $3
) events ON TRUE
WHERE history.singleton
ORDER BY events.id",
)
.bind(*after_id)
.bind(&EVENT_TOPICS[..])
.bind(EVENT_BATCH)
.fetch_all(&state.pool)
.await
.map_err(|_| ())?;
let Some(first) = batch.first() else {
return Err(());
};
if *after_id < first.floor_outbox_id {
return Ok(ReplayStatus::HistoryUnavailable);
}
let rows = batch
.into_iter()
.filter_map(|row| {
Some(ReplayRow {
id: row.id?,
topic: row.topic?,
payload: row.payload?,
created_at: row.created_at?,
})
})
.collect::<Vec<_>>();
let count = rows.len();
for row in rows {
*after_id = row.id;
let now = OffsetDateTime::now_utc();
let cursor = encode_event_cursor(state, principal, (row.id, now)).map_err(|_| ())?;
let candidate_sequence = sequence.saturating_add(1);
let frame = match row.topic.as_str() {
"job.updated" => {
project_job_event(state, principal, &row, candidate_sequence, &cursor).await
}
"deletion.updated" => {
project_deletion_event(state, principal, &row, candidate_sequence, &cursor)
.await
}
_ => None,
};
if let Some(frame) = frame {
*sequence = candidate_sequence;
send_frame(session, &frame).await?;
}
}
if count < EVENT_BATCH as usize {
break;
}
}
Ok(ReplayStatus::Current)
}
async fn project_job_event(
state: &PublicApiState,
principal: &Principal,
event: &ReplayRow,
sequence: u64,
cursor: &str,
) -> Option<EventFrameDto> {
let job_id = parse_payload_id(&event.payload, "job_id")?;
let row = load_job_row(&state.pool, &job_id).await.ok()??;
authorize_row(principal, &row, ManagementOperation::Read).ok()?;
let progress = current_progress(state, &row)
.await
.ok()
.flatten()
.map(progress_fingerprint);
job_frame(
&row,
progress.as_ref(),
stable_event_id(event.id),
sequence,
cursor,
event.created_at,
)
.ok()
}
async fn project_deletion_event(
state: &PublicApiState,
principal: &Principal,
event: &ReplayRow,
sequence: u64,
cursor: &str,
) -> Option<EventFrameDto> {
let operation_id = parse_payload_id(&event.payload, "operation_id")?;
let (job_id, state_name): (String, String) =
sqlx::query_as("SELECT job_id, state FROM deletion_operations WHERE operation_id = $1")
.bind(operation_id.as_str())
.fetch_optional(&state.pool)
.await
.ok()??;
let job_id = parse_id(job_id).ok()?;
let row = load_job_row(&state.pool, &job_id).await.ok()??;
authorize_row(principal, &row, ManagementOperation::Read).ok()?;
deletion_frame(
job_id,
operation_id,
parse_deletion_state(&state_name)?,
stable_event_id(event.id),
sequence,
cursor,
event.created_at,
)
.ok()
}
#[derive(Clone, Debug, Eq, PartialEq)] #[derive(Clone, Debug, Eq, PartialEq)]
struct ProgressFingerprint { struct ProgressFingerprint {
attempt_id: OpaqueId, attempt_id: OpaqueId,
@ -566,370 +98,6 @@ struct ProgressFingerprint {
total_members: Option<u64>, total_members: Option<u64>,
} }
fn progress_fingerprint(snapshot: remora_attempt::ProgressSnapshot) -> ProgressFingerprint {
ProgressFingerprint {
attempt_id: snapshot.attempt_id,
fence: snapshot.fence,
completed_bytes: snapshot.progress.completed_bytes,
total_bytes: snapshot.progress.total_bytes,
completed_members: snapshot.progress.completed_members,
total_members: snapshot.progress.total_members,
}
}
async fn poll_progress(state: &PublicApiState) -> Result<(), ()> {
let delay = state.event_poll_interval.as_secs_f64();
let claimed = sqlx::query_scalar::<_, bool>(
"UPDATE live_event_history_state
SET progress_poll_not_before = now() + ($1 * interval '1 second')
WHERE singleton AND progress_poll_not_before <= now()
RETURNING singleton",
)
.bind(delay)
.fetch_optional(&state.pool)
.await
.map_err(|_| ())?
.is_some();
if !claimed {
return Ok(());
}
let mut after_job = None::<String>;
loop {
let job_ids: Vec<String> = sqlx::query_scalar(
"SELECT job_id FROM jobs
WHERE phase IN ('running', 'cancelling')
AND ($1::text IS NULL OR job_id > $1)
ORDER BY job_id
LIMIT 200",
)
.bind(after_job.as_deref())
.fetch_all(&state.pool)
.await
.map_err(|_| ())?;
if job_ids.is_empty() {
break;
}
let count = job_ids.len();
for job_id in job_ids {
after_job = Some(job_id.clone());
let Some(row) = load_job_row(&state.pool, &parse_id(job_id.clone()).map_err(|_| ())?)
.await
.map_err(|_| ())?
else {
continue;
};
let progress = current_progress(state, &row)
.await
.ok()
.flatten()
.map(progress_fingerprint);
record_progress_change(&state.pool, &row, progress.as_ref())
.await
.map_err(|_| ())?;
}
if count < 200 {
break;
}
}
Ok(())
}
async fn record_progress_change(
pool: &PgPool,
row: &JobRow,
progress: Option<&ProgressFingerprint>,
) -> Result<(), PublicApiError> {
let Some(attempt_id) = row.authority_attempt_id.as_deref() else {
return Ok(());
};
let mut tx = pool.begin().await?;
let now = OffsetDateTime::now_utc();
let changed = if let Some(progress) = progress {
sqlx::query_scalar::<_, String>(
"INSERT INTO live_progress_event_state (
job_id, attempt_id, fence, progress_present, completed_bytes,
total_bytes, completed_members, total_members, updated_at
) VALUES ($1, $2, $3, TRUE, $4, $5, $6, $7, $8)
ON CONFLICT (job_id) DO UPDATE
SET attempt_id = EXCLUDED.attempt_id,
fence = EXCLUDED.fence,
progress_present = TRUE,
completed_bytes = EXCLUDED.completed_bytes,
total_bytes = EXCLUDED.total_bytes,
completed_members = EXCLUDED.completed_members,
total_members = EXCLUDED.total_members,
updated_at = EXCLUDED.updated_at
WHERE (
live_progress_event_state.attempt_id,
live_progress_event_state.fence,
live_progress_event_state.progress_present,
live_progress_event_state.completed_bytes,
live_progress_event_state.total_bytes,
live_progress_event_state.completed_members,
live_progress_event_state.total_members
) IS DISTINCT FROM (
EXCLUDED.attempt_id,
EXCLUDED.fence,
TRUE,
EXCLUDED.completed_bytes,
EXCLUDED.total_bytes,
EXCLUDED.completed_members,
EXCLUDED.total_members
)
RETURNING job_id",
)
.bind(&row.job_id)
.bind(progress.attempt_id.as_str())
.bind(i64::try_from(progress.fence).map_err(|_| PublicApiError::internal())?)
.bind(i64::try_from(progress.completed_bytes).map_err(|_| PublicApiError::internal())?)
.bind(
progress
.total_bytes
.map(i64::try_from)
.transpose()
.map_err(|_| PublicApiError::internal())?,
)
.bind(i64::try_from(progress.completed_members).map_err(|_| PublicApiError::internal())?)
.bind(
progress
.total_members
.map(i64::try_from)
.transpose()
.map_err(|_| PublicApiError::internal())?,
)
.bind(now)
.fetch_optional(tx.as_mut())
.await?
.is_some()
} else {
sqlx::query_scalar::<_, String>(
"UPDATE live_progress_event_state
SET attempt_id = $2, fence = $3, progress_present = FALSE,
completed_bytes = NULL, total_bytes = NULL,
completed_members = NULL, total_members = NULL, updated_at = $4
WHERE job_id = $1
AND (
attempt_id IS DISTINCT FROM $2
OR fence IS DISTINCT FROM $3
OR progress_present
OR completed_bytes IS NOT NULL
OR total_bytes IS NOT NULL
OR completed_members IS NOT NULL
OR total_members IS NOT NULL
)
RETURNING job_id",
)
.bind(&row.job_id)
.bind(attempt_id)
.bind(row.current_fence)
.bind(now)
.fetch_optional(tx.as_mut())
.await?
.is_some()
};
if changed {
remora_durable::enqueue(
&mut tx,
"job.updated",
serde_json::json!({ "job_id": row.job_id }),
)
.await?;
}
tx.commit().await?;
Ok(())
}
fn job_frame(
row: &JobRow,
progress: Option<&ProgressFingerprint>,
event_id: OpaqueId,
sequence: u64,
cursor: &str,
occurred_at: OffsetDateTime,
) -> Result<EventFrameDto, PublicApiError> {
let authoritative_attempt = row
.authority_attempt_id
.as_deref()
.map(|attempt_id| {
Ok::<_, PublicApiError>(EventAttemptDto {
attempt_id: parse_id(attempt_id)?,
fence: PositiveIntegerDto::new(
u64::try_from(row.current_fence).map_err(|_| PublicApiError::internal())?,
)
.map_err(|_| PublicApiError::internal())?,
})
})
.transpose()?;
let progress = progress
.filter(|progress| {
row.authority_attempt_id.as_deref() == Some(progress.attempt_id.as_str())
&& u64::try_from(row.current_fence).ok() == Some(progress.fence)
})
.map(event_progress)
.transpose()?;
Ok(EventFrameDto::JobChanged(JobChangedDto {
schema_version: SchemaVersionDto,
event_id,
sequence: positive(sequence)?,
cursor: event_cursor_dto(cursor)?,
occurred_at: event_time(occurred_at)?,
job_id: parse_id(row.job_id.clone())?,
job_revision: positive(
u64::try_from(row.revision).map_err(|_| PublicApiError::internal())?,
)?,
phase: parse_phase(&row.phase)?,
lifecycle: parse_lifecycle(&row.lifecycle)?,
authoritative_attempt: NullableDto(authoritative_attempt),
progress: NullableDto(progress),
}))
}
fn deletion_frame(
job_id: OpaqueId,
operation_id: OpaqueId,
state: DeletionStateDto,
event_id: OpaqueId,
sequence: u64,
cursor: &str,
occurred_at: OffsetDateTime,
) -> Result<EventFrameDto, PublicApiError> {
Ok(EventFrameDto::DeletionChanged(DeletionChangedDto {
schema_version: SchemaVersionDto,
event_id,
sequence: positive(sequence)?,
cursor: event_cursor_dto(cursor)?,
occurred_at: event_time(occurred_at)?,
job_id,
operation_id,
state,
}))
}
fn resync_frame(
reason: ResyncReasonDto,
sequence: u64,
cursor: &str,
occurred_at: OffsetDateTime,
) -> Result<EventFrameDto, PublicApiError> {
Ok(EventFrameDto::ResyncRequired(ResyncRequiredDto {
schema_version: SchemaVersionDto,
event_id: request_id(),
sequence: positive(sequence)?,
cursor: event_cursor_dto(cursor)?,
occurred_at: event_time(occurred_at)?,
reason,
}))
}
fn heartbeat_frame(
sequence: u64,
cursor: &str,
occurred_at: OffsetDateTime,
) -> Result<EventFrameDto, PublicApiError> {
Ok(EventFrameDto::Heartbeat(HeartbeatDto {
schema_version: SchemaVersionDto,
event_id: request_id(),
sequence: positive(sequence)?,
cursor: event_cursor_dto(cursor)?,
occurred_at: event_time(occurred_at)?,
}))
}
fn event_progress(progress: &ProgressFingerprint) -> Result<EventProgressDto, PublicApiError> {
Ok(EventProgressDto {
attempt_id: progress.attempt_id.clone(),
fence: positive(progress.fence)?,
completed_bytes: safe(progress.completed_bytes)?,
total_bytes: progress
.total_bytes
.map(|value| safe(value).map(|value| NullableDto(Some(value))))
.transpose()?,
completed_members: safe(progress.completed_members)?,
total_members: progress
.total_members
.map(|value| safe(value).map(|value| NullableDto(Some(value))))
.transpose()?,
})
}
fn parse_payload_id(payload: &Value, field: &str) -> Option<OpaqueId> {
payload
.get(field)?
.as_str()
.and_then(|value| OpaqueId::parse(value).ok())
}
fn stable_event_id(id: i64) -> OpaqueId {
OpaqueId::parse(format!("event_{id:016}")).expect("positive outbox ids form opaque event ids")
}
fn parse_phase(value: &str) -> Result<JobPhase, PublicApiError> {
match value {
"queued" => Ok(JobPhase::Queued),
"running" => Ok(JobPhase::Running),
"retry_wait" => Ok(JobPhase::RetryWait),
"cancelling" => Ok(JobPhase::Cancelling),
"succeeded" => Ok(JobPhase::Succeeded),
"failed" => Ok(JobPhase::Failed),
"cancelled" => Ok(JobPhase::Cancelled),
_ => Err(PublicApiError::internal()),
}
}
fn parse_lifecycle(value: &str) -> Result<JobLifecycle, PublicApiError> {
match value {
"visible" => Ok(JobLifecycle::Visible),
"soft_deleted" => Ok(JobLifecycle::SoftDeleted),
"hard_deleting" => Ok(JobLifecycle::HardDeleting),
"hard_deleted" => Ok(JobLifecycle::HardDeleted),
_ => Err(PublicApiError::internal()),
}
}
fn parse_deletion_state(value: &str) -> Option<DeletionStateDto> {
match value {
"queued" => Some(DeletionStateDto::Queued),
"running" => Some(DeletionStateDto::Running),
"retry_wait" => Some(DeletionStateDto::RetryWait),
"blocked" => Some(DeletionStateDto::Blocked),
"completed" => Some(DeletionStateDto::Completed),
_ => None,
}
}
fn positive(value: u64) -> Result<PositiveIntegerDto, PublicApiError> {
PositiveIntegerDto::new(value).map_err(|_| PublicApiError::internal())
}
fn safe(value: u64) -> Result<SafeIntegerDto, PublicApiError> {
SafeIntegerDto::new(value).map_err(|_| PublicApiError::internal())
}
fn event_cursor_dto(value: &str) -> Result<CursorDto, PublicApiError> {
CursorDto::parse(value).map_err(|_| PublicApiError::internal())
}
fn event_time(value: OffsetDateTime) -> Result<DateTimeDto, PublicApiError> {
DateTimeDto::parse(
value
.format(&Rfc3339)
.map_err(|_| PublicApiError::internal())?,
)
.map_err(|_| PublicApiError::internal())
}
fn next_sequence(sequence: &mut u64) -> u64 {
*sequence = sequence.saturating_add(1);
*sequence
}
async fn send_frame(session: &mut actix_ws::Session, frame: &EventFrameDto) -> Result<(), ()> {
session
.text(serde_json::to_string(frame).map_err(|_| ())?)
.await
.map_err(|_| ())
}
#[cfg(test)] #[cfg(test)]
mod tests { mod tests {
use remora_domain::{BucketName, StorageScope, TargetPrefix, TargetStoreId}; use remora_domain::{BucketName, StorageScope, TargetPrefix, TargetStoreId};

View file

@ -0,0 +1,304 @@
use super::*;
pub(in crate::public) async fn connect(
state: web::Data<PublicApiState>,
request: HttpRequest,
body: web::Payload,
query: web::Query<EventQuery>,
) -> Result<HttpResponse, PublicApiError> {
let bearer = request.headers().contains_key(header::AUTHORIZATION);
let principal = match (bearer, query.ticket.as_deref()) {
(true, None) => state.principal(&request).await?,
(false, Some(ticket)) => redeem_ticket(&state, &request, ticket).await?,
(false, None) => return Err(PublicApiError::unauthenticated()),
(true, Some(_)) => return Err(PublicApiError::invalid_request()),
};
let now = OffsetDateTime::now_utc();
let retention =
time::Duration::try_from(state.event_retention).map_err(|_| PublicApiError::internal())?;
prune_history(&state.pool, now - retention).await?;
let start = classify_cursor(&state, &principal, query.cursor.as_deref(), now).await?;
let (response, session, messages) =
actix_ws::handle(&request, body).map_err(|_| PublicApiError::invalid_request())?;
let state = state.get_ref().clone();
actix_web::rt::spawn(async move {
match start {
EventStart::Replay(after_id) => {
run_connection(state, principal, after_id, session, messages).await;
}
EventStart::Resync(reason) => {
send_resync(state, principal, reason, session).await;
}
}
});
Ok(response)
}
pub(in crate::public) async fn prune_history(
pool: &PgPool,
before: OffsetDateTime,
) -> Result<u64, DurableError> {
let mut tx = pool.begin().await?;
let pruned: Vec<i64> = sqlx::query_scalar(
"DELETE FROM outbox_messages
WHERE topic = ANY($1) AND created_at < $2
RETURNING id",
)
.bind(&EVENT_TOPICS[..])
.bind(before)
.fetch_all(tx.as_mut())
.await?;
if let Some(floor) = pruned.iter().max() {
sqlx::query(
"UPDATE live_event_history_state
SET floor_outbox_id = GREATEST(floor_outbox_id, $1)
WHERE singleton",
)
.bind(floor)
.execute(tx.as_mut())
.await?;
}
tx.commit().await?;
Ok(pruned.len() as u64)
}
pub(super) async fn classify_cursor(
state: &PublicApiState,
principal: &Principal,
token: Option<&str>,
now: OffsetDateTime,
) -> Result<EventStart, PublicApiError> {
let Some(token) = token else {
return Ok(EventStart::Resync(ResyncReasonDto::CursorAbsent));
};
let after_id =
match decode_event_cursor(&state.cursors, state.event_retention, principal, token, now)? {
Ok(after_id) => after_id,
Err(reason) => return Ok(EventStart::Resync(reason)),
};
let (floor, maximum): (i64, i64) = sqlx::query_as(
"SELECT history.floor_outbox_id,
GREATEST(
COALESCE((SELECT MAX(id) FROM outbox_messages), 0),
history.floor_outbox_id
)
FROM live_event_history_state history
WHERE history.singleton",
)
.fetch_one(&state.pool)
.await?;
if after_id < floor || after_id > maximum {
return Ok(EventStart::Resync(ResyncReasonDto::HistoryUnavailable));
}
Ok(EventStart::Replay(after_id))
}
pub(super) fn decode_event_cursor(
cursors: &remora_durable::CursorCodec,
retention: std::time::Duration,
principal: &Principal,
token: &str,
now: OffsetDateTime,
) -> Result<Result<i64, ResyncReasonDto>, PublicApiError> {
let cursor = match cursors.decode(token) {
Ok(cursor) => cursor,
Err(_) => return Ok(Err(ResyncReasonDto::CursorInvalid)),
};
let Some(after_id) = event_position(&cursor) else {
return Ok(Err(ResyncReasonDto::CursorInvalid));
};
let expected = remora_durable::CursorCodec::filter_fingerprint(&serde_json::json!({
"resource": "events",
"authorization": authorization_snapshot(principal)
}))?;
if cursor.subject != principal.id.as_str() || cursor.filter_fingerprint != expected {
return Ok(Err(ResyncReasonDto::AuthorizationChanged));
}
let retention = time::Duration::try_from(retention).map_err(|_| PublicApiError::internal())?;
if cursor.high_water.created_at + retention < now {
return Ok(Err(ResyncReasonDto::CursorExpired));
}
Ok(Ok(after_id))
}
pub(super) fn event_position(cursor: &KeysetCursor) -> Option<i64> {
let high_water = cursor
.high_water
.id
.strip_prefix("event-")?
.parse::<i64>()
.ok()?;
let after = cursor.after.as_ref()?;
let after_id = after.id.strip_prefix("event-")?.parse::<i64>().ok()?;
(high_water == after_id && after.created_at == cursor.high_water.created_at).then_some(after_id)
}
pub(super) async fn send_resync(
state: PublicApiState,
principal: Principal,
reason: ResyncReasonDto,
mut session: actix_ws::Session,
) {
let Ok(cursor) = event_cursor(&state, &principal).await else {
let _ = session.close(None).await;
return;
};
let Ok(frame) = resync_frame(reason, 1, &cursor, OffsetDateTime::now_utc()) else {
let _ = session.close(None).await;
return;
};
let _ = send_frame(&mut session, &frame).await;
let _ = session.close(None).await;
}
pub(super) async fn run_connection(
state: PublicApiState,
principal: Principal,
mut after_id: i64,
mut session: actix_ws::Session,
mut messages: actix_ws::MessageStream,
) {
let mut sequence = 0_u64;
let mut poll = tokio::time::interval(state.event_poll_interval);
let mut heartbeat = tokio::time::interval(state.event_heartbeat_interval);
poll.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
heartbeat.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
heartbeat.tick().await;
loop {
tokio::select! {
message = messages.next() => {
match message {
Some(Ok(actix_ws::Message::Ping(bytes))) => {
if session.pong(&bytes).await.is_err() {
break;
}
}
Some(Ok(actix_ws::Message::Pong(_))) => {}
Some(Ok(actix_ws::Message::Close(reason))) => {
let _ = session.close(reason).await;
return;
}
Some(Err(_)) | None => break,
Some(Ok(_)) => break,
}
}
_ = poll.tick() => {
if poll_progress(&state).await.is_err() {
break;
}
match replay_events(
&state,
&principal,
&mut after_id,
&mut sequence,
&mut session,
).await {
Ok(ReplayStatus::Current) => {}
Ok(ReplayStatus::HistoryUnavailable) => {
let now = OffsetDateTime::now_utc();
let Ok(cursor) = event_cursor(&state, &principal).await else {
break;
};
let Ok(frame) = resync_frame(
ResyncReasonDto::HistoryUnavailable,
next_sequence(&mut sequence),
&cursor,
now,
) else {
break;
};
let _ = send_frame(&mut session, &frame).await;
break;
}
Err(()) => break,
}
}
_ = heartbeat.tick() => {
let now = OffsetDateTime::now_utc();
let Ok(cursor) = encode_event_cursor(&state, &principal, (after_id, now)) else {
break;
};
let Ok(frame) = heartbeat_frame(next_sequence(&mut sequence), &cursor, now) else {
break;
};
if send_frame(&mut session, &frame).await.is_err() {
break;
}
}
}
}
let _ = session.close(None).await;
}
pub(super) async fn replay_events(
state: &PublicApiState,
principal: &Principal,
after_id: &mut i64,
sequence: &mut u64,
session: &mut actix_ws::Session,
) -> Result<ReplayStatus, ()> {
for _ in 0..10 {
let batch = sqlx::query_as::<_, ReplayBatchRow>(
"SELECT history.floor_outbox_id, events.id, events.topic,
events.payload, events.created_at
FROM live_event_history_state history
LEFT JOIN LATERAL (
SELECT id, topic, payload, created_at
FROM outbox_messages
WHERE id > $1 AND topic = ANY($2)
ORDER BY id
LIMIT $3
) events ON TRUE
WHERE history.singleton
ORDER BY events.id",
)
.bind(*after_id)
.bind(&EVENT_TOPICS[..])
.bind(EVENT_BATCH)
.fetch_all(&state.pool)
.await
.map_err(|_| ())?;
let Some(first) = batch.first() else {
return Err(());
};
if *after_id < first.floor_outbox_id {
return Ok(ReplayStatus::HistoryUnavailable);
}
let rows = batch
.into_iter()
.filter_map(|row| {
Some(ReplayRow {
id: row.id?,
topic: row.topic?,
payload: row.payload?,
created_at: row.created_at?,
})
})
.collect::<Vec<_>>();
let count = rows.len();
for row in rows {
*after_id = row.id;
let now = OffsetDateTime::now_utc();
let cursor = encode_event_cursor(state, principal, (row.id, now)).map_err(|_| ())?;
let candidate_sequence = sequence.saturating_add(1);
let frame = match row.topic.as_str() {
"job.updated" => {
project_job_event(state, principal, &row, candidate_sequence, &cursor).await
}
"deletion.updated" => {
project_deletion_event(state, principal, &row, candidate_sequence, &cursor)
.await
}
_ => None,
};
if let Some(frame) = frame {
*sequence = candidate_sequence;
send_frame(session, &frame).await?;
}
}
if count < EVENT_BATCH as usize {
break;
}
}
Ok(ReplayStatus::Current)
}

View file

@ -0,0 +1,197 @@
use super::*;
pub(super) fn job_frame(
row: &JobRow,
progress: Option<&ProgressFingerprint>,
event_id: OpaqueId,
sequence: u64,
cursor: &str,
occurred_at: OffsetDateTime,
) -> Result<EventFrameDto, PublicApiError> {
let authoritative_attempt = row
.authority_attempt_id
.as_deref()
.map(|attempt_id| {
Ok::<_, PublicApiError>(EventAttemptDto {
attempt_id: parse_id(attempt_id)?,
fence: PositiveIntegerDto::new(
u64::try_from(row.current_fence).map_err(|_| PublicApiError::internal())?,
)
.map_err(|_| PublicApiError::internal())?,
})
})
.transpose()?;
let progress = progress
.filter(|progress| {
row.authority_attempt_id.as_deref() == Some(progress.attempt_id.as_str())
&& u64::try_from(row.current_fence).ok() == Some(progress.fence)
})
.map(event_progress)
.transpose()?;
Ok(EventFrameDto::JobChanged(JobChangedDto {
schema_version: SchemaVersionDto,
event_id,
sequence: positive(sequence)?,
cursor: event_cursor_dto(cursor)?,
occurred_at: event_time(occurred_at)?,
job_id: parse_id(row.job_id.clone())?,
job_revision: positive(
u64::try_from(row.revision).map_err(|_| PublicApiError::internal())?,
)?,
phase: parse_phase(&row.phase)?,
lifecycle: parse_lifecycle(&row.lifecycle)?,
authoritative_attempt: NullableDto(authoritative_attempt),
progress: NullableDto(progress),
}))
}
pub(super) fn deletion_frame(
job_id: OpaqueId,
operation_id: OpaqueId,
state: DeletionStateDto,
event_id: OpaqueId,
sequence: u64,
cursor: &str,
occurred_at: OffsetDateTime,
) -> Result<EventFrameDto, PublicApiError> {
Ok(EventFrameDto::DeletionChanged(DeletionChangedDto {
schema_version: SchemaVersionDto,
event_id,
sequence: positive(sequence)?,
cursor: event_cursor_dto(cursor)?,
occurred_at: event_time(occurred_at)?,
job_id,
operation_id,
state,
}))
}
pub(super) fn resync_frame(
reason: ResyncReasonDto,
sequence: u64,
cursor: &str,
occurred_at: OffsetDateTime,
) -> Result<EventFrameDto, PublicApiError> {
Ok(EventFrameDto::ResyncRequired(ResyncRequiredDto {
schema_version: SchemaVersionDto,
event_id: request_id(),
sequence: positive(sequence)?,
cursor: event_cursor_dto(cursor)?,
occurred_at: event_time(occurred_at)?,
reason,
}))
}
pub(super) fn heartbeat_frame(
sequence: u64,
cursor: &str,
occurred_at: OffsetDateTime,
) -> Result<EventFrameDto, PublicApiError> {
Ok(EventFrameDto::Heartbeat(HeartbeatDto {
schema_version: SchemaVersionDto,
event_id: request_id(),
sequence: positive(sequence)?,
cursor: event_cursor_dto(cursor)?,
occurred_at: event_time(occurred_at)?,
}))
}
pub(super) fn event_progress(
progress: &ProgressFingerprint,
) -> Result<EventProgressDto, PublicApiError> {
Ok(EventProgressDto {
attempt_id: progress.attempt_id.clone(),
fence: positive(progress.fence)?,
completed_bytes: safe(progress.completed_bytes)?,
total_bytes: progress
.total_bytes
.map(|value| safe(value).map(|value| NullableDto(Some(value))))
.transpose()?,
completed_members: safe(progress.completed_members)?,
total_members: progress
.total_members
.map(|value| safe(value).map(|value| NullableDto(Some(value))))
.transpose()?,
})
}
pub(super) fn parse_payload_id(payload: &Value, field: &str) -> Option<OpaqueId> {
payload
.get(field)?
.as_str()
.and_then(|value| OpaqueId::parse(value).ok())
}
pub(super) fn stable_event_id(id: i64) -> OpaqueId {
OpaqueId::parse(format!("event_{id:016}")).expect("positive outbox ids form opaque event ids")
}
pub(super) fn parse_phase(value: &str) -> Result<JobPhase, PublicApiError> {
match value {
"queued" => Ok(JobPhase::Queued),
"running" => Ok(JobPhase::Running),
"retry_wait" => Ok(JobPhase::RetryWait),
"cancelling" => Ok(JobPhase::Cancelling),
"succeeded" => Ok(JobPhase::Succeeded),
"failed" => Ok(JobPhase::Failed),
"cancelled" => Ok(JobPhase::Cancelled),
_ => Err(PublicApiError::internal()),
}
}
pub(super) fn parse_lifecycle(value: &str) -> Result<JobLifecycle, PublicApiError> {
match value {
"visible" => Ok(JobLifecycle::Visible),
"soft_deleted" => Ok(JobLifecycle::SoftDeleted),
"hard_deleting" => Ok(JobLifecycle::HardDeleting),
"hard_deleted" => Ok(JobLifecycle::HardDeleted),
_ => Err(PublicApiError::internal()),
}
}
pub(super) fn parse_deletion_state(value: &str) -> Option<DeletionStateDto> {
match value {
"queued" => Some(DeletionStateDto::Queued),
"running" => Some(DeletionStateDto::Running),
"retry_wait" => Some(DeletionStateDto::RetryWait),
"blocked" => Some(DeletionStateDto::Blocked),
"completed" => Some(DeletionStateDto::Completed),
_ => None,
}
}
pub(super) fn positive(value: u64) -> Result<PositiveIntegerDto, PublicApiError> {
PositiveIntegerDto::new(value).map_err(|_| PublicApiError::internal())
}
pub(super) fn safe(value: u64) -> Result<SafeIntegerDto, PublicApiError> {
SafeIntegerDto::new(value).map_err(|_| PublicApiError::internal())
}
pub(super) fn event_cursor_dto(value: &str) -> Result<CursorDto, PublicApiError> {
CursorDto::parse(value).map_err(|_| PublicApiError::internal())
}
pub(super) fn event_time(value: OffsetDateTime) -> Result<DateTimeDto, PublicApiError> {
DateTimeDto::parse(
value
.format(&Rfc3339)
.map_err(|_| PublicApiError::internal())?,
)
.map_err(|_| PublicApiError::internal())
}
pub(super) fn next_sequence(sequence: &mut u64) -> u64 {
*sequence = sequence.saturating_add(1);
*sequence
}
pub(super) async fn send_frame(
session: &mut actix_ws::Session,
frame: &EventFrameDto,
) -> Result<(), ()> {
session
.text(serde_json::to_string(frame).map_err(|_| ())?)
.await
.map_err(|_| ())
}

View file

@ -0,0 +1,176 @@
use super::*;
pub(super) fn progress_fingerprint(
snapshot: remora_attempt::ProgressSnapshot,
) -> ProgressFingerprint {
ProgressFingerprint {
attempt_id: snapshot.attempt_id,
fence: snapshot.fence,
completed_bytes: snapshot.progress.completed_bytes,
total_bytes: snapshot.progress.total_bytes,
completed_members: snapshot.progress.completed_members,
total_members: snapshot.progress.total_members,
}
}
pub(super) async fn poll_progress(state: &PublicApiState) -> Result<(), ()> {
let delay = state.event_poll_interval.as_secs_f64();
let claimed = sqlx::query_scalar::<_, bool>(
"UPDATE live_event_history_state
SET progress_poll_not_before = now() + ($1 * interval '1 second')
WHERE singleton AND progress_poll_not_before <= now()
RETURNING singleton",
)
.bind(delay)
.fetch_optional(&state.pool)
.await
.map_err(|_| ())?
.is_some();
if !claimed {
return Ok(());
}
let mut after_job = None::<String>;
loop {
let job_ids: Vec<String> = sqlx::query_scalar(
"SELECT job_id FROM jobs
WHERE phase IN ('running', 'cancelling')
AND ($1::text IS NULL OR job_id > $1)
ORDER BY job_id
LIMIT 200",
)
.bind(after_job.as_deref())
.fetch_all(&state.pool)
.await
.map_err(|_| ())?;
if job_ids.is_empty() {
break;
}
let count = job_ids.len();
for job_id in job_ids {
after_job = Some(job_id.clone());
let Some(row) = load_job_row(&state.pool, &parse_id(job_id.clone()).map_err(|_| ())?)
.await
.map_err(|_| ())?
else {
continue;
};
let progress = current_progress(state, &row)
.await
.ok()
.flatten()
.map(progress_fingerprint);
record_progress_change(&state.pool, &row, progress.as_ref())
.await
.map_err(|_| ())?;
}
if count < 200 {
break;
}
}
Ok(())
}
pub(super) async fn record_progress_change(
pool: &PgPool,
row: &JobRow,
progress: Option<&ProgressFingerprint>,
) -> Result<(), PublicApiError> {
let Some(attempt_id) = row.authority_attempt_id.as_deref() else {
return Ok(());
};
let mut tx = pool.begin().await?;
let now = OffsetDateTime::now_utc();
let changed = if let Some(progress) = progress {
sqlx::query_scalar::<_, String>(
"INSERT INTO live_progress_event_state (
job_id, attempt_id, fence, progress_present, completed_bytes,
total_bytes, completed_members, total_members, updated_at
) VALUES ($1, $2, $3, TRUE, $4, $5, $6, $7, $8)
ON CONFLICT (job_id) DO UPDATE
SET attempt_id = EXCLUDED.attempt_id,
fence = EXCLUDED.fence,
progress_present = TRUE,
completed_bytes = EXCLUDED.completed_bytes,
total_bytes = EXCLUDED.total_bytes,
completed_members = EXCLUDED.completed_members,
total_members = EXCLUDED.total_members,
updated_at = EXCLUDED.updated_at
WHERE (
live_progress_event_state.attempt_id,
live_progress_event_state.fence,
live_progress_event_state.progress_present,
live_progress_event_state.completed_bytes,
live_progress_event_state.total_bytes,
live_progress_event_state.completed_members,
live_progress_event_state.total_members
) IS DISTINCT FROM (
EXCLUDED.attempt_id,
EXCLUDED.fence,
TRUE,
EXCLUDED.completed_bytes,
EXCLUDED.total_bytes,
EXCLUDED.completed_members,
EXCLUDED.total_members
)
RETURNING job_id",
)
.bind(&row.job_id)
.bind(progress.attempt_id.as_str())
.bind(i64::try_from(progress.fence).map_err(|_| PublicApiError::internal())?)
.bind(i64::try_from(progress.completed_bytes).map_err(|_| PublicApiError::internal())?)
.bind(
progress
.total_bytes
.map(i64::try_from)
.transpose()
.map_err(|_| PublicApiError::internal())?,
)
.bind(i64::try_from(progress.completed_members).map_err(|_| PublicApiError::internal())?)
.bind(
progress
.total_members
.map(i64::try_from)
.transpose()
.map_err(|_| PublicApiError::internal())?,
)
.bind(now)
.fetch_optional(tx.as_mut())
.await?
.is_some()
} else {
sqlx::query_scalar::<_, String>(
"UPDATE live_progress_event_state
SET attempt_id = $2, fence = $3, progress_present = FALSE,
completed_bytes = NULL, total_bytes = NULL,
completed_members = NULL, total_members = NULL, updated_at = $4
WHERE job_id = $1
AND (
attempt_id IS DISTINCT FROM $2
OR fence IS DISTINCT FROM $3
OR progress_present
OR completed_bytes IS NOT NULL
OR total_bytes IS NOT NULL
OR completed_members IS NOT NULL
OR total_members IS NOT NULL
)
RETURNING job_id",
)
.bind(&row.job_id)
.bind(attempt_id)
.bind(row.current_fence)
.bind(now)
.fetch_optional(tx.as_mut())
.await?
.is_some()
};
if changed {
remora_durable::enqueue(
&mut tx,
"job.updated",
serde_json::json!({ "job_id": row.job_id }),
)
.await?;
}
tx.commit().await?;
Ok(())
}

View file

@ -0,0 +1,56 @@
use super::*;
pub(super) async fn project_job_event(
state: &PublicApiState,
principal: &Principal,
event: &ReplayRow,
sequence: u64,
cursor: &str,
) -> Option<EventFrameDto> {
let job_id = parse_payload_id(&event.payload, "job_id")?;
let row = load_job_row(&state.pool, &job_id).await.ok()??;
authorize_row(principal, &row, ManagementOperation::Read).ok()?;
let progress = current_progress(state, &row)
.await
.ok()
.flatten()
.map(progress_fingerprint);
job_frame(
&row,
progress.as_ref(),
stable_event_id(event.id),
sequence,
cursor,
event.created_at,
)
.ok()
}
pub(super) async fn project_deletion_event(
state: &PublicApiState,
principal: &Principal,
event: &ReplayRow,
sequence: u64,
cursor: &str,
) -> Option<EventFrameDto> {
let operation_id = parse_payload_id(&event.payload, "operation_id")?;
let (job_id, state_name): (String, String) =
sqlx::query_as("SELECT job_id, state FROM deletion_operations WHERE operation_id = $1")
.bind(operation_id.as_str())
.fetch_optional(&state.pool)
.await
.ok()??;
let job_id = parse_id(job_id).ok()?;
let row = load_job_row(&state.pool, &job_id).await.ok()??;
authorize_row(principal, &row, ManagementOperation::Read).ok()?;
deletion_frame(
job_id,
operation_id,
parse_deletion_state(&state_name)?,
stable_event_id(event.id),
sequence,
cursor,
event.created_at,
)
.ok()
}

View file

@ -0,0 +1,129 @@
use super::*;
pub(in crate::public) async fn issue_ticket(
state: web::Data<PublicApiState>,
request: HttpRequest,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&request).await?;
let origin = allowed_origin(&state, &request)?;
let ticket = random_ticket();
let digest = ticket_digest(ticket.as_str());
let mut tx = state.pool.begin().await?;
sqlx::query("DELETE FROM live_event_tickets WHERE expires_at <= now()")
.execute(tx.as_mut())
.await?;
let expires_at: OffsetDateTime = sqlx::query_scalar(
"INSERT INTO live_event_tickets (
ticket_digest, principal, origin, created_at, expires_at
) VALUES (
$1, $2, $3, now(),
now() + $4 * interval '1 second'
)
RETURNING expires_at",
)
.bind(digest.as_slice())
.bind(Json(ticket_principal_projection(&principal)))
.bind(&origin)
.bind(state.event_ticket_ttl.as_secs_f64())
.fetch_one(tx.as_mut())
.await?;
tx.commit().await?;
Ok(HttpResponse::Created()
.insert_header((header::CACHE_CONTROL, "no-store"))
.insert_header((header::PRAGMA, "no-cache"))
.json(EventTicketDto {
ticket,
expires_at: event_time(expires_at)?,
}))
}
pub(super) fn allowed_origin(
state: &PublicApiState,
request: &HttpRequest,
) -> Result<String, PublicApiError> {
let mut origins = request.headers().get_all(header::ORIGIN);
let origin = origins
.next()
.and_then(|value| value.to_str().ok())
.ok_or_else(PublicApiError::forbidden)?;
if origins.next().is_some() || !state.event_allowed_origins.contains(origin) {
return Err(PublicApiError::forbidden());
}
Ok(origin.to_owned())
}
pub(super) async fn redeem_ticket(
state: &PublicApiState,
request: &HttpRequest,
ticket: &str,
) -> Result<Principal, PublicApiError> {
let origin = allowed_origin(state, request)?;
let digest = ticket_digest(ticket);
let projection: Option<Json<TicketPrincipal>> = sqlx::query_scalar(
"DELETE FROM live_event_tickets
WHERE ticket_digest = $1 AND origin = $2 AND expires_at > now()
RETURNING principal",
)
.bind(digest.as_slice())
.bind(origin)
.fetch_optional(&state.pool)
.await?;
projection
.ok_or_else(PublicApiError::unauthenticated)
.and_then(|projection| ticket_principal(projection.0))
}
pub(super) fn ticket_principal_projection(principal: &Principal) -> TicketPrincipal {
TicketPrincipal {
id: principal.id.as_str().to_owned(),
issuer: principal.issuer.clone(),
subject: principal.subject.clone(),
administrator: principal.administrator,
scopes: principal
.scopes
.iter()
.map(|scope| TicketScope {
store: scope.store.as_str().to_owned(),
bucket: scope.bucket.as_str().to_owned(),
prefix: scope.prefix.canonical(),
})
.collect(),
}
}
pub(super) fn ticket_principal(projection: TicketPrincipal) -> Result<Principal, PublicApiError> {
let scopes = projection
.scopes
.into_iter()
.map(|scope| {
Ok(StorageScope {
store: TargetStoreId::parse(scope.store).map_err(|_| PublicApiError::internal())?,
bucket: BucketName::parse(scope.bucket).map_err(|_| PublicApiError::internal())?,
prefix: TargetPrefix::parse(&scope.prefix, std::iter::empty())
.map_err(|_| PublicApiError::internal())?,
})
})
.collect::<Result<Vec<_>, PublicApiError>>()?;
Ok(Principal {
id: OpaqueId::parse(projection.id).map_err(|_| PublicApiError::internal())?,
issuer: projection.issuer,
subject: projection.subject,
administrator: projection.administrator,
scopes,
})
}
pub(super) fn random_ticket() -> OpaqueId {
let mut bytes = [0_u8; 32];
getrandom::fill(&mut bytes).expect("operating system randomness is required");
let mut value = String::from("t");
for byte in bytes {
use std::fmt::Write as _;
write!(value, "{byte:02x}").expect("writing to String cannot fail");
}
OpaqueId::parse(value).expect("random event ticket is valid")
}
pub(super) fn ticket_digest(ticket: &str) -> [u8; 32] {
Sha256::digest(ticket.as_bytes()).into()
}

View file

@ -0,0 +1,354 @@
use super::*;
pub(super) async fn idempotent_job_mutation(
state: &PublicApiState,
principal: &Principal,
job_id: &OpaqueId,
key: &str,
operation: &'static str,
action: JobMutationAction,
) -> Result<Value, PublicApiError> {
let request_fingerprint =
fingerprint(&json!({ "job_id": job_id })).map_err(|_| PublicApiError::internal())?;
let state_for_projection = state.clone();
let principal_for_projection = principal.clone();
let job_for_projection = job_id.clone();
let outcome = execute_mutation(
&state.pool,
MutationScope {
subject: &principal.id,
operation,
key,
},
operation,
&[],
request_fingerprint,
move |tx| {
Box::pin(async move {
match action {
JobMutationAction::Cancel => {
AttemptService::request_cancel_in_tx(tx, &job_for_projection).await?;
}
JobMutationAction::SoftDelete => {
DeletionService::soft_delete_in_tx(
tx,
&principal_for_projection,
&job_for_projection,
)
.await
.map_err(DurableError::effect)?;
}
JobMutationAction::Restore => {
DeletionService::restore_in_tx(
tx,
&principal_for_projection,
&job_for_projection,
)
.await
.map_err(DurableError::effect)?;
}
JobMutationAction::Revoke { grant_id } => {
remora_admission::AdmissionService::revoke_execution_grant_in_tx(
tx,
&principal_for_projection,
&grant_id,
)
.await?;
AttemptService::request_cancel_in_tx(tx, &job_for_projection).await?;
}
}
let projection = load_authorized_job_in_tx(
tx,
&state_for_projection,
&principal_for_projection,
&job_for_projection,
true,
)
.await
.map_err(|error| DurableError::Effect(error.to_string()))?;
Ok((MutationStatus::Accepted, projection))
})
},
)
.await?;
if outcome.disposition == MutationDisposition::Conflict {
Err(PublicApiError::idempotency_conflict())
} else {
Ok(outcome.response)
}
}
pub(super) async fn load_job_row(
pool: &PgPool,
job_id: &OpaqueId,
) -> Result<Option<JobRow>, PublicApiError> {
Ok(sqlx::query_as::<_, JobRow>(
"SELECT job_id, owner_id, label, revision, phase, lifecycle, source_projection,
target_store, target_bucket, target_prefix, execution_grant_id,
authority_attempt_id,
(SELECT worker_incarnation FROM attempts WHERE attempt_id = jobs.authority_attempt_id)
AS authority_worker_incarnation,
current_fence, checkpoint, committed_result, failure,
deletion_operation_id, hard_delete_started_at, created_at, updated_at
FROM jobs WHERE job_id = $1",
)
.bind(job_id.as_str())
.fetch_optional(pool)
.await?)
}
pub(super) async fn load_job_row_in_tx(
tx: &mut PgTx<'_>,
job_id: &OpaqueId,
) -> Result<Option<JobRow>, PublicApiError> {
Ok(sqlx::query_as::<_, JobRow>(
"SELECT job_id, owner_id, label, revision, phase, lifecycle, source_projection,
target_store, target_bucket, target_prefix, execution_grant_id,
authority_attempt_id,
(SELECT worker_incarnation FROM attempts WHERE attempt_id = jobs.authority_attempt_id)
AS authority_worker_incarnation,
current_fence, checkpoint, committed_result, failure,
deletion_operation_id, hard_delete_started_at, created_at, updated_at
FROM jobs WHERE job_id = $1",
)
.bind(job_id.as_str())
.fetch_optional(tx.as_mut())
.await?)
}
pub(super) async fn load_authorized_job_in_tx(
tx: &mut PgTx<'_>,
state: &PublicApiState,
principal: &Principal,
job_id: &OpaqueId,
include_deleted: bool,
) -> Result<Value, PublicApiError> {
let row = load_job_row_in_tx(tx, job_id)
.await?
.ok_or_else(PublicApiError::not_found)?;
if !include_deleted && row.lifecycle != "visible" {
return Err(PublicApiError::not_found());
}
authorize_row(principal, &row, ManagementOperation::Read)?;
let cursor = event_cursor_in_tx(tx, state, principal).await?;
job_row_projection(&row, None, &cursor)
}
pub(super) async fn load_authorized_job(
state: &PublicApiState,
principal: &Principal,
job_id: &OpaqueId,
include_deleted: bool,
) -> Result<Value, PublicApiError> {
let cursor = event_cursor(state, principal).await?;
let row = load_job_row(&state.pool, job_id)
.await?
.ok_or_else(PublicApiError::not_found)?;
if !include_deleted && row.lifecycle != "visible" {
return Err(PublicApiError::not_found());
}
authorize_row(principal, &row, ManagementOperation::Read)?;
let progress = current_progress(state, &row).await?;
job_row_projection(&row, progress, &cursor)
}
pub(super) async fn load_visible_job_row(
state: &PublicApiState,
principal: &Principal,
job_id: &OpaqueId,
) -> Result<JobRow, PublicApiError> {
let row = load_job_row(&state.pool, job_id)
.await?
.ok_or_else(PublicApiError::not_found)?;
if row.lifecycle != "visible" {
return Err(PublicApiError::not_found());
}
authorize_row(principal, &row, ManagementOperation::Read)?;
Ok(row)
}
pub(super) async fn authorize_job(
state: &PublicApiState,
principal: &Principal,
job_id: &OpaqueId,
operation: ManagementOperation,
) -> Result<JobRow, PublicApiError> {
let row = load_job_row(&state.pool, job_id)
.await?
.ok_or_else(PublicApiError::not_found)?;
authorize_row(principal, &row, operation)?;
Ok(row)
}
pub(super) fn authorize_row(
principal: &Principal,
row: &JobRow,
operation: ManagementOperation,
) -> Result<(), PublicApiError> {
let resource = AuthorizedResource {
owner: parse_id(row.owner_id.clone())?,
target: row_target(row)?,
};
if may_manage(principal, &resource, operation) {
Ok(())
} else {
Err(PublicApiError::not_found())
}
}
pub(super) async fn load_job_page_rows(
pool: &PgPool,
principal: &Principal,
include_deleted: bool,
cursor: Option<&KeysetCursor>,
wanted: u32,
) -> Result<Vec<JobRow>, PublicApiError> {
let mut after = cursor.and_then(|value| value.after.clone());
let high_water = cursor.map(|value| value.high_water.clone());
let mut visible = Vec::new();
loop {
let rows = sqlx::query_as::<_, JobRow>(
"SELECT job_id, owner_id, label, revision, phase, lifecycle, source_projection,
target_store, target_bucket, target_prefix, execution_grant_id,
authority_attempt_id,
(SELECT worker_incarnation FROM attempts WHERE attempt_id = jobs.authority_attempt_id)
AS authority_worker_incarnation,
current_fence, checkpoint, committed_result, failure,
deletion_operation_id, hard_delete_started_at, created_at, updated_at
FROM jobs
WHERE ($1::text IS NULL OR owner_id = $1)
AND ($2 OR lifecycle = 'visible')
AND ($3::timestamptz IS NULL OR (created_at, job_id) <= ($3, $4))
AND ($5::timestamptz IS NULL OR (created_at, job_id) < ($5, $6))
ORDER BY created_at DESC, job_id DESC
LIMIT 200",
)
.bind((!principal.administrator).then(|| principal.id.as_str()))
.bind(include_deleted)
.bind(high_water.as_ref().map(|value| value.created_at))
.bind(high_water.as_ref().map(|value| value.id.as_str()))
.bind(after.as_ref().map(|value| value.created_at))
.bind(after.as_ref().map(|value| value.id.as_str()))
.fetch_all(pool)
.await?;
if rows.is_empty() {
break;
}
let count = rows.len();
for row in rows {
after = Some(position(row.created_at, &row.job_id));
if authorize_row(principal, &row, ManagementOperation::List).is_ok() {
visible.push(row);
if visible.len() >= wanted as usize {
return Ok(visible);
}
}
}
if count < 200 {
break;
}
}
Ok(visible)
}
pub(super) async fn current_progress(
state: &PublicApiState,
row: &JobRow,
) -> Result<Option<ProgressSnapshot>, PublicApiError> {
let Some(authority_attempt_id) = row.authority_attempt_id.as_deref() else {
return Ok(None);
};
let Ok(progress) = state
.attempts
.progress_snapshot(&parse_id(row.job_id.clone())?)
.await
else {
return Ok(None);
};
Ok(progress.filter(|progress| {
progress.attempt_id.as_str() == authority_attempt_id
&& i64::try_from(progress.fence).ok() == Some(row.current_fence)
}))
}
pub(super) async fn event_cursor(
state: &PublicApiState,
principal: &Principal,
) -> Result<String, PublicApiError> {
let high_water = sqlx::query_as::<_, (i64, OffsetDateTime)>(
"SELECT GREATEST(
COALESCE((SELECT MAX(id) FROM outbox_messages), 0),
history.floor_outbox_id
),
now()
FROM live_event_history_state history
WHERE history.singleton",
)
.fetch_one(&state.pool)
.await?;
encode_event_cursor(state, principal, high_water)
}
pub(super) async fn event_cursor_in_tx(
tx: &mut PgTx<'_>,
state: &PublicApiState,
principal: &Principal,
) -> Result<String, PublicApiError> {
let high_water = sqlx::query_as::<_, (i64, OffsetDateTime)>(
"SELECT GREATEST(
COALESCE((SELECT MAX(id) FROM outbox_messages), 0),
history.floor_outbox_id
),
now()
FROM live_event_history_state history
WHERE history.singleton",
)
.fetch_one(tx.as_mut())
.await?;
encode_event_cursor(state, principal, high_water)
}
pub(super) fn encode_event_cursor(
state: &PublicApiState,
principal: &Principal,
high_water: (i64, OffsetDateTime),
) -> Result<String, PublicApiError> {
let created_at = high_water.1;
let id = format!("event-{}", high_water.0);
let filter = CursorCodec::filter_fingerprint(&json!({
"resource": "events",
"authorization": authorization_snapshot(principal)
}))?;
state
.cursors
.encode(&first_page_cursor(
principal.id.as_str(),
filter,
KeysetPosition {
created_at,
id: id.clone(),
},
KeysetPosition { created_at, id },
))
.map_err(Into::into)
}
pub(super) fn authorization_snapshot(principal: &Principal) -> Value {
let mut scopes = principal
.scopes
.iter()
.map(|scope| {
(
scope.store.as_str(),
scope.bucket.as_str(),
scope.prefix.canonical(),
)
})
.collect::<Vec<_>>();
scopes.sort_unstable();
scopes.dedup();
json!({
"administrator": principal.administrator,
"scopes": scopes
})
}

View file

@ -0,0 +1,281 @@
use super::*;
pub(super) async fn submit_job(
state: web::Data<PublicApiState>,
http: HttpRequest,
request: web::Json<JobSubmissionDto>,
) -> Result<HttpResponse, PublicApiError> {
validate_request("JobSubmission", &*request)?;
let principal = state.principal(&http).await?;
let key = idempotency_key(&http)?;
let request = request.into_inner();
let request_fingerprint =
fingerprint(&request).map_err(|_| PublicApiError::invalid_request())?;
let admission = state.admission.clone();
let state_for_projection = state.get_ref().clone();
let principal_for_effect = principal.clone();
let outcome = execute_mutation(
&state.pool,
MutationScope {
subject: &principal.id,
operation: "job.submit",
key,
},
"submit_job",
&[],
request_fingerprint,
move |tx| {
Box::pin(async move {
match admission
.submit_in_tx(tx, &principal_for_effect, request)
.await?
{
AdmissionDecision::Accepted(accepted) => {
let projection = load_authorized_job_in_tx(
tx,
&state_for_projection,
&principal_for_effect,
&accepted.job_id,
false,
)
.await
.map_err(|error| DurableError::Effect(error.to_string()))?;
Ok((MutationStatus::Accepted, projection))
}
AdmissionDecision::Rejected { .. } => Ok((
MutationStatus::Rejected,
public_error_body(ApiErrorCodeDto::InvalidRequest, "invalid request"),
)),
}
})
},
)
.await?;
if outcome.disposition == MutationDisposition::Conflict {
return Err(PublicApiError::idempotency_conflict());
}
if outcome.status == MutationStatus::Accepted {
let job_id = outcome
.response
.get("id")
.and_then(Value::as_str)
.ok_or_else(PublicApiError::internal)?;
let job_id = parse_id(job_id.to_owned())?;
authorize_job(&state, &principal, &job_id, ManagementOperation::Read).await?;
}
let response = if outcome.status == MutationStatus::Rejected {
HttpResponse::BadRequest()
.content_type("application/problem+json")
.json(outcome.response)
} else {
HttpResponse::Accepted().json(outcome.response)
};
Ok(with_idempotency_expiry(response))
}
pub(super) async fn get_job(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let job_id = parse_id(path.into_inner())?;
let job = load_authorized_job(&state, &principal, &job_id, true).await?;
Ok(HttpResponse::Ok().json(job))
}
pub(super) async fn list_jobs(
state: web::Data<PublicApiState>,
http: HttpRequest,
query: web::Query<JobPageQuery>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let limit = page_limit(query.limit)?;
let filters = json!({
"resource": "jobs",
"include_deleted": query.include_deleted,
"authorization": authorization_snapshot(&principal)
});
let filter = CursorCodec::filter_fingerprint(&filters)?;
let cursor = query
.cursor
.as_deref()
.map(|token| state.cursors.bind(token, principal.id.as_str(), filter))
.transpose()
.map_err(PublicApiError::from_cursor)?;
let event_cursor = event_cursor(&state, &principal).await?;
let mut rows = load_job_page_rows(
&state.pool,
&principal,
query.include_deleted,
cursor.as_ref(),
limit + 1,
)
.await?;
let has_more = rows.len() > limit as usize;
rows.truncate(limit as usize);
let mut items = Vec::with_capacity(rows.len());
for row in &rows {
items.push(job_row_projection(
row,
current_progress(&state, row).await?,
&event_cursor,
)?);
}
let next_cursor = if has_more {
let last = rows.last().ok_or_else(PublicApiError::internal)?;
let last = position(last.created_at, &last.job_id);
let cursor = match cursor.as_ref() {
Some(current) => next_page_cursor(current, last),
None => first_page_cursor(
principal.id.as_str(),
filter,
position(rows[0].created_at, &rows[0].job_id),
last,
),
};
Some(state.cursors.encode(&cursor)?)
} else {
None
};
Ok(HttpResponse::Ok().json(json!({
"items": items,
"next_cursor": next_cursor,
"event_cursor": event_cursor,
"management_capabilities": {
"revoke_execution_grants": principal.administrator
}
})))
}
pub(super) async fn cancel_job(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let job_id = parse_id(path.into_inner())?;
authorize_job(&state, &principal, &job_id, ManagementOperation::Cancel).await?;
let key = idempotency_key(&http)?.to_owned();
let job = idempotent_job_mutation(
&state,
&principal,
&job_id,
&key,
"job.cancel",
JobMutationAction::Cancel,
)
.await?;
Ok(with_idempotency_expiry(HttpResponse::Accepted().json(job)))
}
pub(super) async fn revoke_job(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
if !principal.administrator {
return Err(PublicApiError::forbidden());
}
let job_id = parse_id(path.into_inner())?;
let row = load_job_row(&state.pool, &job_id)
.await?
.ok_or_else(PublicApiError::not_found)?;
let key = idempotency_key(&http)?.to_owned();
let grant_id = parse_id(row.execution_grant_id.clone())?;
let job = idempotent_job_mutation(
&state,
&principal,
&job_id,
&key,
"job.revoke",
JobMutationAction::Revoke { grant_id },
)
.await?;
Ok(with_idempotency_expiry(HttpResponse::Accepted().json(job)))
}
pub(super) async fn list_job_attempts(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
query: web::Query<PageQuery>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let job_id = parse_id(path.into_inner())?;
authorize_job(&state, &principal, &job_id, ManagementOperation::Read).await?;
let limit = page_limit(query.limit)?;
let filter = CursorCodec::filter_fingerprint(&json!({
"resource": "attempts",
"job_id": job_id,
"authorization": authorization_snapshot(&principal)
}))?;
let cursor = query
.cursor
.as_deref()
.map(|token| state.cursors.bind(token, principal.id.as_str(), filter))
.transpose()
.map_err(PublicApiError::from_cursor)?;
let after = cursor
.as_ref()
.and_then(|value| value.after.as_ref())
.map(|value| value.id.parse::<i64>())
.transpose()
.map_err(|_| PublicApiError::cursor_invalid())?;
let rows = sqlx::query_as::<_, AttemptRow>(
"SELECT a.attempt_id, a.fence, a.worker_incarnation, a.state, a.failure,
CASE WHEN j.committed_result->>'attempt_id' = a.attempt_id
THEN j.committed_result END AS result,
a.authorized_at,
COALESCE(a.committed_at, a.stopped_at, a.reaped_at, a.executing_at, a.authorized_at)
AS updated_at
FROM attempts a
JOIN jobs j ON j.job_id = a.job_id
WHERE a.job_id = $1 AND ($2::bigint IS NULL OR a.fence < $2)
ORDER BY a.fence DESC
LIMIT $3",
)
.bind(job_id.as_str())
.bind(after)
.bind(i64::from(limit) + 1)
.fetch_all(&state.pool)
.await?;
let has_more = rows.len() > limit as usize;
let visible = rows.iter().take(limit as usize).collect::<Vec<_>>();
let progress = state
.attempts
.progress_snapshot(&job_id)
.await
.unwrap_or(None);
let items = visible
.iter()
.map(|row| attempt_projection(row, progress.as_ref()))
.collect::<Result<Vec<_>, _>>()?;
let next_cursor = if has_more {
let last = visible.last().ok_or_else(PublicApiError::internal)?;
let next = KeysetPosition {
created_at: last.authorized_at,
id: last.fence.to_string(),
};
let cursor = match cursor.as_ref() {
Some(current) => next_page_cursor(current, next),
None => first_page_cursor(
principal.id.as_str(),
filter,
KeysetPosition {
created_at: rows[0].authorized_at,
id: rows[0].fence.to_string(),
},
next,
),
};
Some(state.cursors.encode(&cursor)?)
} else {
None
};
Ok(HttpResponse::Ok().json(json!({
"items": items,
"next_cursor": next_cursor
})))
}

View file

@ -0,0 +1,297 @@
use super::*;
pub(super) fn progress_projection(progress: &ProgressSnapshot) -> Result<Value, PublicApiError> {
Ok(json!({
"attempt_id": progress.attempt_id,
"fence": progress.fence,
"completed_bytes": progress.progress.completed_bytes,
"total_bytes": progress.progress.total_bytes,
"completed_members": progress.progress.completed_members,
"total_members": progress.progress.total_members,
"observed_at": format_time(progress.observed_at)?
}))
}
pub(super) fn job_row_projection(
row: &JobRow,
progress: Option<ProgressSnapshot>,
event_cursor: &str,
) -> Result<Value, PublicApiError> {
let target = json!({
"store": row.target_store,
"bucket": row.target_bucket,
"prefix": row.target_prefix.0.join("/")
});
if row.lifecycle == "hard_deleted" || row.lifecycle == "hard_deleting" {
return Ok(json!({
"resource_kind": "tombstone",
"id": row.job_id,
"owner": row.owner_id,
"target": target,
"terminal_phase": phase_projection(&row.phase)?,
"deletion_operation_id": row.deletion_operation_id,
"created_at": format_time(row.created_at)?,
"hard_delete_started_at": row.hard_delete_started_at.map(format_time).transpose()?,
"updated_at": format_time(row.updated_at)?,
"event_cursor": event_cursor
}));
}
let authority = row
.authority_attempt_id
.as_ref()
.zip(row.authority_worker_incarnation.as_ref())
.map(|(attempt_id, worker_incarnation)| {
json!({
"attempt_id": attempt_id,
"fence": row.current_fence,
"worker_incarnation": worker_incarnation
})
});
let checkpoint = row
.checkpoint
.as_ref()
.map(|value| checkpoint_projection(&value.0))
.transpose()?;
let result = committed_result(row)?.as_ref().map(result_projection);
let failure = row
.failure
.as_ref()
.map(|value| failure_projection(&value.0))
.transpose()?;
Ok(json!({
"resource_kind": "job",
"id": row.job_id,
"owner": row.owner_id,
"label": row.label,
"revision": row.revision,
"phase": phase_projection(&row.phase)?,
"lifecycle": row.lifecycle,
"source": row.source_projection.0,
"target": target,
"authoritative_attempt": authority,
"progress": progress.as_ref().map(progress_projection).transpose()?,
"checkpoint": checkpoint,
"result": result,
"failure": failure,
"created_at": format_time(row.created_at)?,
"updated_at": format_time(row.updated_at)?,
"event_cursor": event_cursor
}))
}
pub(super) fn phase_projection(value: &str) -> Result<&'static str, PublicApiError> {
match value {
"queued" => Ok("Queued"),
"running" => Ok("Running"),
"retry_wait" => Ok("RetryWait"),
"cancelling" => Ok("Cancelling"),
"succeeded" => Ok("Succeeded"),
"failed" => Ok("Failed"),
"cancelled" => Ok("Cancelled"),
_ => Err(PublicApiError::internal()),
}
}
pub(super) fn checkpoint_projection(value: &Value) -> Result<Value, PublicApiError> {
let state = value
.get("state")
.and_then(Value::as_str)
.ok_or_else(PublicApiError::internal)?;
match state {
"reusable" => {
let checkpoint = value
.get("checkpoint")
.ok_or_else(PublicApiError::internal)?;
Ok(json!({
"reusable": true,
"validated_bytes": checkpoint.get("validated_bytes"),
"validated_members": checkpoint.get("validated_members")
}))
}
"complete" => Ok(json!({
"reusable": false,
"validated_bytes": value.get("validated_bytes"),
"validated_members": value.get("validated_members")
})),
_ => Err(PublicApiError::internal()),
}
}
pub(super) fn failure_projection(value: &Value) -> Result<Value, PublicApiError> {
let class = value
.get("class")
.and_then(Value::as_str)
.ok_or_else(PublicApiError::internal)?;
let disposition = value
.get("disposition")
.and_then(Value::as_str)
.ok_or_else(PublicApiError::internal)?;
Ok(json!({
"class": class,
"message": "job failed",
"retryable": disposition == "retryable"
}))
}
pub(super) fn row_target(row: &JobRow) -> Result<Target, PublicApiError> {
Ok(Target {
store: TargetStoreId::parse(row.target_store.clone())
.map_err(|_| PublicApiError::internal())?,
bucket: BucketName::parse(row.target_bucket.clone())
.map_err(|_| PublicApiError::internal())?,
prefix: TargetPrefix::parse(&row.target_prefix.0.join("/"), std::iter::empty())
.map_err(|_| PublicApiError::internal())?,
})
}
pub(super) fn committed_result(row: &JobRow) -> Result<Option<PublicationInput>, PublicApiError> {
let Some(value) = row.committed_result.as_ref() else {
return Ok(None);
};
let record: CommittedResultRecord =
serde_json::from_value(value.0.clone()).map_err(|_| PublicApiError::internal())?;
Ok(Some(PublicationInput {
attempt_id: record.attempt_id,
manifest_id: record.manifest_id,
schema_version: record.schema_version,
manifest: StoredObject {
identity: ObjectIdentity {
target_store_id: record.manifest.store,
bucket: record.manifest.bucket,
key: record.manifest.key,
version_id: record.manifest.version_id,
},
content_sha256: ContentSha256::parse(record.manifest.sha256)
.map_err(|_| PublicApiError::internal())?,
size: record.manifest.size,
},
member_count: record.member_count,
total_bytes: record.total_bytes,
}))
}
pub(super) fn result_projection(result: &CommittedResult) -> Value {
json!({
"kind": "manifest",
"manifest_id": result.manifest_id,
"schema_version": result.schema_version,
"manifest": object_identity_projection(
&result.manifest.identity.target_store_id,
&result.manifest.identity.bucket,
&result.manifest.identity.key,
result.manifest.identity.version_id.as_deref(),
result.manifest.content_sha256.as_str(),
result.manifest.size
),
"member_count": result.member_count,
"total_bytes": result.total_bytes
})
}
pub(super) fn object_identity_projection(
store: &str,
bucket: &str,
key: &str,
version_id: Option<&str>,
checksum: &str,
size: u64,
) -> Value {
let mut object = json!({
"store": store,
"bucket": bucket,
"key": key,
"checksum": checksum,
"size": size
});
if let Some(version_id) = version_id {
object["version_id"] = json!(version_id);
}
object
}
pub(super) fn attempt_projection(
row: &AttemptRow,
progress: Option<&ProgressSnapshot>,
) -> Result<Value, PublicApiError> {
let state = match row.state.as_str() {
"authorized" => "AttemptAuthorized",
"executing" => "AttemptExecuting",
"reaped" | "stopped" => "AttemptStopped",
"committed" => "AttemptCommitted",
_ => return Err(PublicApiError::internal()),
};
let result = row
.result
.as_ref()
.map(|value| result_record_projection(&value.0))
.transpose()?;
let progress = progress
.filter(|progress| {
progress.attempt_id.as_str() == row.attempt_id
&& i64::try_from(progress.fence).ok() == Some(row.fence)
})
.map(progress_projection)
.transpose()?;
Ok(json!({
"id": row.attempt_id,
"fence": row.fence,
"worker_incarnation": row.worker_incarnation,
"state": state,
"progress": progress,
"result": result,
"failure": row
.failure
.as_ref()
.map(|value| failure_projection(&value.0))
.transpose()?,
"created_at": format_time(row.authorized_at)?,
"updated_at": format_time(row.updated_at)?
}))
}
pub(super) fn result_record_projection(value: &Value) -> Result<Value, PublicApiError> {
let record: CommittedResultRecord =
serde_json::from_value(value.clone()).map_err(|_| PublicApiError::internal())?;
Ok(json!({
"kind": "manifest",
"manifest_id": record.manifest_id,
"schema_version": record.schema_version,
"manifest": object_identity_projection(
&record.manifest.store,
&record.manifest.bucket,
&record.manifest.key,
record.manifest.version_id.as_deref(),
&record.manifest.sha256,
record.manifest.size
),
"member_count": record.member_count,
"total_bytes": record.total_bytes
}))
}
pub(super) fn deletion_projection(operation: &DeletionOperation) -> Result<Value, PublicApiError> {
let state = match operation.state {
DeletionState::Queued => "queued",
DeletionState::Running => "running",
DeletionState::RetryWait => "retry_wait",
DeletionState::Blocked => "blocked",
DeletionState::Completed => "completed",
};
let problem = operation.problem.as_ref().map(|problem| {
let class = match problem.class {
DeletionProblemClass::Authorization => "authorization",
DeletionProblemClass::TargetUnavailable => "target_unavailable",
DeletionProblemClass::Configuration => "configuration",
DeletionProblemClass::Internal => "internal",
};
json!({ "class": class, "message": problem.message })
});
Ok(json!({
"id": operation.id,
"job_id": operation.job_id,
"state": state,
"problem": problem,
"created_at": format_time(operation.created_at)?,
"updated_at": format_time(operation.updated_at)?
}))
}

View file

@ -0,0 +1,80 @@
use super::*;
pub(super) fn public_error_body(code: ApiErrorCodeDto, message: &'static str) -> Value {
json!({
"code": code,
"message": message,
"request_id": request_id()
})
}
pub(super) fn page_limit(limit: Option<u32>) -> Result<u32, PublicApiError> {
let limit = limit.unwrap_or(DEFAULT_LIMIT);
if !(1..=MAX_LIMIT).contains(&limit) {
return Err(PublicApiError::invalid_request());
}
Ok(limit)
}
pub(super) fn parse_id(value: impl Into<String>) -> Result<OpaqueId, PublicApiError> {
OpaqueId::parse(value).map_err(|_| PublicApiError::invalid_request())
}
pub(super) fn idempotency_key(request: &HttpRequest) -> Result<&str, PublicApiError> {
let value = request
.headers()
.get("Idempotency-Key")
.and_then(|value| value.to_str().ok())
.ok_or_else(PublicApiError::invalid_request)?;
if !(16..=128).contains(&value.len())
|| !value
.bytes()
.all(|byte| byte.is_ascii_alphanumeric() || b"._~-".contains(&byte))
{
return Err(PublicApiError::invalid_request());
}
Ok(value)
}
pub(super) fn with_idempotency_expiry(mut response: HttpResponse) -> HttpResponse {
let expiry = OffsetDateTime::now_utc() + time::Duration::hours(24);
if let Ok(value) = expiry.format(&Rfc3339)
&& let Ok(value) = header::HeaderValue::from_str(&value)
{
response.headers_mut().insert(
header::HeaderName::from_static("idempotency-expires-at"),
value,
);
}
response
}
pub(super) fn format_time(value: OffsetDateTime) -> Result<String, PublicApiError> {
value
.format(&Rfc3339)
.map_err(|_| PublicApiError::internal())
}
pub(super) fn datetime(value: OffsetDateTime) -> Result<DateTimeDto, PublicApiError> {
DateTimeDto::parse(format_time(value)?).map_err(|_| PublicApiError::internal())
}
pub(super) fn request_id() -> OpaqueId {
let mut bytes = [0_u8; 16];
getrandom::fill(&mut bytes).expect("operating system randomness is required");
let mut value = String::from("r");
for byte in bytes {
use std::fmt::Write as _;
write!(value, "{byte:02x}").expect("writing to String cannot fail");
}
OpaqueId::parse(value).expect("random request id is valid")
}
pub(super) fn validate_request(
schema: &'static str,
request: &impl Serialize,
) -> Result<(), PublicApiError> {
remora_contracts::request_is_valid(schema, request)
.then_some(())
.ok_or_else(PublicApiError::invalid_request)
}

View file

@ -0,0 +1,122 @@
use super::*;
pub(super) async fn get_job_result(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let job_id = parse_id(path.into_inner())?;
let row = load_visible_job_row(&state, &principal, &job_id).await?;
let result = committed_result(&row)?.ok_or_else(PublicApiError::precondition_failed)?;
state.results.resolve_committed(&result).await?;
Ok(HttpResponse::Ok().json(result_projection(&result)))
}
pub(super) async fn list_result_members(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<String>,
query: web::Query<PageQuery>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let job_id = parse_id(path.into_inner())?;
let row = load_visible_job_row(&state, &principal, &job_id).await?;
let committed = committed_result(&row)?.ok_or_else(PublicApiError::precondition_failed)?;
let filter = CursorCodec::filter_fingerprint(&json!({
"resource": "result_members",
"job_id": job_id,
"authorization": authorization_snapshot(&principal)
}))?;
let cursor = query
.cursor
.as_deref()
.map(|token| state.cursors.bind(token, principal.id.as_str(), filter))
.transpose()
.map_err(PublicApiError::from_cursor)?;
let after = cursor
.as_ref()
.and_then(|value| value.after.as_ref())
.map(|value| value.id.as_str());
let page_size = PageSize::new(page_limit(query.limit)?)?;
let page = state
.results
.page_members(&committed, after, page_size)
.await?;
let next_cursor = if let Some(after) = page.next_after.as_deref() {
let next = KeysetPosition {
created_at: row.created_at,
id: after.to_owned(),
};
let cursor = match cursor.as_ref() {
Some(current) => next_page_cursor(current, next),
None => first_page_cursor(
principal.id.as_str(),
filter,
KeysetPosition {
created_at: row.created_at,
id: String::new(),
},
next,
),
};
Some(state.cursors.encode(&cursor)?)
} else {
None
};
let items = page
.items
.iter()
.map(|member| {
json!({
"id": member.id(),
"path": member.path(),
"size": member.size(),
"object": object_identity_projection(
member.object().store(),
member.object().bucket(),
member.object().key(),
member.object().version_id(),
member.content_sha256().as_str(),
member.size()
)
})
})
.collect::<Vec<_>>();
Ok(HttpResponse::Ok().json(json!({
"items": items,
"next_cursor": next_cursor
})))
}
pub(super) async fn access_result_member(
state: web::Data<PublicApiState>,
http: HttpRequest,
path: web::Path<(String, String)>,
) -> Result<HttpResponse, PublicApiError> {
let principal = state.principal(&http).await?;
let (job_id, member_id) = path.into_inner();
let job_id = parse_id(job_id)?;
let member_id = parse_id(member_id)?;
let row = load_visible_job_row(&state, &principal, &job_id).await?;
let committed = committed_result(&row)?.ok_or_else(PublicApiError::precondition_failed)?;
let member = state.results.lookup_member(&committed, &member_id).await?;
let target = state
.targets
.resolve(member.object().store(), member.object().bucket())
.ok_or_else(PublicApiError::unavailable)?;
let issued_at = OffsetDateTime::now_utc();
let access = target
.presign_read(&member.exact_identity(), state.member_access_ttl)
.await
.map_err(|_| PublicApiError::unavailable())?;
let duration =
time::Duration::try_from(access.expires_in).map_err(|_| PublicApiError::internal())?;
Ok(HttpResponse::Ok()
.insert_header((header::CACHE_CONTROL, "no-store"))
.insert_header((header::PRAGMA, "no-cache"))
.json(ResultMemberAccessDto {
url: access.url,
expires_at: datetime(issued_at + duration)?,
}))
}

View file

@ -1,41 +1,45 @@
use actix_web::web; use actix_web::web;
use super::{ use super::{credentials, deletions, events as event_handlers, jobs, results};
access_result_member, cancel_job, create_source_credential, events as event_handlers,
get_deletion_operation, get_job, get_job_result, get_source_credential, hard_delete_job,
list_job_attempts, list_jobs, list_result_members, list_source_credentials, restore_job,
revoke_job, revoke_source_credential, rotate_source_credential, soft_delete_job, submit_job,
};
pub(super) fn jobs(config: &mut web::ServiceConfig) { pub(super) fn jobs(config: &mut web::ServiceConfig) {
config config
.route("/jobs", web::post().to(submit_job)) .route("/jobs", web::post().to(jobs::submit_job))
.route("/jobs", web::get().to(list_jobs)) .route("/jobs", web::get().to(jobs::list_jobs))
.route("/jobs/{job_id}", web::get().to(get_job)) .route("/jobs/{job_id}", web::get().to(jobs::get_job))
.route("/jobs/{job_id}/cancel", web::post().to(cancel_job)) .route("/jobs/{job_id}/cancel", web::post().to(jobs::cancel_job))
.route( .route(
"/jobs/{job_id}/soft-delete", "/jobs/{job_id}/soft-delete",
web::post().to(soft_delete_job), web::post().to(deletions::soft_delete_job),
)
.route(
"/jobs/{job_id}/restore",
web::post().to(deletions::restore_job),
) )
.route("/jobs/{job_id}/restore", web::post().to(restore_job))
.route( .route(
"/jobs/{job_id}/hard-delete", "/jobs/{job_id}/hard-delete",
web::post().to(hard_delete_job), web::post().to(deletions::hard_delete_job),
) )
.route("/jobs/{job_id}/revoke", web::post().to(revoke_job)) .route("/jobs/{job_id}/revoke", web::post().to(jobs::revoke_job))
.route("/jobs/{job_id}/attempts", web::get().to(list_job_attempts)); .route(
"/jobs/{job_id}/attempts",
web::get().to(jobs::list_job_attempts),
);
} }
pub(super) fn results(config: &mut web::ServiceConfig) { pub(super) fn results(config: &mut web::ServiceConfig) {
config config
.route("/jobs/{job_id}/result", web::get().to(get_job_result)) .route(
"/jobs/{job_id}/result",
web::get().to(results::get_job_result),
)
.route( .route(
"/jobs/{job_id}/result/members", "/jobs/{job_id}/result/members",
web::get().to(list_result_members), web::get().to(results::list_result_members),
) )
.route( .route(
"/jobs/{job_id}/result/members/{member_id}/content", "/jobs/{job_id}/result/members/{member_id}/content",
web::get().to(access_result_member), web::get().to(results::access_result_member),
); );
} }
@ -43,30 +47,30 @@ pub(super) fn credentials(config: &mut web::ServiceConfig) {
config config
.route( .route(
"/source-credentials", "/source-credentials",
web::post().to(create_source_credential), web::post().to(credentials::create_source_credential),
) )
.route( .route(
"/source-credentials", "/source-credentials",
web::get().to(list_source_credentials), web::get().to(credentials::list_source_credentials),
) )
.route( .route(
"/source-credentials/{credential_id}", "/source-credentials/{credential_id}",
web::get().to(get_source_credential), web::get().to(credentials::get_source_credential),
) )
.route( .route(
"/source-credentials/{credential_id}/versions", "/source-credentials/{credential_id}/versions",
web::post().to(rotate_source_credential), web::post().to(credentials::rotate_source_credential),
) )
.route( .route(
"/source-credentials/{credential_id}/revoke", "/source-credentials/{credential_id}/revoke",
web::post().to(revoke_source_credential), web::post().to(credentials::revoke_source_credential),
); );
} }
pub(super) fn deletions(config: &mut web::ServiceConfig) { pub(super) fn deletions(config: &mut web::ServiceConfig) {
config.route( config.route(
"/deletions/{operation_id}", "/deletions/{operation_id}",
web::get().to(get_deletion_operation), web::get().to(deletions::get_deletion_operation),
); );
} }

View file

@ -30,7 +30,10 @@ use remora_target::{ContentSha256, ObjectIdentity, StoredObject, attempt_staging
use sha2::{Digest, Sha256}; use sha2::{Digest, Sha256};
use time::{OffsetDateTime, format_description::well_known::Rfc3339}; use time::{OffsetDateTime, format_description::well_known::Rfc3339};
mod authority;
mod execution;
mod routes; mod routes;
mod staging;
#[derive(Clone, Debug, Eq, PartialEq)] #[derive(Clone, Debug, Eq, PartialEq)]
pub struct AuthenticatedWorker { pub struct AuthenticatedWorker {
@ -104,478 +107,6 @@ pub fn configure_worker_api(config: &mut web::ServiceConfig) {
); );
} }
async fn worker_ready(_: AuthenticatedWorker) -> HttpResponse {
HttpResponse::Ok().finish()
}
async fn register_staging_object(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptStagingObjectRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptStagingObjectRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let accepted = state
.attempts
.register_staging_object_operation(
&authority,
&request.operation_id,
StagingObject {
target_store: request.object.store.clone(),
bucket: request.object.bucket.clone(),
key: request.object.key.clone(),
version_id: request.object.version_id.clone(),
},
request.completed_upload_id.clone(),
)
.await
.map_err(WorkerApiError::from_attempt)?;
if accepted {
Ok(HttpResponse::NoContent().finish())
} else {
Err(WorkerApiError::precondition_failed())
}
}
async fn register_staging_write_intent(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptStagingWriteIntentRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptStagingWriteIntentRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let accepted = state
.attempts
.register_staging_write_intent_operation(
&authority,
&request.operation_id,
StagingWriteIntent {
target_store: request.intent.store.clone(),
bucket: request.intent.bucket.clone(),
key: request.intent.key.clone(),
},
)
.await
.map_err(WorkerApiError::from_attempt)?;
if accepted {
Ok(HttpResponse::NoContent().finish())
} else {
Err(WorkerApiError::precondition_failed())
}
}
async fn register_staging_multipart(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptStagingMultipartRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptStagingMultipartRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let accepted = state
.attempts
.register_staging_multipart_operation(
&authority,
&request.operation_id,
StagingMultipart {
target_store: request.multipart.store.clone(),
bucket: request.multipart.bucket.clone(),
key: request.multipart.key.clone(),
upload_id: request.multipart.upload_id.clone(),
},
)
.await
.map_err(WorkerApiError::from_attempt)?;
if accepted {
Ok(HttpResponse::NoContent().finish())
} else {
Err(WorkerApiError::precondition_failed())
}
}
async fn stop(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptStopRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptStopRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
match state
.attempts
.cooperative_stop_operation(&authority, &request.operation_id)
.await
.map_err(WorkerApiError::from_attempt)?
{
remora_attempt::CooperativeStopOutcome::Stopped
| remora_attempt::CooperativeStopOutcome::StopNotRequired => {
Ok(HttpResponse::NoContent().finish())
}
remora_attempt::CooperativeStopOutcome::AuthorityMismatch => {
Err(WorkerApiError::precondition_failed())
}
}
}
async fn claim(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<ClaimRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("ClaimRequest", &*request)?;
let job_id = parse_id(path.into_inner())?;
let outcome = state
.attempts
.claim(&job_id, &request.worker_incarnation, &request.operation_id)
.await
.map_err(WorkerApiError::from_attempt)?;
let ClaimOutcome::Claimed { authority, .. } = outcome else {
return Err(WorkerApiError::precondition_failed());
};
let execution = load_execution_envelope(&state.attempts, &authority).await?;
Ok(HttpResponse::Ok().json(ClaimResponseDto {
authority: authority_dto(&authority)?,
execution,
}))
}
async fn renew(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptRenewRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptRenewRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
match state
.attempts
.renew_operation(&authority, &request.operation_id)
.await
.map_err(WorkerApiError::from_attempt)?
{
RenewalMutationOutcome::Renewed { lease, .. } => {
Ok(HttpResponse::Ok().json(lease_dto(lease)?))
}
RenewalMutationOutcome::AuthorityLost { .. } => Err(WorkerApiError::precondition_failed()),
}
}
async fn activate(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptActivateRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptActivateRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let lease = match state
.attempts
.activate_operation(&authority, &request.operation_id)
.await
.map_err(WorkerApiError::from_attempt)?
{
ActivationMutationOutcome::Activated { lease, .. } => lease,
ActivationMutationOutcome::PreconditionFailed => {
return Err(WorkerApiError::precondition_failed());
}
};
let execution = load_execution_envelope(&state.attempts, &authority).await?;
Ok(HttpResponse::Ok().json(ActivationResponseDto {
authority: authority_dto(&authority)?,
execution,
lease: lease_dto(lease)?,
}))
}
async fn pin_resolved_source(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptResolvedSourceRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptResolvedSourceRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
accepted(
state
.attempts
.pin_resolved_source(
&authority,
&request.operation_id,
descriptor_from_resolved(&request.resolved_source_ref)?,
)
.await
.map_err(WorkerApiError::from_attempt)?,
)
}
async fn persist_checkpoint(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptCheckpointRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptCheckpointRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
accepted(
state
.attempts
.persist_checkpoint(
&authority,
&request.operation_id,
CheckpointRecord {
descriptor: descriptor_from_checkpoint(&request.checkpoint_ref)?,
resolved_source_id: request.resolved_source_id.clone(),
validated_bytes: request.validated_bytes.get(),
validated_members: request.validated_members.get(),
staging_multipart: request.staging_multipart.as_ref().map(|identity| {
CheckpointMultipart {
target_store: identity.store.clone(),
bucket: identity.bucket.clone(),
key: identity.key.clone(),
upload_id: identity.upload_id.clone(),
}
}),
},
)
.await
.map_err(WorkerApiError::from_attempt)?,
)
}
async fn report_progress(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptProgressRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptProgressRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
accepted(
state
.attempts
.report_progress(
&authority,
&request.operation_id,
ProgressRecord {
completed_bytes: request.progress.completed_bytes.get(),
total_bytes: request.progress.total_bytes.map(SafeIntegerDto::get),
completed_members: request.progress.completed_members.get(),
total_members: request.progress.total_members.map(SafeIntegerDto::get),
},
)
.await
.map_err(WorkerApiError::from_attempt)?,
)
}
async fn report_outcome(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptOutcomeRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptOutcomeRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let outcome = match &request.outcome {
AttemptOutcomeDto::Success(success) => {
let result = &success.result;
state
.attempts
.publish_success_operation(
&authority,
&request.operation_id,
PublicationInput {
attempt_id: authority.attempt_id.clone(),
manifest_id: result.manifest_id.clone(),
schema_version: u32::try_from(result.schema_version.get())
.map_err(|_| WorkerApiError::invalid_request())?,
manifest: StoredObject {
identity: ObjectIdentity {
target_store_id: result.manifest.store.clone(),
bucket: result.manifest.bucket.clone(),
key: result.manifest.key.clone(),
version_id: result.manifest.version_id.clone(),
},
content_sha256: ContentSha256::parse(&result.manifest.checksum)
.map_err(|_| WorkerApiError::invalid_request())?,
size: result.manifest.size.get(),
},
member_count: result.member_count.get(),
total_bytes: result.total_bytes.get(),
},
)
.await
}
AttemptOutcomeDto::Failure(failure) => {
state
.attempts
.publish_failure_operation(&authority, &request.operation_id, failure.failure.class)
.await
}
}
.map_err(WorkerApiError::from_attempt)?;
let phase = match outcome {
OutcomePublication::Accepted { phase, .. } => phase,
OutcomePublication::Rejected { reason, .. } => {
return Err(WorkerApiError::from_rejection(reason));
}
};
Ok(HttpResponse::Ok().json(JobPhaseProjectionDto { phase }))
}
fn accepted(outcome: AttemptMutationOutcome) -> Result<HttpResponse, WorkerApiError> {
match outcome {
AttemptMutationOutcome::Accepted(_) => Ok(HttpResponse::NoContent().finish()),
AttemptMutationOutcome::Rejected { reason, .. } => {
Err(WorkerApiError::from_rejection(reason))
}
}
}
async fn request_authority(
state: &WorkerApiState,
attempt_id: String,
fence: u64,
worker: &OpaqueId,
) -> Result<AttemptAuthority, WorkerApiError> {
let attempt_id = parse_id(attempt_id)?;
state
.attempts
.historical_authority(&attempt_id)
.await
.map_err(WorkerApiError::from_attempt)?
.filter(|authority| authority.fence == fence && authority.worker_incarnation == *worker)
.ok_or_else(WorkerApiError::precondition_failed)
}
async fn load_execution_envelope(
attempts: &AttemptService,
authority: &AttemptAuthority,
) -> Result<ExecutionEnvelopeDto, WorkerApiError> {
let row = attempts
.execution_envelope(&authority.attempt_id)
.await
.map_err(WorkerApiError::from_attempt)?
.ok_or_else(WorkerApiError::precondition_failed)?;
if row.job_id != authority.job_id
|| row.attempt_id != authority.attempt_id
|| row.fence != authority.fence
|| row.worker_incarnation != authority.worker_incarnation
{
return Err(WorkerApiError::precondition_failed());
}
let source: SourceProjectionDto =
serde_json::from_value(row.source).map_err(|_| WorkerApiError::unavailable())?;
let target_prefix = row.target_prefix.join("/");
let staging_prefix = attempt_staging_prefix(&target_prefix, &row.staging_origin_attempt_id);
let resolved_source_ref = row.resolved_source.map(|descriptor| {
NullableDto(Some(descriptor_ref(
descriptor,
DurableDescriptorKindDto::ResolvedSource,
)))
});
let checkpoint_ref = row.checkpoint.and_then(|checkpoint| match checkpoint {
CheckpointProjection::Reusable { checkpoint } => Some(NullableDto(Some(descriptor_ref(
checkpoint.descriptor,
DurableDescriptorKindDto::Checkpoint,
)))),
CheckpointProjection::Complete { .. } => None,
});
Ok(ExecutionEnvelopeDto {
id: row.id,
job_id: authority.job_id.clone(),
attempt_id: authority.attempt_id.clone(),
fence: positive(authority.fence)?,
worker_incarnation: authority.worker_incarnation.clone(),
source,
sensitive_source_ref: row.sensitive_source_ref,
credential_version: NullableDto(match (row.credential_id, row.credential_version_id) {
(Some(credential_id), Some(version_id)) => Some(CredentialGrantVersionDto {
credential_id,
version_id,
}),
(None, None) => None,
_ => return Err(WorkerApiError::unavailable()),
}),
resolved_source_ref: resolved_source_ref.unwrap_or(NullableDto(None)),
target: TargetDto {
store: row.target_store.clone(),
bucket: row.target_bucket.clone(),
prefix: target_prefix,
},
staging_target: StagingTargetDto {
store: row.target_store,
bucket: row.target_bucket,
prefix: staging_prefix,
},
execution_grant_ref: row.execution_grant_id,
checkpoint_ref: checkpoint_ref.unwrap_or(NullableDto(None)),
})
}
fn descriptor_ref( fn descriptor_ref(
descriptor: DurableDescriptor, descriptor: DurableDescriptor,
kind: DurableDescriptorKindDto, kind: DurableDescriptorKindDto,

View file

@ -0,0 +1,199 @@
use super::*;
pub(super) async fn worker_ready(_: AuthenticatedWorker) -> HttpResponse {
HttpResponse::Ok().finish()
}
pub(super) async fn stop(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptStopRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptStopRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
match state
.attempts
.cooperative_stop_operation(&authority, &request.operation_id)
.await
.map_err(WorkerApiError::from_attempt)?
{
remora_attempt::CooperativeStopOutcome::Stopped
| remora_attempt::CooperativeStopOutcome::StopNotRequired => {
Ok(HttpResponse::NoContent().finish())
}
remora_attempt::CooperativeStopOutcome::AuthorityMismatch => {
Err(WorkerApiError::precondition_failed())
}
}
}
pub(super) async fn claim(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<ClaimRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("ClaimRequest", &*request)?;
let job_id = parse_id(path.into_inner())?;
let outcome = state
.attempts
.claim(&job_id, &request.worker_incarnation, &request.operation_id)
.await
.map_err(WorkerApiError::from_attempt)?;
let ClaimOutcome::Claimed { authority, .. } = outcome else {
return Err(WorkerApiError::precondition_failed());
};
let execution = load_execution_envelope(&state.attempts, &authority).await?;
Ok(HttpResponse::Ok().json(ClaimResponseDto {
authority: authority_dto(&authority)?,
execution,
}))
}
pub(super) async fn renew(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptRenewRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptRenewRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
match state
.attempts
.renew_operation(&authority, &request.operation_id)
.await
.map_err(WorkerApiError::from_attempt)?
{
RenewalMutationOutcome::Renewed { lease, .. } => {
Ok(HttpResponse::Ok().json(lease_dto(lease)?))
}
RenewalMutationOutcome::AuthorityLost { .. } => Err(WorkerApiError::precondition_failed()),
}
}
pub(super) async fn activate(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptActivateRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptActivateRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let lease = match state
.attempts
.activate_operation(&authority, &request.operation_id)
.await
.map_err(WorkerApiError::from_attempt)?
{
ActivationMutationOutcome::Activated { lease, .. } => lease,
ActivationMutationOutcome::PreconditionFailed => {
return Err(WorkerApiError::precondition_failed());
}
};
let execution = load_execution_envelope(&state.attempts, &authority).await?;
Ok(HttpResponse::Ok().json(ActivationResponseDto {
authority: authority_dto(&authority)?,
execution,
lease: lease_dto(lease)?,
}))
}
pub(super) async fn request_authority(
state: &WorkerApiState,
attempt_id: String,
fence: u64,
worker: &OpaqueId,
) -> Result<AttemptAuthority, WorkerApiError> {
let attempt_id = parse_id(attempt_id)?;
state
.attempts
.historical_authority(&attempt_id)
.await
.map_err(WorkerApiError::from_attempt)?
.filter(|authority| authority.fence == fence && authority.worker_incarnation == *worker)
.ok_or_else(WorkerApiError::precondition_failed)
}
pub(super) async fn load_execution_envelope(
attempts: &AttemptService,
authority: &AttemptAuthority,
) -> Result<ExecutionEnvelopeDto, WorkerApiError> {
let row = attempts
.execution_envelope(&authority.attempt_id)
.await
.map_err(WorkerApiError::from_attempt)?
.ok_or_else(WorkerApiError::precondition_failed)?;
if row.job_id != authority.job_id
|| row.attempt_id != authority.attempt_id
|| row.fence != authority.fence
|| row.worker_incarnation != authority.worker_incarnation
{
return Err(WorkerApiError::precondition_failed());
}
let source: SourceProjectionDto =
serde_json::from_value(row.source).map_err(|_| WorkerApiError::unavailable())?;
let target_prefix = row.target_prefix.join("/");
let staging_prefix = attempt_staging_prefix(&target_prefix, &row.staging_origin_attempt_id);
let resolved_source_ref = row.resolved_source.map(|descriptor| {
NullableDto(Some(descriptor_ref(
descriptor,
DurableDescriptorKindDto::ResolvedSource,
)))
});
let checkpoint_ref = row.checkpoint.and_then(|checkpoint| match checkpoint {
CheckpointProjection::Reusable { checkpoint } => Some(NullableDto(Some(descriptor_ref(
checkpoint.descriptor,
DurableDescriptorKindDto::Checkpoint,
)))),
CheckpointProjection::Complete { .. } => None,
});
Ok(ExecutionEnvelopeDto {
id: row.id,
job_id: authority.job_id.clone(),
attempt_id: authority.attempt_id.clone(),
fence: positive(authority.fence)?,
worker_incarnation: authority.worker_incarnation.clone(),
source,
sensitive_source_ref: row.sensitive_source_ref,
credential_version: NullableDto(match (row.credential_id, row.credential_version_id) {
(Some(credential_id), Some(version_id)) => Some(CredentialGrantVersionDto {
credential_id,
version_id,
}),
(None, None) => None,
_ => return Err(WorkerApiError::unavailable()),
}),
resolved_source_ref: resolved_source_ref.unwrap_or(NullableDto(None)),
target: TargetDto {
store: row.target_store.clone(),
bucket: row.target_bucket.clone(),
prefix: target_prefix,
},
staging_target: StagingTargetDto {
store: row.target_store,
bucket: row.target_bucket,
prefix: staging_prefix,
},
execution_grant_ref: row.execution_grant_id,
checkpoint_ref: checkpoint_ref.unwrap_or(NullableDto(None)),
})
}

View file

@ -0,0 +1,171 @@
use super::authority::request_authority;
use super::*;
pub(super) async fn pin_resolved_source(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptResolvedSourceRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptResolvedSourceRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
accepted(
state
.attempts
.pin_resolved_source(
&authority,
&request.operation_id,
descriptor_from_resolved(&request.resolved_source_ref)?,
)
.await
.map_err(WorkerApiError::from_attempt)?,
)
}
pub(super) async fn persist_checkpoint(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptCheckpointRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptCheckpointRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
accepted(
state
.attempts
.persist_checkpoint(
&authority,
&request.operation_id,
CheckpointRecord {
descriptor: descriptor_from_checkpoint(&request.checkpoint_ref)?,
resolved_source_id: request.resolved_source_id.clone(),
validated_bytes: request.validated_bytes.get(),
validated_members: request.validated_members.get(),
staging_multipart: request.staging_multipart.as_ref().map(|identity| {
CheckpointMultipart {
target_store: identity.store.clone(),
bucket: identity.bucket.clone(),
key: identity.key.clone(),
upload_id: identity.upload_id.clone(),
}
}),
},
)
.await
.map_err(WorkerApiError::from_attempt)?,
)
}
pub(super) async fn report_progress(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptProgressRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptProgressRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
accepted(
state
.attempts
.report_progress(
&authority,
&request.operation_id,
ProgressRecord {
completed_bytes: request.progress.completed_bytes.get(),
total_bytes: request.progress.total_bytes.map(SafeIntegerDto::get),
completed_members: request.progress.completed_members.get(),
total_members: request.progress.total_members.map(SafeIntegerDto::get),
},
)
.await
.map_err(WorkerApiError::from_attempt)?,
)
}
pub(super) async fn report_outcome(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptOutcomeRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptOutcomeRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let outcome = match &request.outcome {
AttemptOutcomeDto::Success(success) => {
let result = &success.result;
state
.attempts
.publish_success_operation(
&authority,
&request.operation_id,
PublicationInput {
attempt_id: authority.attempt_id.clone(),
manifest_id: result.manifest_id.clone(),
schema_version: u32::try_from(result.schema_version.get())
.map_err(|_| WorkerApiError::invalid_request())?,
manifest: StoredObject {
identity: ObjectIdentity {
target_store_id: result.manifest.store.clone(),
bucket: result.manifest.bucket.clone(),
key: result.manifest.key.clone(),
version_id: result.manifest.version_id.clone(),
},
content_sha256: ContentSha256::parse(&result.manifest.checksum)
.map_err(|_| WorkerApiError::invalid_request())?,
size: result.manifest.size.get(),
},
member_count: result.member_count.get(),
total_bytes: result.total_bytes.get(),
},
)
.await
}
AttemptOutcomeDto::Failure(failure) => {
state
.attempts
.publish_failure_operation(&authority, &request.operation_id, failure.failure.class)
.await
}
}
.map_err(WorkerApiError::from_attempt)?;
let phase = match outcome {
OutcomePublication::Accepted { phase, .. } => phase,
OutcomePublication::Rejected { reason, .. } => {
return Err(WorkerApiError::from_rejection(reason));
}
};
Ok(HttpResponse::Ok().json(JobPhaseProjectionDto { phase }))
}
pub(super) fn accepted(outcome: AttemptMutationOutcome) -> Result<HttpResponse, WorkerApiError> {
match outcome {
AttemptMutationOutcome::Accepted(_) => Ok(HttpResponse::NoContent().finish()),
AttemptMutationOutcome::Rejected { reason, .. } => {
Err(WorkerApiError::from_rejection(reason))
}
}
}

View file

@ -1,33 +1,44 @@
use actix_web::web; use actix_web::web;
use super::{ use super::{authority as authority_handlers, execution, staging};
activate, claim, persist_checkpoint, pin_resolved_source, register_staging_multipart,
register_staging_object, register_staging_write_intent, renew, report_outcome, report_progress,
stop, worker_ready,
};
pub(super) fn authority(config: &mut web::ServiceConfig) { pub(super) fn authority(config: &mut web::ServiceConfig) {
config config
.route("/health/ready", web::get().to(worker_ready)) .route(
.route("/jobs/{job_id}/claim", web::post().to(claim)) "/health/ready",
.route("/attempts/{attempt_id}/renew", web::post().to(renew)) web::get().to(authority_handlers::worker_ready),
.route("/attempts/{attempt_id}/activate", web::post().to(activate)) )
.route("/attempts/{attempt_id}/stop", web::post().to(stop)); .route(
"/jobs/{job_id}/claim",
web::post().to(authority_handlers::claim),
)
.route(
"/attempts/{attempt_id}/renew",
web::post().to(authority_handlers::renew),
)
.route(
"/attempts/{attempt_id}/activate",
web::post().to(authority_handlers::activate),
)
.route(
"/attempts/{attempt_id}/stop",
web::post().to(authority_handlers::stop),
);
} }
pub(super) fn staging(config: &mut web::ServiceConfig) { pub(super) fn staging(config: &mut web::ServiceConfig) {
config config
.route( .route(
"/attempts/{attempt_id}/staging/write-intent", "/attempts/{attempt_id}/staging/write-intent",
web::post().to(register_staging_write_intent), web::post().to(staging::register_staging_write_intent),
) )
.route( .route(
"/attempts/{attempt_id}/staging/object", "/attempts/{attempt_id}/staging/object",
web::post().to(register_staging_object), web::post().to(staging::register_staging_object),
) )
.route( .route(
"/attempts/{attempt_id}/staging/multipart", "/attempts/{attempt_id}/staging/multipart",
web::post().to(register_staging_multipart), web::post().to(staging::register_staging_multipart),
); );
} }
@ -35,18 +46,18 @@ pub(super) fn execution(config: &mut web::ServiceConfig) {
config config
.route( .route(
"/attempts/{attempt_id}/resolved-source", "/attempts/{attempt_id}/resolved-source",
web::post().to(pin_resolved_source), web::post().to(execution::pin_resolved_source),
) )
.route( .route(
"/attempts/{attempt_id}/checkpoint", "/attempts/{attempt_id}/checkpoint",
web::post().to(persist_checkpoint), web::post().to(execution::persist_checkpoint),
) )
.route( .route(
"/attempts/{attempt_id}/progress", "/attempts/{attempt_id}/progress",
web::post().to(report_progress), web::post().to(execution::report_progress),
) )
.route( .route(
"/attempts/{attempt_id}/outcome", "/attempts/{attempt_id}/outcome",
web::post().to(report_outcome), web::post().to(execution::report_outcome),
); );
} }

View file

@ -0,0 +1,107 @@
use super::authority::request_authority;
use super::*;
pub(super) async fn register_staging_object(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptStagingObjectRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptStagingObjectRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let accepted = state
.attempts
.register_staging_object_operation(
&authority,
&request.operation_id,
StagingObject {
target_store: request.object.store.clone(),
bucket: request.object.bucket.clone(),
key: request.object.key.clone(),
version_id: request.object.version_id.clone(),
},
request.completed_upload_id.clone(),
)
.await
.map_err(WorkerApiError::from_attempt)?;
if accepted {
Ok(HttpResponse::NoContent().finish())
} else {
Err(WorkerApiError::precondition_failed())
}
}
pub(super) async fn register_staging_write_intent(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptStagingWriteIntentRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptStagingWriteIntentRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let accepted = state
.attempts
.register_staging_write_intent_operation(
&authority,
&request.operation_id,
StagingWriteIntent {
target_store: request.intent.store.clone(),
bucket: request.intent.bucket.clone(),
key: request.intent.key.clone(),
},
)
.await
.map_err(WorkerApiError::from_attempt)?;
if accepted {
Ok(HttpResponse::NoContent().finish())
} else {
Err(WorkerApiError::precondition_failed())
}
}
pub(super) async fn register_staging_multipart(
_: AuthenticatedWorker,
state: web::Data<WorkerApiState>,
path: web::Path<String>,
request: web::Json<AttemptStagingMultipartRequestDto>,
) -> Result<HttpResponse, WorkerApiError> {
validate_request("AttemptStagingMultipartRequest", &*request)?;
let authority = request_authority(
&state,
path.into_inner(),
request.fence.get(),
&request.worker_incarnation,
)
.await?;
let accepted = state
.attempts
.register_staging_multipart_operation(
&authority,
&request.operation_id,
StagingMultipart {
target_store: request.multipart.store.clone(),
bucket: request.multipart.bucket.clone(),
key: request.multipart.key.clone(),
upload_id: request.multipart.upload_id.clone(),
},
)
.await
.map_err(WorkerApiError::from_attempt)?;
if accepted {
Ok(HttpResponse::NoContent().finish())
} else {
Err(WorkerApiError::precondition_failed())
}
}

Some files were not shown because too many files have changed in this diff Show more