ms2k/remora
SHA256
Watch
0
Code Releases Activity

Add fenced attempt mutation and outcome publication

Browse source
This commit is contained in:
ms2k 2026-07-17 07:40:36 +09:00
commit 80c631f780
1 changed files with 1868 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

45
crates/remora-attempt/src/service.rs
View file

@ -10,12 +10,13 @@ use remora_durable::{
};
use serde::Serialize;
use serde_json::json;
use sqlx::PgPool;
use sqlx::{PgPool, types::Json};
use time::OffsetDateTime;
use crate::{
ActivationOutcome, AttemptAuthority, AttemptError, AttemptFaultInjector, AttemptFaultPoint,
AttemptState, ClaimOutcome, ClaimRejection, RecoveryOutcome, RenewalOutcome,
AttemptState, CheckpointProjection, ClaimOutcome, ClaimRejection, RecoveryOutcome,
RenewalOutcome,
};
struct NoFaults;
@ -41,12 +42,12 @@ impl Default for AttemptConfig {
#[derive(Clone)]
pub struct AttemptService {
pool: PgPool,
coordination: CoordinationStore,
pub(crate) pool: PgPool,
pub(crate) coordination: CoordinationStore,
local_permits: LocalPermitGate,
operation_subject: OpaqueId,
pub(crate) operation_subject: OpaqueId,
faults: Arc<dyn AttemptFaultInjector>,
config: AttemptConfig,
pub(crate) config: AttemptConfig,
}
impl AttemptService {
@ -477,14 +478,26 @@ async fn authorize_attempt(
tx: &mut remora_durable::PgTx<'_>,
input: AuthorizeInput<'_>,
) -> Result<ClaimOutcome, DurableError> {
let row = sqlx::query_as::<_, (String, bool, i64, Option<String>)>(
"SELECT phase, dispatch_needed, current_fence, authority_attempt_id
let row = sqlx::query_as::<
_,
(
String,
bool,
i64,
Option<String>,
Option<Json<crate::DurableDescriptor>>,
Option<Json<CheckpointProjection>>,
),
>(
"SELECT phase, dispatch_needed, current_fence, authority_attempt_id,
resolved_source, checkpoint
FROM jobs WHERE job_id = $1 FOR UPDATE",
)
.bind(input.job_id.as_str())
.fetch_optional(tx.as_mut())
.await?;
let Some((phase, dispatch_needed, current_fence, authority)) = row else {
let Some((phase, dispatch_needed, current_fence, authority, resolved_source, checkpoint)) = row
else {
return Ok(claim_rejected(ClaimRejection::JobUnavailable));
};
if phase != "queued" || !dispatch_needed || authority.is_some() {
@ -509,17 +522,29 @@ async fn authorize_attempt(
input.reserved.store(true, Ordering::Release);
input.faults.hit(AttemptFaultPoint::ClaimAfterReserve);
let fence = current_fence + 1;
let initial_progress = match (resolved_source, checkpoint) {
(Some(source), Some(checkpoint))
if checkpoint
.0
.reusable()
.is_some_and(|value| value.resolved_source_id == source.0.id) =>
{
checkpoint.0.reusable().unwrap().validated_bytes
}
_ => 0,
};
let now = OffsetDateTime::now_utc();
sqlx::query(
"INSERT INTO attempts (
attempt_id, job_id, fence, worker_incarnation, state,
initial_progress, authorized_at, executing_at, reaped_at
) VALUES ($1, $2, $3, $4, 'authorized', 0, $5, NULL, NULL)",
) VALUES ($1, $2, $3, $4, 'authorized', $5, $6, NULL, NULL)",
)
.bind(input.attempt_id.as_str())
.bind(input.job_id.as_str())
.bind(i64::try_from(fence).map_err(DurableError::effect)?)
.bind(input.worker.as_str())
.bind(i64::try_from(initial_progress).map_err(DurableError::effect)?)
.bind(now)
.execute(tx.as_mut())
.await?;