Implement crash-safe streaming worker runtime
This commit is contained in:
parent
16cba3c317
commit
cdeef3cc1a
1 changed files with 7448 additions and 194 deletions
|
|
@ -18,7 +18,7 @@ use remora_domain::{
|
|||
use remora_durable::{MutationDisposition, PoolConfig, connect, migrate as durable_migrate};
|
||||
use remora_identity::{DefaultTarget, Principal};
|
||||
use remora_result::{PublicationInput, migrate as result_migrate};
|
||||
use remora_target::{ContentSha256, ObjectIdentity, StoredObject, attempt_manifest_key};
|
||||
use remora_target::{ContentSha256, ObjectIdentity, StoredObject, attempt_manifest_key_in};
|
||||
use sqlx::PgPool;
|
||||
|
||||
struct Harness {
|
||||
|
|
@ -168,7 +168,7 @@ impl Harness {
|
|||
) -> PublicationInput {
|
||||
let manifest_id = unique_id("manifest");
|
||||
let sha256 = format!("sha256:{}", "11".repeat(32));
|
||||
let key = attempt_manifest_key(attempt_id, &manifest_id);
|
||||
let key = attempt_manifest_key_in("users/alice", attempt_id, &manifest_id);
|
||||
let input = PublicationInput {
|
||||
attempt_id: attempt_id.clone(),
|
||||
manifest_id: manifest_id.clone(),
|
||||
|
|
@ -190,9 +190,10 @@ impl Harness {
|
|||
"INSERT INTO result_inventories (
|
||||
attempt_id, state, manifest_schema_version, manifest_id, manifest_store,
|
||||
manifest_bucket, manifest_key, manifest_version_id, manifest_sha256,
|
||||
manifest_size, member_count, total_bytes
|
||||
manifest_size, member_count, total_bytes,
|
||||
staging_origin_attempt_id, target_prefix
|
||||
) VALUES ($1, 'sealed', 1, $2, 'primary', 'remora-results', $3,
|
||||
'version-1', $4, 128, $5, $6)",
|
||||
'version-1', $4, 128, $5, $6, $1, 'users/alice')",
|
||||
)
|
||||
.bind(attempt_id.as_str())
|
||||
.bind(manifest_id.as_str())
|
||||
|
|
@ -283,6 +284,7 @@ async fn resolved_source_checkpoint_and_progress_are_fenced_and_replayable() {
|
|||
resolved_source_id: resolved.id.clone(),
|
||||
validated_bytes: 0,
|
||||
validated_members: 0,
|
||||
staging_multipart: None,
|
||||
},
|
||||
)
|
||||
.await
|
||||
|
|
@ -295,6 +297,7 @@ async fn resolved_source_checkpoint_and_progress_are_fenced_and_replayable() {
|
|||
resolved_source_id: resolved.id.clone(),
|
||||
validated_bytes: 1024,
|
||||
validated_members: 1,
|
||||
staging_multipart: None,
|
||||
};
|
||||
let mut wrong_source = checkpoint.clone();
|
||||
wrong_source.resolved_source_id = unique_id("resolved");
|
||||
|
|
@ -539,6 +542,7 @@ async fn failure_disposition_retries_with_budget_and_becomes_terminal_when_exhau
|
|||
resolved_source_id: retry_source.id.clone(),
|
||||
validated_bytes: 512,
|
||||
validated_members: 1,
|
||||
staging_multipart: None,
|
||||
};
|
||||
assert_rejected(
|
||||
service
|
||||
|
|
@ -555,6 +559,40 @@ async fn failure_disposition_retries_with_budget_and_becomes_terminal_when_exhau
|
|||
.persist_checkpoint(&first, &unique_id("checkpointop"), retry_checkpoint)
|
||||
.await
|
||||
.unwrap();
|
||||
sqlx::query(
|
||||
"INSERT INTO result_inventories (
|
||||
attempt_id, staging_origin_attempt_id, target_prefix
|
||||
) VALUES ($1, $1, 'users/alice')",
|
||||
)
|
||||
.bind(first.attempt_id.as_str())
|
||||
.execute(&harness.pool)
|
||||
.await
|
||||
.unwrap();
|
||||
sqlx::query(
|
||||
"INSERT INTO result_inventory_members (
|
||||
attempt_id, member_id, path, size, object_store, object_bucket,
|
||||
object_key, object_sha256
|
||||
) VALUES ($1, $2, 'retry.bin', 512, 'target', 'bucket', $3, $4)",
|
||||
)
|
||||
.bind(first.attempt_id.as_str())
|
||||
.bind(unique_id("member").as_str())
|
||||
.bind(format!(
|
||||
"users/alice/_remora/a/{}/o/retry/retry.bin",
|
||||
first.attempt_id.as_str()
|
||||
))
|
||||
.bind(format!("sha256:{}", "0".repeat(64)))
|
||||
.execute(&harness.pool)
|
||||
.await
|
||||
.unwrap();
|
||||
sqlx::query(
|
||||
"UPDATE result_inventories
|
||||
SET member_count = 1, total_bytes = 512
|
||||
WHERE attempt_id = $1",
|
||||
)
|
||||
.bind(first.attempt_id.as_str())
|
||||
.execute(&harness.pool)
|
||||
.await
|
||||
.unwrap();
|
||||
assert!(matches!(
|
||||
service
|
||||
.publish_failure(
|
||||
|
|
@ -610,6 +648,20 @@ async fn failure_disposition_retries_with_budget_and_becomes_terminal_when_exhau
|
|||
.await
|
||||
.unwrap();
|
||||
assert_eq!(initial_progress, 512);
|
||||
let staging_origin: String =
|
||||
sqlx::query_scalar("SELECT staging_origin_attempt_id FROM attempts WHERE attempt_id = $1")
|
||||
.bind(second.attempt_id.as_str())
|
||||
.fetch_one(&harness.pool)
|
||||
.await
|
||||
.unwrap();
|
||||
assert_eq!(staging_origin, first.attempt_id.as_str());
|
||||
let inventory_owner: String =
|
||||
sqlx::query_scalar("SELECT attempt_id FROM result_inventories WHERE attempt_id = $1")
|
||||
.bind(second.attempt_id.as_str())
|
||||
.fetch_one(&harness.pool)
|
||||
.await
|
||||
.unwrap();
|
||||
assert_eq!(inventory_owner, second.attempt_id.as_str());
|
||||
service
|
||||
.publish_failure(
|
||||
&second,
|
||||
|
|
@ -760,7 +812,7 @@ async fn publication_from_database(pool: &PgPool, attempt_id: &OpaqueId) -> Publ
|
|||
identity: ObjectIdentity {
|
||||
target_store_id: row.1,
|
||||
bucket: row.2,
|
||||
key: attempt_manifest_key(attempt_id, &manifest_id),
|
||||
key: attempt_manifest_key_in("users/alice", attempt_id, &manifest_id),
|
||||
version_id: row.3,
|
||||
},
|
||||
content_sha256: ContentSha256::parse(row.4).unwrap(),
|
||||
|
|
|
|||
Loading…
Reference in a new issue