i/Aether
1
0
Fork 0
mirror of https://github.com/fawney19/Aether.git synced 2026-09-02 09:20:22 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: add provider-key concurrent limit (#352)

Browse Source 
* feat: add provider-key concurrent limit

* Fix provider key concurrent limit checks

---------

Co-authored-by: fawney19 <elky0401@gmail.com>
This commit is contained in:
Kayphoon
2026-05-03 01:55:23 +08:00
committed by GitHub
parent 11c5884d4f
commit fe27fb17fb
23 changed files with 1827 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
crates/aether-data/src/repository/provider_catalog/memory.rs
View File

@@ -845,6 +845,36 @@ mod tests {
assert_eq!(stored[0].provider_id, "provider-1");
}
#[tokio::test]
async fn provider_api_keys_concurrent_limit_defaults_and_round_trips_in_memory() {
let repository = InMemoryProviderCatalogReadRepository::seed(
vec![sample_provider("test-provider-a")],
vec![],
vec![],
);
let mut key = sample_key("provider-key-a", "test-provider-a");
assert_eq!(key.concurrent_limit, None);
key.concurrent_limit = Some(1);
let created = repository
.create_key(&key)
.await
.expect("key should create");
assert_eq!(created.concurrent_limit, Some(1));
let mut updated = created.clone();
updated.concurrent_limit = None;
repository
.update_key(&updated)
.await
.expect("key should update");
let reloaded = repository
.list_keys_by_ids(&["provider-key-a".to_string()])
.await
.expect("keys should read");
assert_eq!(reloaded[0].concurrent_limit, None);
}
#[tokio::test]
async fn creates_endpoint() {
let repository = InMemoryProviderCatalogReadRepository::seed(

113
crates/aether-data/src/repository/provider_catalog/sql.rs
View File

@@ -155,6 +155,7 @@ SELECT
proxy,
fingerprint,
rpm_limit,
concurrent_limit,
learned_rpm_limit,
concurrent_429_count,
rpm_429_count,
@@ -213,6 +214,7 @@ SELECT
proxy,
fingerprint,
rpm_limit,
concurrent_limit,
learned_rpm_limit,
concurrent_429_count,
rpm_429_count,
@@ -273,6 +275,7 @@ SELECT
NULL::jsonb AS proxy,
NULL::jsonb AS fingerprint,
NULL::integer AS rpm_limit,
NULL::integer AS concurrent_limit,
NULL::integer AS learned_rpm_limit,
NULL::integer AS concurrent_429_count,
NULL::integer AS rpm_429_count,
@@ -611,6 +614,7 @@ SELECT
proxy,
fingerprint,
rpm_limit,
concurrent_limit,
learned_rpm_limit,
concurrent_429_count,
rpm_429_count,
@@ -1199,6 +1203,7 @@ INSERT INTO provider_api_keys (
internal_priority,
global_priority_by_format,
rpm_limit,
concurrent_limit,
learned_rpm_limit,
allowed_models,
capabilities,
@@ -1331,6 +1336,7 @@ INSERT INTO provider_api_keys (
.bind(key.internal_priority)
.bind(&key.global_priority_by_format)
.bind(key.rpm_limit.map(|value| value as i32))
.bind(key.concurrent_limit)
.bind(key.learned_rpm_limit.map(|value| value as i32))
.bind(&key.allowed_models)
.bind(&key.capabilities)
@@ -1726,8 +1732,8 @@ UPDATE provider_api_keys
SET
provider_id = $2,
api_formats = $3,
auth_type_by_format = $39,
allow_auth_channel_mismatch_formats = $40,
auth_type_by_format = $40,
allow_auth_channel_mismatch_formats = $41,
auth_type = $4,
api_key = $5,
auth_config = $6,
@@ -1737,46 +1743,47 @@ SET
internal_priority = $10,
global_priority_by_format = $11,
rpm_limit = $12,
learned_rpm_limit = $13,
allowed_models = $14,
capabilities = $15,
cache_ttl_minutes = $16,
max_probe_interval_minutes = $17,
auto_fetch_models = $18,
locked_models = $19,
model_include_patterns = $20,
model_exclude_patterns = $21,
proxy = $22,
fingerprint = $23,
upstream_metadata = $24,
concurrent_limit = $13,
learned_rpm_limit = $14,
allowed_models = $15,
capabilities = $16,
cache_ttl_minutes = $17,
max_probe_interval_minutes = $18,
auto_fetch_models = $19,
locked_models = $20,
model_include_patterns = $21,
model_exclude_patterns = $22,
proxy = $23,
fingerprint = $24,
upstream_metadata = $25,
expires_at = CASE
WHEN $38::double precision IS NULL THEN NULL
ELSE TO_TIMESTAMP($38::double precision)
WHEN $39::double precision IS NULL THEN NULL
ELSE TO_TIMESTAMP($39::double precision)
END,
oauth_invalid_at = CASE
WHEN $25::double precision IS NULL THEN NULL
ELSE TO_TIMESTAMP($25::double precision)
WHEN $26::double precision IS NULL THEN NULL
ELSE TO_TIMESTAMP($26::double precision)
END,
oauth_invalid_reason = $26,
status_snapshot = $27,
concurrent_429_count = COALESCE($28, 0),
rpm_429_count = COALESCE($29, 0),
oauth_invalid_reason = $27,
status_snapshot = $28,
concurrent_429_count = COALESCE($29, 0),
rpm_429_count = COALESCE($30, 0),
last_429_at = CASE
WHEN $30::double precision IS NULL THEN NULL
ELSE TO_TIMESTAMP($30::double precision)
WHEN $31::double precision IS NULL THEN NULL
ELSE TO_TIMESTAMP($31::double precision)
END,
last_429_type = $31,
adjustment_history = $32,
utilization_samples = $33,
last_429_type = $32,
adjustment_history = $33,
utilization_samples = $34,
last_probe_increase_at = CASE
WHEN $34::double precision IS NULL THEN NULL
ELSE TO_TIMESTAMP($34::double precision)
WHEN $35::double precision IS NULL THEN NULL
ELSE TO_TIMESTAMP($35::double precision)
END,
last_rpm_peak = $35,
is_active = $36,
last_rpm_peak = $36,
is_active = $37,
updated_at = CASE
WHEN $37::double precision IS NULL THEN NOW()
ELSE TO_TIMESTAMP($37::double precision)
WHEN $38::double precision IS NULL THEN NOW()
ELSE TO_TIMESTAMP($38::double precision)
END
WHERE id = $1
"#,
@@ -1793,6 +1800,7 @@ WHERE id = $1
.bind(key.internal_priority)
.bind(&key.global_priority_by_format)
.bind(key.rpm_limit.map(|value| value as i32))
.bind(key.concurrent_limit)
.bind(key.learned_rpm_limit.map(|value| value as i32))
.bind(&key.allowed_models)
.bind(&key.capabilities)
@@ -2301,6 +2309,7 @@ fn map_key_row(row: &PgRow) -> Result<StoredProviderCatalogKey, DataLayerError>
})
})
.transpose()?;
let concurrent_limit = row_get::<Option<i32>>(row, "concurrent_limit")?;
let learned_rpm_limit = row_get::<Option<i32>>(row, "learned_rpm_limit")?
.map(|value| {
u32::try_from(value).map_err(|_| {
@@ -2466,6 +2475,7 @@ fn map_key_row(row: &PgRow) -> Result<StoredProviderCatalogKey, DataLayerError>
let mut key = key
.with_rate_limit_fields(
rpm_limit,
concurrent_limit,
learned_rpm_limit,
concurrent_429_count,
rpm_429_count,
@@ -2545,4 +2555,41 @@ mod tests {
assert!(sql.contains("total_cost_usd"));
}
}
#[test]
fn provider_api_keys_concurrent_limit_queries_include_field() {
for sql in [
super::LIST_KEYS_BY_IDS_PREFIX,
super::LIST_KEYS_BY_PROVIDER_IDS_PREFIX,
super::LIST_KEY_SUMMARIES_BY_PROVIDER_IDS_PREFIX,
] {
assert!(sql.contains("concurrent_limit"));
}
let source = include_str!("sql.rs");
assert!(source.contains("concurrent_limit,"));
assert!(source.contains("concurrent_limit = $13"));
assert!(source.contains(".bind(key.concurrent_limit)"));
assert!(source.contains("row_get::<Option<i32>>(row, \"concurrent_limit\")"));
}
#[test]
fn provider_api_keys_concurrent_limit_schema_is_nullable_without_default() {
let migration = include_str!(
"../../../migrations/20260502000000_add_provider_key_auth_channel_mismatch_formats.sql"
);
let concurrent_limit_line = migration
.lines()
.find(|line| line.contains("ADD COLUMN IF NOT EXISTS concurrent_limit"))
.expect("concurrent_limit migration line should exist")
.to_ascii_lowercase();
assert_eq!(
concurrent_limit_line.trim(),
"add column if not exists concurrent_limit integer;"
);
let baseline = include_str!("../../../bootstrap/20260413020000_baseline_v2.sql");
assert!(baseline.contains("CREATE TABLE IF NOT EXISTS public.provider_api_keys"));
assert!(baseline.contains("concurrent_limit integer,"));
}
}