fix: refine oauth auto-removal behavior

This commit is contained in:
beilo
2026-05-19 13:41:18 +08:00
parent 2ca4b486ec
commit 75aa3dc0cc
13 changed files with 388 additions and 40 deletions
@@ -1,7 +1,10 @@
use super::super::super::errors::{
merge_provider_oauth_refresh_failure_reason, normalize_provider_oauth_refresh_error_message,
};
use super::super::super::quota::shared::persist_provider_quota_refresh_state;
use super::super::super::quota::shared::{
persist_provider_quota_refresh_state, provider_auto_remove_banned_keys,
should_auto_remove_oauth_refresh_failed_key,
};
use super::super::super::runtime::refresh_provider_oauth_account_state_after_update;
use super::helpers::{self, RefreshDispatch, RefreshRequestContext, RefreshSuccessContext};
use super::response;
@@ -76,6 +79,42 @@ pub(super) async fn execute_admin_provider_oauth_refresh(
None,
)
.await?;
if provider_auto_remove_banned_keys(provider.config.as_ref()) {
let now_unix_secs = helpers::unix_now_secs();
let auto_removed = state
.cleanup_provider_catalog_key_if_current(
&provider,
&key_id,
|latest_key| {
should_auto_remove_oauth_refresh_failed_key(
latest_key,
now_unix_secs,
)
},
)
.await?;
if auto_removed {
tracing::info!(
trace_id = %trace_id,
key_id = %key_id,
provider_id = %provider.id,
provider_type = %provider_type,
event_name = "auto_removed_oauth_refresh_failed",
"gateway manual provider oauth refresh auto-removed expired key"
);
return Ok(RefreshDispatch::Respond(
response::oauth_refresh_auto_removed_response(&error_reason),
));
}
}
tracing::info!(
trace_id = %trace_id,
key_id = %key_id,
provider_id = %provider.id,
provider_type = %provider_type,
event_name = "refresh_failed_retained",
"gateway manual provider oauth refresh failure retained key"
);
}
}
return Ok(RefreshDispatch::Respond(
@@ -124,9 +163,27 @@ pub(super) async fn execute_admin_provider_oauth_refresh(
};
if !helpers::key_is_account_blocked(&key, OAUTH_ACCOUNT_BLOCK_PREFIX) {
let _ = state
let previous_oauth_refresh_issue = key
.oauth_invalid_reason
.as_deref()
.is_some_and(|reason| {
reason.lines().map(str::trim).any(|line| {
line.starts_with("[OAUTH_EXPIRED]") || line.starts_with("[REFRESH_FAILED]")
})
});
let cleared = state
.clear_provider_catalog_key_oauth_invalid_marker(&key_id)
.await?;
if cleared && previous_oauth_refresh_issue {
tracing::info!(
trace_id = %trace_id,
key_id = %key_id,
provider_id = %provider.id,
provider_type = %provider_type,
event_name = "refresh_fixed",
"gateway manual provider oauth refresh cleared oauth invalid marker"
);
}
}
let refreshed_key = state
@@ -24,6 +24,15 @@ pub(super) fn oauth_refresh_failed_bad_request_response(
)
}
pub(super) fn oauth_refresh_auto_removed_response(error_reason: impl AsRef<str>) -> Response<Body> {
Json(json!({
"status": "auto_removed",
"message": "已自动删除",
"detail": format!("Token 刷新失败且 Access Token 已过期,已自动删除:{}", error_reason.as_ref()),
}))
.into_response()
}
pub(super) fn oauth_refresh_failed_service_unavailable_response(
error_reason: impl Into<String>,
) -> Response<Body> {
@@ -1,6 +1,4 @@
use crate::handlers::admin::provider::shared::payloads::{
OAUTH_ACCOUNT_BLOCK_PREFIX, OAUTH_EXPIRED_PREFIX,
};
use crate::handlers::admin::provider::shared::payloads::OAUTH_ACCOUNT_BLOCK_PREFIX;
use axum::{
body::Body,
http,
@@ -146,9 +144,6 @@ pub(crate) fn merge_provider_oauth_refresh_failure_reason(
if current_reason.is_empty() {
return Some(refresh_reason.to_string());
}
if current_reason.starts_with(OAUTH_EXPIRED_PREFIX) {
return Some(current_reason.to_string());
}
if oauth_invalid_reason_is_account_level_block(Some(current_reason)) {
return None;
}
@@ -190,13 +185,13 @@ mod tests {
}
#[test]
fn refresh_failure_does_not_replace_access_token_expired_marker() {
fn refresh_failure_replaces_access_token_expired_marker() {
assert_eq!(
merge_provider_oauth_refresh_failure_reason(
Some("[OAUTH_EXPIRED] access token invalid"),
"[REFRESH_FAILED] Token 续期失败 (401): refresh_token 无效",
),
Some("[OAUTH_EXPIRED] access token invalid".to_string()),
Some("[REFRESH_FAILED] Token 续期失败 (401): refresh_token 无效".to_string()),
);
}
}
@@ -189,7 +189,6 @@ pub(crate) async fn update_existing_provider_oauth_catalog_key(
updated.encrypted_api_key = Some(encrypted_api_key);
updated.encrypted_auth_config = Some(encrypted_auth_config);
updated.api_formats = provider_oauth_catalog_key_api_formats(provider_type, api_formats);
updated.is_active = true;
updated.expires_at_unix_secs = expires_at_unix_secs;
updated.oauth_invalid_at_unix_secs = None;
updated.oauth_invalid_reason = None;
@@ -44,6 +44,15 @@ fn merge_codex_quota_metadata(
serde_json::Value::Object(merged)
}
fn codex_oauth_refresh_issue_reason(reason: Option<&str>) -> bool {
reason.is_some_and(|reason| {
reason
.lines()
.map(str::trim)
.any(|line| line.starts_with("[OAUTH_EXPIRED]") || line.starts_with("[REFRESH_FAILED]"))
})
}
pub(crate) async fn refresh_codex_provider_quota_locally(
state: &AdminAppState<'_>,
provider: &StoredProviderCatalogProvider,
@@ -56,8 +65,13 @@ pub(crate) async fn refresh_codex_provider_quota_locally(
let mut success_count = 0usize;
let mut failed_count = 0usize;
let mut auto_removed_count = 0usize;
let mut refresh_fixed_count = 0usize;
let mut refresh_failed_retained_count = 0usize;
let mut auto_removed_hard_banned_count = 0usize;
for key in keys {
let had_oauth_refresh_issue =
codex_oauth_refresh_issue_reason(key.oauth_invalid_reason.as_deref());
let transport = match state
.read_provider_transport_snapshot(&provider.id, &endpoint.id, &key.id)
.await?
@@ -261,13 +275,9 @@ pub(crate) async fn refresh_codex_provider_quota_locally(
}
}
let auto_removed = auto_remove_abnormal_keys
let auto_remove_candidate = auto_remove_abnormal_keys
&& should_auto_remove_structured_reason(oauth_invalid_reason.as_deref());
if auto_removed {
if state.delete_provider_catalog_key(&key.id).await? {
auto_removed_count += 1;
}
} else if !persist_provider_quota_refresh_state(
let persisted = persist_provider_quota_refresh_state(
state,
&key.id,
metadata_update.as_ref(),
@@ -275,8 +285,8 @@ pub(crate) async fn refresh_codex_provider_quota_locally(
oauth_invalid_reason.clone(),
None,
)
.await?
{
.await?;
if !persisted {
failed_count += 1;
results.push(json!({
"key_id": key.id,
@@ -286,6 +296,29 @@ pub(crate) async fn refresh_codex_provider_quota_locally(
}));
continue;
}
let auto_removed = if auto_remove_candidate {
state
.cleanup_provider_catalog_key_if_current(provider, &key.id, |latest_key| {
should_auto_remove_structured_reason(latest_key.oauth_invalid_reason.as_deref())
})
.await?
} else {
false
};
if auto_removed {
auto_removed_count += 1;
auto_removed_hard_banned_count += 1;
}
let refresh_fixed =
status == "success" && had_oauth_refresh_issue && oauth_invalid_reason.is_none();
if refresh_fixed {
refresh_fixed_count += 1;
}
let refresh_failed_retained =
status != "success" && oauth_invalid_reason.is_some() && !auto_removed;
if refresh_failed_retained {
refresh_failed_retained_count += 1;
}
if status == "success" {
success_count += 1;
@@ -321,6 +354,13 @@ pub(crate) async fn refresh_codex_provider_quota_locally(
}
if auto_removed {
payload.insert("auto_removed".to_string(), json!(true));
payload.insert("auto_removed_hard_banned".to_string(), json!(true));
}
if refresh_fixed {
payload.insert("refresh_fixed".to_string(), json!(true));
}
if refresh_failed_retained {
payload.insert("refresh_failed_retained".to_string(), json!(true));
}
results.push(serde_json::Value::Object(payload));
}
@@ -331,5 +371,8 @@ pub(crate) async fn refresh_codex_provider_quota_locally(
"total": results.len(),
"results": results,
"auto_removed": auto_removed_count,
"refresh_fixed": refresh_fixed_count,
"refresh_failed_retained": refresh_failed_retained_count,
"auto_removed_hard_banned": auto_removed_hard_banned_count,
})))
}
@@ -44,7 +44,7 @@ pub(super) fn default_provider_quota_execution_timeouts(
}
}
pub(super) fn provider_auto_remove_banned_keys(config: Option<&serde_json::Value>) -> bool {
pub(crate) fn provider_auto_remove_banned_keys(config: Option<&serde_json::Value>) -> bool {
admin_provider_quota_pure::provider_auto_remove_banned_keys(config)
}
@@ -52,6 +52,13 @@ pub(super) fn should_auto_remove_structured_reason(reason: Option<&str>) -> bool
admin_provider_quota_pure::should_auto_remove_structured_reason(reason)
}
pub(crate) fn should_auto_remove_oauth_refresh_failed_key(
key: &StoredProviderCatalogKey,
now_unix_secs: u64,
) -> bool {
admin_provider_quota_pure::should_auto_remove_oauth_refresh_failed_key(key, now_unix_secs)
}
pub(crate) fn normalize_string_id_list(values: Option<Vec<String>>) -> Option<Vec<String>> {
admin_provider_quota_pure::normalize_string_id_list(values)
}
@@ -1,5 +1,7 @@
use super::*;
use aether_data_contracts::repository::provider_catalog::StoredProviderCatalogProvider;
use aether_data_contracts::repository::provider_catalog::{
StoredProviderCatalogKey, StoredProviderCatalogProvider,
};
use axum::{
body::Body,
http,
@@ -264,6 +266,41 @@ impl<'a> AdminAppState<'a> {
Ok(affected)
}
pub(crate) async fn cleanup_provider_catalog_key_if_current<F>(
&self,
provider: &StoredProviderCatalogProvider,
key_id: &str,
should_delete: F,
) -> Result<bool, GatewayError>
where
F: FnOnce(&StoredProviderCatalogKey) -> bool,
{
let key_ids = [key_id.to_string()];
let Some(key) = self
.read_provider_catalog_keys_by_ids(&key_ids)
.await?
.into_iter()
.next()
else {
return Ok(false);
};
if key.provider_id != provider.id || !should_delete(&key) {
return Ok(false);
}
self.clear_admin_provider_pool_cooldown(&provider.id, &key.id)
.await;
self.reset_admin_provider_pool_cost(&provider.id, &key.id)
.await;
let deleted = self.delete_provider_catalog_key(&key.id).await?;
if deleted {
let deleted_key_ids = [key.id.clone()];
self.cleanup_deleted_provider_catalog_refs(&provider.id, &[], &deleted_key_ids)
.await?;
}
Ok(deleted)
}
pub(crate) async fn build_admin_pool_batch_action_response(
&self,
provider_id: &str,