feat(routing): add strategy failover controls

This commit is contained in:
elky
2026-09-09 09:12:09 +08:00
parent e58570d79d
commit f2839ae6a7
31 changed files with 1881 additions and 160 deletions
@@ -300,6 +300,34 @@ pub(crate) fn classify_local_failover(
policy: &LocalFailoverPolicy,
input: LocalFailoverInput<'_>,
) -> LocalFailoverClassification {
if input.status_code >= 400
&& policy.routing_rules.error_stop_patterns.iter().any(|rule| {
failover_pattern_matches(
&rule.pattern,
&rule.status_codes,
input.response_text,
input.status_code,
)
})
{
return LocalFailoverClassification::StopErrorPattern;
}
if input.status_code == 200
&& policy
.routing_rules
.success_failover_patterns
.iter()
.any(|rule| {
failover_pattern_matches(
&rule.pattern,
&rule.status_codes,
input.response_text,
input.status_code,
)
})
{
return LocalFailoverClassification::RetrySuccessPattern;
}
if policy.stop_status_codes.contains(&input.status_code) {
return LocalFailoverClassification::StopStatusCode;
}
@@ -487,13 +515,27 @@ fn local_failover_regex_rule_matches(
response_text: Option<&str>,
status_code: u16,
) -> bool {
if !rule.status_codes.is_empty() && !rule.status_codes.contains(&status_code) {
failover_pattern_matches(
&rule.pattern,
&rule.status_codes,
response_text,
status_code,
)
}
fn failover_pattern_matches(
pattern: &str,
status_codes: &std::collections::BTreeSet<u16>,
response_text: Option<&str>,
status_code: u16,
) -> bool {
if !status_codes.is_empty() && !status_codes.contains(&status_code) {
return false;
}
let pattern = rule.pattern.trim();
let pattern = pattern.trim();
if pattern.is_empty() {
return !rule.status_codes.is_empty();
return !status_codes.is_empty();
}
let Some(response_text) = response_text else {
@@ -507,6 +549,72 @@ fn local_failover_regex_rule_matches(
#[cfg(test)]
mod tests {
#[test]
fn routing_rules_precede_provider_rules_and_keep_provider_fallback() {
let policy = super::LocalFailoverPolicy {
routing_rules: aether_routing_core::RoutingFailoverRules {
success_failover_patterns: vec![aether_routing_core::RoutingFailoverRule {
pattern: "(?i)capacity.*exhausted".to_string(),
..Default::default()
}],
error_stop_patterns: vec![aether_routing_core::RoutingFailoverRule {
pattern: "invalid.*parameter".to_string(),
status_codes: [400].into_iter().collect(),
}],
..Default::default()
},
stop_status_codes: [200, 403].into_iter().collect(),
continue_status_codes: [400].into_iter().collect(),
..Default::default()
};
for (status, body, expected) in [
(
200,
"CAPACITY exhausted",
super::LocalFailoverClassification::RetrySuccessPattern,
),
(
400,
"invalid request parameter",
super::LocalFailoverClassification::StopErrorPattern,
),
(
400,
"capacity exhausted",
super::LocalFailoverClassification::RetryStatusCode,
),
(
403,
"permission denied",
super::LocalFailoverClassification::StopStatusCode,
),
(
429,
"rate limited",
super::LocalFailoverClassification::RetryUpstreamFailure,
),
] {
assert_eq!(
super::classify_local_failover(
&policy,
super::LocalFailoverInput::new(status, Some(body))
),
expected
);
}
}
#[test]
fn routing_transport_stop_cannot_be_overridden_by_provider() {
let policy = super::LocalFailoverPolicy {
stop_on_transport_errors: true,
..Default::default()
};
assert_eq!(
super::classify_local_transport_error(&policy),
super::LocalTransportFailoverClassification::StopTransportError
);
}
use std::collections::BTreeSet;
use super::{
@@ -4,7 +4,7 @@ use aether_contracts::ExecutionPlan;
use serde_json::{json, Value};
use tracing::debug;
use aether_routing_core::RoutingExecutionPolicy;
use aether_routing_core::{RoutingExecutionPolicy, RoutingFailoverRules};
use crate::provider_transport::GatewayProviderTransportSnapshot;
use crate::AppState;
@@ -14,6 +14,7 @@ pub(crate) const ROUTING_EXECUTION_POLICY_REPORT_FIELD: &str = "routing_executio
#[derive(Debug, Clone, PartialEq, Eq)]
pub(crate) struct LocalFailoverPolicy {
pub(crate) routing_rules: RoutingFailoverRules,
pub(crate) max_retries: Option<u64>,
pub(crate) max_transfer_count: u64,
pub(crate) max_transfer_timeout_seconds: u64,
@@ -29,6 +30,7 @@ pub(crate) struct LocalFailoverPolicy {
impl Default for LocalFailoverPolicy {
fn default() -> Self {
Self {
routing_rules: RoutingFailoverRules::default(),
max_retries: None,
max_transfer_count: 0,
max_transfer_timeout_seconds: 0,
@@ -61,8 +63,10 @@ pub(crate) async fn resolve_local_failover_policy(
Ok(Some(transport)) => local_failover_policy_from_transport(&transport),
Ok(None) | Err(_) => LocalFailoverPolicy::default(),
};
let cyber_continue_failover = routing_execution_policy_from_report_context(report_context)
.is_some_and(|policy| policy.cyber_continue_failover);
let routing_policy =
routing_execution_policy_from_report_context(report_context).unwrap_or_default();
let cyber_continue_failover = routing_policy.cyber_continue_failover;
policy.routing_rules = routing_policy.failover_rules;
policy.stop_cyber_policy_errors = !cyber_continue_failover;
debug!(
event_name = "local_failover_policy_loaded",
@@ -80,6 +84,8 @@ pub(crate) async fn resolve_local_failover_policy(
stop_on_transport_errors = policy.stop_on_transport_errors,
success_failover_pattern_count = policy.success_failover_patterns.len(),
error_stop_pattern_count = policy.error_stop_patterns.len(),
global_success_pattern_count = policy.routing_rules.success_failover_patterns.len(),
global_stop_pattern_count = policy.routing_rules.error_stop_patterns.len(),
cyber_continue_failover,
"gateway loaded local failover policy from transport snapshot"
);
@@ -122,6 +128,7 @@ pub(crate) fn local_failover_policy_from_transport(
});
LocalFailoverPolicy {
routing_rules: RoutingFailoverRules::default(),
max_retries,
max_transfer_count: provider_config
.and_then(|value| value.get("max_transfer_count"))
@@ -184,6 +191,10 @@ pub(crate) fn local_failover_policy_from_report_context(
.as_object()?;
Some(LocalFailoverPolicy {
routing_rules: object
.get("routing_rules")
.and_then(|value| serde_json::from_value(value.clone()).ok())
.unwrap_or_default(),
max_retries: object.get("max_retries").and_then(parse_u64_value),
max_transfer_count: object
.get("max_transfer_count")
@@ -267,6 +278,7 @@ fn parse_status_code_list(value: &Value) -> BTreeSet<u16> {
fn local_failover_policy_to_value(policy: &LocalFailoverPolicy) -> Value {
json!({
"routing_rules": policy.routing_rules,
"max_retries": policy.max_retries,
"max_transfer_count": policy.max_transfer_count,
"max_transfer_timeout_seconds": policy.max_transfer_timeout_seconds,
@@ -525,6 +537,7 @@ mod tests {
assert_eq!(
local_failover_policy_from_report_context(Some(&report_context)),
Some(LocalFailoverPolicy {
routing_rules: Default::default(),
max_retries: Some(2),
max_transfer_count: 10,
max_transfer_timeout_seconds: 60,