feat: add river-backed async task queue

This commit is contained in:
2026-05-12 10:11:54 +08:00
parent d69aaed444
commit 7e220b7477
30 changed files with 1342 additions and 200 deletions
@@ -1,6 +1,7 @@
package runner
import (
"errors"
"fmt"
"strings"
@@ -54,6 +55,9 @@ func shouldRetrySameClient(candidate store.RuntimeModelCandidate, err error) boo
func retryDecisionForCandidate(candidate store.RuntimeModelCandidate, err error) retryDecision {
policy := effectiveRetryPolicy(candidate)
info := failureInfoFromError(err)
if errors.Is(err, store.ErrRateLimited) {
return retryDecision{Retry: false, Reason: "local_rate_limit_wait_queue", Match: policyRuleMatch{Source: "gateway_rate_limits", Policy: "rateLimitPolicy", Rule: "localCapacity", Value: "exceeded"}, Info: info}
}
if !boolFromPolicy(policy, "enabled", true) {
return retryDecision{Retry: false, Reason: "retry_disabled", Match: policyRuleMatch{Source: "model_runtime_policy_sets.retry_policy", Policy: "retryPolicy", Rule: "enabled", Value: "false"}, Info: info}
}
@@ -94,6 +98,9 @@ func failoverDecisionForCandidate(runnerPolicy store.RunnerPolicy, candidate sto
if cooldownSeconds <= 0 {
cooldownSeconds = 300
}
if errors.Is(err, store.ErrRateLimited) && store.RateLimitRetryable(err) {
return failoverDecision{Retry: true, Action: "next", Reason: "local_rate_limit_try_next_candidate", CooldownSeconds: cooldownSeconds, Match: policyRuleMatch{Source: "gateway_rate_limits", Policy: "rateLimitPolicy", Rule: "localCapacity", Value: "exceeded"}, Info: info}
}
if match, ok := failoverAllowMatchWithSources(runnerPolicy.FailoverPolicy, overridePolicy, info); ok {
return failoverDecision{Retry: true, Action: action, Reason: "failover_allow_policy", CooldownSeconds: cooldownSeconds, Match: match, Info: info}
}