Add backend capacity policy configuration

This commit is contained in:
2026-07-29 20:48:10 +00:00
parent ffe6d261a9
commit 238fa90bfa
6 changed files with 268 additions and 17 deletions

View File

@@ -22,6 +22,9 @@ const (
openRouterEndpoint = "https://openrouter.ai/api/v1"
openRouterAPIKeyEnv = "OPENROUTER_API_KEY"
openRouterConcurrencyLimit = 16
defaultQueueCapacity = 1024
)
// ErrBackendNotFound identifies a registry lookup for an unknown backend ID.
@@ -43,9 +46,10 @@ func NewRegistry(additions []domain.Backend) (*Registry, error) {
definitions := make([]domain.Backend, 0, len(additions)+1)
definitions = append(definitions, domain.Backend{
ID: OpenRouterID,
Endpoint: openRouterEndpoint,
APIKeyEnv: openRouterAPIKeyEnv,
ID: OpenRouterID,
Endpoint: openRouterEndpoint,
APIKeyEnv: openRouterAPIKeyEnv,
ConcurrencyLimit: openRouterConcurrencyLimit,
})
definitions = append(definitions, additions...)
@@ -85,6 +89,25 @@ func (r *Registry) GetBackend(id string) (domain.Backend, error) {
return definition, nil
}
// CapacityPolicies returns a copy of the normalized policies for limited
// backends.
func (r *Registry) CapacityPolicies() map[string]domain.BackendCapacityPolicy {
policies := make(map[string]domain.BackendCapacityPolicy)
if r == nil {
return policies
}
for id, definition := range r.backends {
if definition.ConcurrencyLimit == 0 {
continue
}
policies[id] = domain.BackendCapacityPolicy{
ConcurrencyLimit: definition.ConcurrencyLimit,
QueueCapacity: definition.QueueCapacity,
}
}
return policies
}
func normalizeBackend(definition domain.Backend) (domain.Backend, error) {
definition.Endpoint = strings.TrimSpace(definition.Endpoint)
if err := validateEndpoint(definition.Endpoint); err != nil {
@@ -100,6 +123,40 @@ func normalizeBackend(definition domain.Backend) (domain.Backend, error) {
)
}
if definition.ConcurrencyLimit < 0 {
return domain.Backend{}, fmt.Errorf(
"backend %q concurrency limit must not be negative",
definition.ID,
)
}
if definition.QueueCapacity < 0 {
return domain.Backend{}, fmt.Errorf(
"backend %q queue capacity must not be negative",
definition.ID,
)
}
if definition.ConcurrencyLimit == 0 {
if definition.QueueCapacitySet {
return domain.Backend{}, fmt.Errorf(
"backend %q queue capacity requires a positive concurrency limit",
definition.ID,
)
}
definition.QueueCapacity = 0
} else {
if !definition.QueueCapacitySet {
definition.QueueCapacity = defaultQueueCapacity
definition.QueueCapacitySet = true
}
maxInt := int(^uint(0) >> 1)
if definition.QueueCapacity > maxInt-definition.ConcurrencyLimit {
return domain.Backend{}, fmt.Errorf(
"backend %q total capacity overflows int",
definition.ID,
)
}
}
keys := make([]string, 0, len(definition.ExtraParams))
for key := range definition.ExtraParams {
keys = append(keys, key)