Coordinate retry attempts with backend permits

This commit is contained in:
2026-09-23 16:48:36 +00:00
parent 65d5894648
commit ae4540822e
10 changed files with 541 additions and 4 deletions

View File

@@ -28,13 +28,13 @@ func (c *client) Generate(
ctx context.Context,
req domain.GenerateRequest,
) (*domain.GenerateResponse, error) {
pool := c.manager.getPool(req.Target.BackendID)
if pool == nil {
if c.manager.getPool(req.Target.BackendID) == nil {
return c.next.Generate(ctx, req)
}
if err := pool.acquire(ctx); err != nil {
release, err := c.manager.Acquire(ctx, req.Target.BackendID)
if err != nil {
return nil, err
}
defer pool.releaseActive()
defer release()
return c.next.Generate(ctx, req)
}

View File

@@ -107,6 +107,23 @@ func (m *Manager) getPool(backendID string) *pool {
return m.pools[backendID]
}
// Acquire obtains one active-generation permit for a backend. Its release
// function is idempotent. Backends without a configured pool are unlimited.
func (m *Manager) Acquire(ctx context.Context, backendID string) (func(), error) {
if err := ctx.Err(); err != nil {
return nil, err
}
pool := m.getPool(backendID)
if pool == nil {
return releaseNothing, nil
}
if err := pool.acquire(ctx); err != nil {
return nil, err
}
var once sync.Once
return func() { once.Do(pool.releaseActive) }, nil
}
func (p *pool) acquire(ctx context.Context) error {
p.mu.Lock()
if err := ctx.Err(); err != nil {

View File

@@ -60,6 +60,26 @@ func TestNewManagerRejectsInvalidPolicies(t *testing.T) {
}
}
func TestManagerAcquireReleaseIsIdempotent(t *testing.T) {
manager, err := NewManager(map[string]domain.BackendCapacityPolicy{
"limited": {ConcurrencyLimit: 1},
})
if err != nil {
t.Fatal(err)
}
release, err := manager.Acquire(context.Background(), "limited")
if err != nil {
t.Fatal(err)
}
release()
release()
other, err := manager.Acquire(context.Background(), "limited")
if err != nil {
t.Fatal(err)
}
other()
}
func TestManagerAdmissionIsBoundedAndReleaseIsIdempotent(t *testing.T) {
policies := map[string]domain.BackendCapacityPolicy{
"limited": {