Add retry for IBM Cloud API rate limiting (HTTP 429) #502
Description
Description
Summary:
IBM Cloud APIs return HTTP 429 (Too Many Requests) when rate limited, along with a Retry-After header indicating how long to wait. Currently, the codebase does not handle this.
Goal:
Introduce a shared retry helper.
pkg/cloudprovider/ibm/retry.go
// DoWithRetry handles HTTP 429 rate limiting with exponential backoff.
// It retries up to 5 times, respecting the Retry-After header if present.
func DoWithRetry[T any](ctx context.Context, fn func() (T, *core.DetailedResponse, error)) (T, error) {
var zero T
backoff := 100 * time.Millisecond
for attempt := 0; attempt < 5; attempt++ {
result, response, err := fn()
// Success or non-rate-limit error
if response == nil || response.StatusCode != 429 {
return result, err
}
// Parse Retry-After header if present
if ra := response.Headers.Get("Retry-After"); ra != "" {
if secs, _ := strconv.Atoi(ra); secs > 0 {
backoff = time.Duration(secs) * time.Second
}
}
select {
case <-ctx.Done():
return zero, ctx.Err()
case <-time.After(backoff):
backoff = min(backoff*2, 30*time.Second)
}
}
return zero, fmt.Errorf("rate limited after retries")
}Possible Solutions:
Option 1
Wrap at the client level. This is what I think is a better approach. But then this retry logic will be called whether or not it is batched.
Before:
func (c *GlobalCatalogClient) GetPricing(ctx context.Context, catalogEntryID string) (*globalcatalogv1.PricingGet, error) {
if err := c.ensureClient(ctx); err != nil {
return nil, err
}
if sdkClient, ok := c.client.(*globalcatalogv1.GlobalCatalogV1); ok {
pricingOptions := &globalcatalogv1.GetPricingOptions{
ID: &catalogEntryID,
}
pricingData, _, err := sdkClient.GetPricing(pricingOptions)
if err != nil {
return nil, fmt.Errorf("calling GetPricing API: %w", err)
}
return pricingData, nil
}
return nil, fmt.Errorf("invalid client type for GetPricing")
}After:
func (c *GlobalCatalogClient) GetPricing(ctx context.Context, catalogEntryID string) (*globalcatalogv1.PricingGet, error) {
if err := c.ensureClient(ctx); err != nil {
return nil, err
}
if sdkClient, ok := c.client.(*globalcatalogv1.GlobalCatalogV1); ok {
pricingOptions := &globalcatalogv1.GetPricingOptions{
ID: &catalogEntryID,
}
pricingData, err := DoWithRetry(ctx, func() (*globalcatalogv1.PricingGet, *core.DetailedResponse, error) {
return sdkClient.GetPricing(pricingOptions)
})
if err != nil {
return nil, fmt.Errorf("calling GetPricing API: %w", err)
}
return pricingData, nil
}
return nil, fmt.Errorf("invalid client type for GetPricing")
}Option 2
Retry at the executor/batcher level. But then we will need to change all client interface signatures used by batchers.
Before:
type pricingClient interface {
GetPricing(ctx context.Context, catalogEntryID string) (*globalcatalogv1.PricingGet, error)
}
func (p *PricingBatcher) execPricingBatch() BatchExecutor[PricingQueryInput, globalcatalogv1.PricingGet] {
return func(ctx context.Context, inputs []*PricingQueryInput) []Result[globalcatalogv1.PricingGet] {
results := make([]Result[globalcatalogv1.PricingGet], len(inputs))
if len(inputs) == 0 {
return results
}
id := inputs[0].CatalogEntryID
out, err := p.client.GetPricing(ctx, id)
for i := range inputs {
results[i] = Result[globalcatalogv1.PricingGet]{Output: out, Err: err}
}
return results
}
}After:
type pricingClient interface {
GetPricing(ctx context.Context, catalogEntryID string) (*globalcatalogv1.PricingGet, *core.DetailedResponse, error)
}
func (p *PricingBatcher) execPricingBatch() BatchExecutor[PricingQueryInput, globalcatalogv1.PricingGet] {
return func(ctx context.Context, inputs []*PricingQueryInput) []Result[globalcatalogv1.PricingGet] {
results := make([]Result[globalcatalogv1.PricingGet], len(inputs))
if len(inputs) == 0 {
return results
}
id := inputs[0].CatalogEntryID
out, err := ibm.DoWithRetry(ctx, func() (*globalcatalogv1.PricingGet, *core.DetailedResponse, error) {
return p.client.GetPricing(ctx, id)
})
for i := range inputs {
results[i] = Result[globalcatalogv1.PricingGet]{Output: out, Err: err}
}
return results
}
}