Bug #21023
Updated by Peter Amstutz about 1 year ago
In sdk/go/keepclient getOrHead(), when a request fails with a retryable error, we don't wait between retries. We should use a randomized exponential backoff for congestion control similar to what we did with the API client.
This was found when looking at a program that talked to keep directly and made so many keepstore requests that it filled up the request queue, then started getting 503 errors. Because keepclient retried immediately instead of waiting, it did not give the overwhelmed keepstore a chance to drain its own queue and catch up.