Documentation
¶
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func NewErrorRetryingBlobAccess ¶
func NewErrorRetryingBlobAccess(base blobstore.BlobAccess, clock clock.Clock, randomNumberGenerator random.ThreadSafeGenerator, errorLogger util.ErrorLogger, initialInterval, maximumInterval, maximumDelay time.Duration) blobstore.BlobAccess
NewErrorRetryingBlobAccess creates a decorator for BlobAccess that performs retrying of Get() and FindMissing() operations that fail with INTERNAL, UNAVAILABLE or UNKNOWN gRPC status codes. Put() operations cannot be retried, as the buffer provided to this method is destroyed upon failure.
Retries are performed using exponential backoff, using an algorithm that the following blog post refers to as "Full Jitter": https://aws.amazon.com/blogs/architecture/exponential-backoff-and-jitter/
In pure gRPC contexts, a decorator like this isn't needed. Errors can be propagated all the way up, and tools such as Bazel can simply retry execution of build actions. Unfortunately, Bazel's OutputService interface doesn't provide facilities for bb_clientd to propagate transient errors. We have no choice but to do some basic retrying of FUSE file system operations.
Types ¶
This section is empty.
Source Files