Retry IOException and http status >= 500

[mrichards42]

Dependencies

                 [com.cognitect.aws/api "0.8.505"]
                 [com.cognitect.aws/endpoints "1.1.11.976"]
                 [com.cognitect.aws/s3 "811.2.865.0"]

Description

Since switching from the Java SDKs, I've noticed an increase in the number of API calls that result in failures. The majority of errors I've seen (IOException and http status 500) are considered retriable by the Java SDK, but in this library are marked :cognitect.anomalies/fault, which isn't retried by default.

This issue is similar to #162 and #79, but should provide more information.

I understand that I can handle this by using a custom retriable? function, but I hope you'll consider retrying at least 500 and 502 statuses, and IOException by default. Thanks!

Retriable conditions

Exceptions

The Java SDK retries all IOExceptions (source).

HTTP status codes

The Java SDK retries the following http statuses: 500, 502, 503, 504 (source), while the JS SDK retries statuses >= 500 (source).

Error codes

Error codes are, unfortunately, a little less consistent between the Java and JS SDKs, so I understand that these might be out of scope. For the sake of completeness, this is what I've gleaned from the two SDKs:

Codes retried by both Java and JS SDKs

EC2ThrottledException
ProvisionedThroughputExceededException
RequestLimitExceeded
RequestThrottled
RequestThrottledException
Throttling
ThrottlingException
TooManyRequestsException
TransactionInProgressException

These are ClockSkew errors, and are retried by default in Java, but not JS

AuthFailure
InvalidSignatureException
RequestExpired
RequestInTheFuture
RequestTimeTooSkewed
SignatureDoesNotMatch

Codes retried by only the Java SDK

BandwidthLimitExceeded
PriorRequestNotComplete
RequestTimeout
RequestTimeoutException
SlowDown
ThrottledException

Codes retried by only the JS SDK

ExpiredTokenException
NetworkingError
TimeoutError