HttpStatus.UNKNOWN should be retried

[dsaxton-1password]

Expected Behavior

Unknown HTTP status codes should also be retried. This is because they can sometimes be caused by things like TLS handshake timeouts which may be transient.

Current Behavior

Many records are failing to be sent to Amplitude in an AWS Lambda cold start situation because of errors such as this:

• The handshake operation timed out
• The write operation timed out

(Each with a -1 status code)

In theory these could be recoverable so we should try.

Possible Solution

One line code change:

--- a/src/amplitude/processor.py
+++ b/src/amplitude/processor.py
@@ -15,7 +15,7 @@ class ResponseProcessor:
         if res.status == HttpStatus.SUCCESS:
             self.callback(events, res.code, "Event sent successfully.")
             self.log(events, res.code, "Event sent successfully.")
-        elif res.status == HttpStatus.TIMEOUT or res.status == HttpStatus.FAILED:
+        elif res.status in (HttpStatus.TIMEOUT, HttpStatus.FAILED, HttpStatus.UNKNOWN):
             self.push_to_storage(events, 0, res)
         elif res.status == HttpStatus.PAYLOAD_TOO_LARGE:
             if len(events) == 1:

Steps to Reproduce

Environment

[crleona]

Hi @dsaxton-1password - the current behavior is consistent across our tests and other SDKs, so we'll discuss this internally.

[crleona]

We've added this in #70, released in v1.2.2.