Regression in max_retries usage on GetSerialPortOutput()

[same-id]

After v2.0.2, max_retries is ignored in GetSerialPortOutput().
This previously worked on v1.12.8.

All our tests are failing since they used this feature of the python library.

I traced the problem to this commit:

#1116 and #1117

c6912836 - fix: handle error on service not enabled (#1117)

1117 changed the default reason to be error.status which is PERMISSION_DENIAD instead using error.errors[0].reason which is rateLimitExceeded and therefore later ignored in:

# Only retry on rate limit related failures.
if reason in ("userRateLimitExceeded", "rateLimitExceeded"):
    return True

Example for such json:

{'error': {'code': 403,
           'errors': [{'domain': 'usageLimits',
                       'message': 'Quota exceeded for quota group '
                                  "'GetSerialPortOutputGroup' and limit "
                                  "'Instance GetSerialPortOutput requests per "
                                  "user per 100 seconds' of service "
                                  "'compute.googleapis.com' for consumer "
                                  "'project_number:REDACTED'.",
                       'reason': 'rateLimitExceeded'}],
           'message': 'Quota exceeded for quota group '
                      "'GetSerialPortOutputGroup' and limit 'Instance "
                      "GetSerialPortOutput requests per user per 100 seconds' "
                      "of service 'compute.googleapis.com' for consumer "
                      "'project_number:REDACTED'.",
           'status': 'PERMISSION_DENIED'}}

I think that correct fix for #1117 should not have changed the order of lookup - first check for error.errors[0].reason and if it doesn't exist fallback to error.status

Sam

[parthea]

I'm sorry that we introduced a regression! I have a potential fix in #1244. I still need to add unit tests and update the documentation.