Open ziltodian opened 2 months ago
Hi @ziltodian 👋🏻
Thanks a lot for your contribution 🙇🏻
We looked thoroughly into your use-case and PR with @codebien and reached the conclusion we'd like to pursue going in this direction indeed. However, we'd like to go even step further, and would like to propose more changes to your PR if you have the capacity to contribute further to it.
At the moment, the NetworkError
you have defined is a bit of a catch-all error. And while in your specific case, it would likely good enough, we'd like to provide more granularity to users using the library as we're at it, and eventually (not in this PR) extend to even more of the clients we provide.
As you might have found in the k6 documentation, the error_code that k6 returns offers some level of granularity and context as to what happened.
The error codes described in the doc fall essentially into X categories:
As a result, we'd like to define DNSError
, TCPError
, TLSError
, and HTTP2Error
types, which would be returned in case the error code received by k6 querying AWS is within the range defined in the range.
But what should we do with the standard HTTP error status codes, then? We suggest creating dedicated LambdaError
type, following the same convention as established by S3Client, with a dedicated mapping between the error name, and the HTTP status code as defined in the AWS API documentation.
We believe this would be the right design moving forward, not only for the lambda client, but for the whole library. In this PR our expectation is that we would modify only the lambda client (and maybe the internal/error.ts
file too). But that if we're happy with the result here, we would expand it to the whole codebase at a later point in time.
Is this something you'd have the capacity to work on? We wouldn't be able to merge the PR in its current state, but if you don't have the capacity to work on the proposed changes, let us know, and we'll try to contribute some of the changes we had in mind ourselves.
Let us know 🙇🏻
Quick heads-up that we have startd preparing a branch introducing the base error types I mentioned earlier, and will open a PR soon-ish.
The Lambda client treats all errors (apart from Lambda function errors) as AWS errors. This small change checks the
response.error_code
to see if the error is an HTTP error or some other kind of network error (https://k6.io/docs/javascript-api/error-codes/). It also exports bothAWSError
andNetworkError
. Our use case for this is that sometimes we encounter network errors where the Lambda invocation request never makes it to AWS. We need to identify when this occurs so we can exclude these failures from our test results. Currently we check if the error.name is "AWSError" and error.code is empty string. We'd like to be able to check for an instance ofNetworkError
instead.