Closed jlost closed 7 years ago
You are right - this will result in a subject being throttled until the accounting is reset. This is a related proposed patch: https://github.com/rabbah/openwhisk/commit/c8a31abf51b5b00eceb6d82a833a3e96fc642c6b (although does not address this particular issue yet which requires a bit more refactoring).
This issue is a known defect and related to https://github.com/apache/incubator-openwhisk/issues/2118#issuecomment-292360083.
Thanks for the info! Is there any reason that patch hasn't been accepted into the codebase yet?
I haven't opened the pull request yet (It's part of several inter-related fixes). I expect to do that very soon.
Environment details:
Steps to reproduce the issue:
Provide the expected results and outputs:
The result I would expect is that in step 1, a "message length exceeded" error is returned to the caller, and in step 3, "Too many requests" would either not occur (because they're quickly being completed, not still in-flight), or would go away after a short time.
Provide the actual results and outputs:
Blocking invocations only return the Activation ID after a timeout, and the OW instance becomes locked with "Too many requests". This seems like a Denial of Service vulnerability as well.
Code Samples
break-openwhisk.js:
spam.sh: