More Error Handling

[isra17]

@aviau @slaroche

Add a few more advanced error handling, namely:

      error::ZeroDivisionError:
        - topic: logging-error
        - error_handler:
            set_handled: false
            republish:
              channel: default
              max_retry: 5

What the code above does is:

• topic: logging-error: Output a message to the logging-error topic. This message has the error and cause attributes which is the error details and the original message causing the exception. One way to use this would be to have specific topic that can output to specific sentry or error logging system.
• set_handled: fase: By default any error that match one of the error:... channel is set as "handled". This means that services like error logging or sentry will not handle them. Metric will increment a error_handled metric instead of the usual failed one. By using set_handled: false, the error will be reraise and the services will handle it as usual. Can be useful if we want to do something special with an error, but also use the default behavior.
• republish: ...: Not the same as topic: logging-error. republish will publish the original message, instead of creating a new message with {"error": ..., "cause": ...}. Therefor it can be used to publish to a queue fed to the same pipeline so it gets retried. It also set and increment a retried metadata attribute to the message, so we can stop publishing the message after a threshold. If the message if not published because of max retried, the error is set as unhandled and should be logged or sent to sentry.