feat: add `corrupt` toxic

[thalesmg]

trafficstars

Adds a toxic that flips random bits of the input stream, corrupting it.

Example run:

ͳ curl -v http://localhost:4001/api/echo -H "Content-Type: application/json" -d '{"echo": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"}'
*   Trying 127.0.0.1:4001...
* Connected to localhost (127.0.0.1) port 4001 (#0)
> POST /api/echo HTTP/1.1
> Host: localhost:4001
> User-Agent: curl/7.85.0
> Accept: */*
> Content-Type: application/json
> Content-Length: 77
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 200 OK
< cache-control: max-age=0, private, must-revalidate
< content-length: 76
< content-type: application/json; charset=utf-8
< date: Sun, 09 Oct 2022 17:55:36 GMT
< server: Cowboy
< x-request-id: Fxx4NnPZWF6J4lwAAABG
<
* Connection #0 to host localhost left intact
{"echo":"aaqaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"}

(Note that it changed one of the a's to a q. Many times it may corrupt the request such that a response is not sent back.)

[miry]

@thalesmg Thank you very much for contribution. I am going to review in more details, but here is something you can mean time do:

I am going to change Toxics to use a new way to manage Toxics. Here is an example: https://github.com/Shopify/toxiproxy/blob/ca6020993945335765b620146baa68656134cd38/_examples/toxics/http_toxic.go

It is explained here: https://github.com/Shopify/toxiproxy/blob/master/CREATING_TOXICS.md#using-ioreader-and-iowriter

Also to make sure it works, can you add a e2e test to https://github.com/Shopify/toxiproxy/blob/master/scripts/test-e2e#L170

[thalesmg]

@miry thanks for the pointers. I'll look at them and update the PR.

[xthexder]

I'd like to point out, this toxic produces behavior that would never be seen in the real world, unless the server or client is actually sending corrupt data.

TCP packets include a checksum, and will be automatically dropped by the network and retransmitted by the server if corruption occurs. The result is just added latency from the client's perspective.

[thalesmg]

@xthexder Indeed, if the corruption would be applied in the packet itself (I think like tc/netem does), then the observable effect would be just an increase in latency as the packets are dropped.

My intention with this toxic was to try to make it easier to observe strange error responses from services that typically use binary protocols, such as Kafka, Pulsar, etc. . Such responses could be the result of buggy drivers / protocol implementations, and the error responses may be hard to provoke without doing some kind of mocking in the request/response of the drivers.

Is there an interest in moving forward with this PR, under those assumptions?

[miry]

@thalesmg I think it is a valid example.

I agree, it would not be useful for most of users. In same time it is nice to have a fleet of toxics to simulate not just TCP, but Application layer protocols.

As alternative: it is still possible to build a custom toxiproxy server with custom toxics.

I would still think the better place for new toxics in the that repo.