Implement different traffic distribution policies based on the header

[hkk1010]

Hi,there:
During canary deployment I want to add a policy of traffic forwarding based on header, that is access with header only to canary instance and others to primary and canary as follows.

So I added a flag (trafficType) in the CRD to identify the traffic distribution policy. and update istio.go.
This is a requirement of our business and I think there are many other users who also want to use this traffic forwarding strategy(#840). Can you add this function to it. Thank you very much

Signed-off-by: Henry He kuahe@cisco.com

[aryan9600]

Hi @hkk1010, thanks for this PR. This PR seems to try and address #1198, which will require some prior planning and discussion. We do want to add support for this, but this PR doesn't seem to solve it fully. I'm currently working on a PoC for this, and will open a PR soon. Meanwhile, if you want to discuss about this issue more, feel free to do so in the flagger channel in CNCF slack.

[hkk1010]

Hi @aryan9600 ,thanks for your reply.Do you have a specific schedule to add this feature.A month or two?We hope to have this feature soon,thanks

[devttao]

@aryan9600
Hello Sanskar , is there any updates on this issue?
could we get to know the plan or schedule for this feature?
Thank you!

[aryan9600]

hi @devttao, i opened a draft PR last week: #1280
i'm currently a bit bogged down with KubeCon stuff, i'll bring this into shape as soon as i'm done with that.

[chlunde]

@aryan9600 We really would like this feature as described here. I don't fully see the overlap.

An example use case would be to ensure internal developers (or all on-site employees) always hit the canary if there is one, and use the primary by default.

[aryan9600]

@chlunde maybe i'm misunderstanding the use case for this change, so let me dump my understanding here.
during a canary run, we want all requests from users that have a particular header to always hit the canary instance but all other users should experience a normal canary release (weight based routing)?

[chlunde]

@aryan9600 Sorry, I think I'm the one misunderstanding here, based on the drawing it looked like what I wanted, but I think this implementation is something entirely different. Sorry for the noise.