Upstream redirects are passed to client about rt-5gms-application-server

[davidjwbbc]

Have created a simple version of this solution (no housekeeping) using lua, but it requires the Openresty version of nginx to be installed (which has the ngx_http_lua module). I've tested the individual lua functions manually and they work just fine, but I need to check that this all works when used within nginx on real requests.

[davidjwbbc]

This is not going to be easy as it will require dynamic programming of proxy URLs in the AS when it sees a redirect response to a upstream request.

My initial suggestion would be to build a table mapping a UUID to seen redirection URL prefixes (redirection URL minus the final path element). Then it could map requests for {m4d-prefix}/{redirect-uuid}/object to the new redirected URL.

I'm not sure how much of this is possible with lua inside nginx, but it will be a marked degree of extra complexity.

[davidjwbbc]

@rjb1000 what are your thoughts on this issue?

[rjb1000]

@rjb1000 what are your thoughts on this issue?

Could the AS follow the redirect itself and make it sticky?

[davidjwbbc]

Not really because the client then uses the manifest URL as a base to request media segments and how do you then know if a media segment request is for that manifest or another one and therefore how to redirect it. I think the only way would be to rewrite the redirect Location in such a way that a request to that, or any relative resource locator, contains a unique identifier that allows tracing back to the original redirect.

[dsilhavy]

Nor sure if this helps but note that for DASH you can use Location and BaseURL elements in the manifest. Location points to the location of the manifest while BaseURL is used for segment requests.

[rjb1000]

Nor sure if this helps but note that for DASH you can use Location and BaseURL elements in the manifest. Location points to the location of the manifest while BaseURL is used for segment requests.

Manifest manipulation is still available as a last resort. But I think @davidjwbbc is keen to find a protocol-based solution if at all possible, to avoid having to resort to that.

[davidjwbbc]

Raised 5G-MAG/Standards#69 for SA4 discussion of this issue.

[davidjwbbc]

Demonstrated dynamic redirect learning working under test environment, will now deploy and run some more tests in a more representative environment (i.e. full AF & AS deployment with multiple provisioning sessions).