Defining multiple endpoints in an app definition about tyk

[lonelycode]

Currently it's not possible, one endpoint per app definition I'm afraid.

It's an interesting idea though, since Tyk can be deployed load-balanced behind NGinX, the individual nodes would round-robin out to the hosts (which could essentially create a random pattern), so it might be better to assign targets randomly from the pool, at least we can hope for a standard distribution across endpoints that way.

Of course, running the API endpoint behind it's load balancer would would be a preferred option instead of trying to give Tyk load-balancing features (there's better, more mature kit out there to handle that sort of thing).

Alternatively, Tyk could be deployed on the same machine as the API endpoint, with the target URL being localhost, then these could be auto-scaled behind an ELB, you'd need to duplicate your app definition, but that could be done using the DB back-end option, then the only configuration that you would need to handle would be the Analytics purge, but you could tell all the nodes to not purge and have a dedicated analytics purge instance that does nothing but shovel data out of Redis.

I think I might be over-thinking it ;-)

It really depends on how you want to deploy tyk, as a self-contained application it is quite flexible and can adapt to many set ups.

What do you think?

[patrickmslatteryvt]

OK, that's what I suspected.
The simplest scenario that I can see is:
Client -> Nginx_gateway(443) -> Tyk(8080) -> Nginx_upstream(8443) -> Endpoints(01-06)

Where Nginx_gateway(443), Tyk(8080) and Nginx_upstream(8443) are all running on a single VM and the endpoints are running in their six separate VMs.
In our case the application endpoints are IIS on Windows so it's not possible to consolidate to a single host.

[lonelycode]

I see, that makes sense.

There is the possibility of trying to compile Tyk for windows, but I believe some of the deps aren't 100% compatible, though it should be a target I guess, will put it on the list :-)

On 12 Nov 2014, at 19:31, Patrick M. Slattery [email protected] wrote:

OK, that's what I suspected.
The simplest scenario that I can see is:
Client -> Nginx_gateway(443) -> Tyk(8080) -> Nginx_upstream(8443) -> Endpoints(01-06)

Where Nginx_gateway(443), Tyk(8080) and Nginx_upstream(8443) are all running on a single VM and the endpoints are running in their six separate VMs.
In our case the application endpoints are IIS on Windows so it's not possible to consolidate to a single host.

—
Reply to this email directly or view it on GitHub.

[lonelycode]

Fixed in master - you can now round-robin load balance to multiple upstream services using

    ...
    "enable_load_balancing": true,
    "target_list": [
        "http://server1", 
        "http://server2", 
        "http://server3"
    ],

In the proxy section of the API Definition