Potential undocumented breaking change in rover dev when going from 0.27.1 to 0.27.2

Router
1

Hey folks…curious if anyone has noticed this issue with rover dev

in < 0.27.2 I could omit the routing_url from a subgraph config if I was specifying a subgraph_url for introspection

However in >= 0.27.2, it seems that having a routing_url is required even if I have a subgraph_url defined, otherwise the query execution will route to the deployed subgraph url rather than my local subgraph override

2

Hi @Chris_Coker , thanks for posting about Rover dev.

Can you share how you reproduce the issue because this works for me on Rover v0.27.2

supergraph config

federation_version: =2.10.0
subgraphs:
  extra:
    schema:
      subgraph_url: http://localhost:4001/extra/graphql

start with rover dev

rover dev \
  --supergraph-config supergraph.yaml \
  --router-config router.yaml
3

Hi here are configs. One thing to note is that in my config I’m claiming a graphref in the supergraph.yaml file and on the rover dev command. I don’t see that in your supergraph config. So, I don’t see how rover would be able to “fallback” ( even unintentionally) to the deployed subgraph routing_url rather than using localhost:4001 in your configs.

supergraph.yaml

federation_version: =2.10.0
graphref: <my-graph-ref>
subgraphs:
    extra:
        schema:
            subgraph_url: http://localhost:4001/graphql

rover dev

rover dev \
   --supergraph-config supergraph.yaml \
   --router-config router.yaml \
   --graph-ref <my-graph-ref>

Also have this shell ENV var set

APOLLO_KEY=service:*****
APOLLO_ROVER_DEV_ROUTER_VERSION=2.0.0
4

This is as intended

  • The graph_ref in you supergraph.yaml is not doing anything. You would need to specify that on a per subgraph: Rover supergraph Commands - Apollo GraphQL Docs
  • Setting a --graph-ref on the rover dev command will fetch all schemas and routing urls from that source and then the supergraph.yaml is overriding those.
    • So what it is assuming is that you want to use the routing_url from the graph-ref and the schema from introspecting the url. If that is not the case then you would need to provide both the routing_url and schema of the subgraph to override the values from GraphOS
1 Like
5

Thanks for the reply.

One thing that is weird is that I don’t experience this issue for some reason with rover 0.27.1, using the same configs ( with the exception of the APOLLO_ROVER_DEV_ROUTER_VERSION=2.0.0 and federation_version: =2.10.0 having lower values 1.58.0 and 2.9.0 respectively

Also there is documentation in the repo that indicates that if subgraph_url is provided the routing_url can be omitted

That’s why I’m concerned about it potentially being a breaking change

6

@Chris_Coker I will share with the Rover team to get clarification if this was intended, but just to be clear that in my opinion the current behavior is as intended in which case the old behavior is a bug…

By specifying the --graph-ref flag you are basically specifying the routing_url and the schema option for all your subgraphs.

Then in the supergraph.yaml you can choose to override each one for each subgraph.

If you had two subgraph A + B and used the graph ref option plus this config

federation_version: =2.10.0
subgraphs:
    A:
      schema:
          subgraph_url: http://localhost:4001/graphql

What you are saying is

  • For subgraph A
    • Get the routing URL from graph ref
    • Get the schema from an intospection url
  • For subgraph B
    • Get the routing URL from graph ref
    • Get the schema from graph ref

What you are referring to in the docs is when there is no routing url set, then we assume we can use the introspection url, but by setting graph-ref you have set a routing url

1 Like
7

Thanks so much. I would be very interested in learning what you hear from the rover team. I was looking into the code earlier and did actually find a comment in the code on 0.26.3 that actually outlines the exact scenario. The comment seems to imply that the behavior in 0.26.3 was a known limitation,

1 Like
8

actually appears the functionality was changed in 0.27.2. The above code snippet is present in the code in 0.27.1

9

Hey @Chris_Coker, I talked with the Rover engineering team and got some answers.

This does appear to be an unintended change due to a side effect of fixing the rover startup with our recent replatform.

To help make sure others users see this change we are updating the CHANGELOG: Add changelog entry for new behaviour w.r.t subgraph_url by jonathanrainer · Pull Request #2420 · apollographql/rover · GitHub

The code was removed in 0.27.2 but it was unused since 0.27.0 and wasn’t cleaned up, so 0.26.3 → 0.27.0 is where the change gets introduced.

If you would like to go back to the old behavour you can set both the subgraph_url and routing_url to your local host

1 Like