Introspection query doesn't return latest schema changes when query plan is being persisted on Redis

[nathanmarcos]

Describe the bug
Starting from version 1.44.0, up to latest 1.45.0, the introspection query is not returning new parts of the schema when persisting the query plan on Redis and router is using a supergraph from a file.

To Reproduce
Steps to reproduce the behavior:

1. Start redis:
   • docker run --rm -d -p 6379:6379 redis
2. Start stock router version 1.44.0 or 1.45.0 as following:
   • APOLLO_KEY="..." APOLLO_GRAPH_REF="..." APOLLO_ROUTER_SUPERGRAPH_PATH="examples/graphql/supergraph.graphql" APOLLO_ROUTER_CONFIG_PATH="router.yaml" cargo run -- --dev

# router.yaml
supergraph:
  query_planning:
    cache:
      in_memory:
        limit: 1024
      redis:
        timeout: 10ms
        urls:
          - "redis://127.0.0.1"

3. Open sandbox at http://127.0.0.1:4000/ and make sure schema polling is on
4. Add a new query to examples/graphql/supergraph.graphql and save the file
5. Router will hot reload
6. The new query doesn't appear on sandbox

PS: If you manually delete the entry on Redis for the plan:v2.7.2:06402fd188... that contains the {"Ok":{"Response":{"response":{"data":{"__schema":{"queryType":{"name":"Query"},"..., a new entry will be created and the new schema will immediately be delivered and the new query will be visible on sandbox.

Expected behavior
Router should serve via introspection all schema changes after a reload or hot reload even when the query plan is being persisted on Redis.

Desktop (please complete the following information):

• OS: macOS Sonoma
• Version: 14.4.1

Additional context
Router version: 1.44.0 or 1.45.0

[Geal]

I think that is related to #4883
The query plan cache key uses a hash of the query that takes into account the relevant parts of the schema to avoid planning again queries that are not affected by a schema change, but I don't think it takes into account introspection queries.
We will soon separate introspection queries from the query planner cache so that will be solved definitely then, but in the meantime I think we should fix this quickly

The schema aware hash is in :

router/apollo-router/src/spec/query/change.rs

Lines 72 to 80 in a709cff

	pub(crate) fn hash_query(
	schema: &'a schema::Schema,
	executable: &'a executable::ExecutableDocument,
	operation_name: Option<&str>,
	) -> Result<Vec<u8>, BoxError> {
	let mut visitor = QueryHashVisitor::new(schema, executable);
	traverse::document(&mut visitor, executable, operation_name)?;
	Ok(visitor.finish())
	}

[abernix]

Thanks for opening this, @nathanmarcos. We're working on a 1.45.1 patch release with a couple things right now. In the meantime, are you in any position to be able to try the PR that is opened that purports to fix this (#5007)? (If you necessitate a Docker image or something, we could probably make that happen.)

[nathanmarcos]

Hello @abernix, I tested it, and it works just fine. 🎉

Thanks @Geal for the quick fix. 🚀

[abernix]

Thanks for the validation, @nathanmarcos. Just to set expectations, it'll probably be a Friday release with that above patch and probably one other smaller fix.

[nathanmarcos]

No problem. That sounds good, thanks @abernix . 😉

[abernix]

This was released in https://github.com/apollographql/router/releases/tag/v1.45.1 via #5007. Note that we also brought in another (more critical) hashing fix to that release.