rover dev high cpu after composition failed

[benrea]

Description

When running rover dev cpu usage spikes to ~100% following composition failed i.e. all subgraphs removed, and remains high even if subgraphs & composition are restored.

💀 composition failed, killing the router
error: No subgraphs were found in the supergraph config.

Not specific to local development - also occurs with remote subgraph(s)
With multiple subgraphs, only occurs once all subgraphs removed

Doesn't occur if composition fails/no subgraphs were reachable on startup, though rover dev never retries in this scenario.
My desired setup is to run rover in docker for better local DX, which works but suffers the same issue.

Tested with rover 0.20.0, 0.21.0, 0.22.0, 0.23.0

Steps to reproduce

Example repo to reproduce with a single subgraph: https://github.com/benrea/rover-example
This mimics our setup using go + gqlgen + rover dev for local development against a federated graph.

Expected result

Cpu usage expected to spike during introspection(?) or genuine usage e.g. handling requests.

Actual result

Cpu usage spikes to ~100% and remains high, even after subgraphs/composition restored and operational.

Restarting rover / new process returns to normal.

Environment

Rover Info:
Version: 0.23.0
Install Location: /Users/benrea/.rover/bin/rover
OS: Mac OS 14.4.0 [64-bit]
Shell: /bin/zsh

[benrea]

Hack workaround... add a dummy subgraph to supergraph.yaml also fixes the 'nothing on startup' issue 😅

Was curious if schema-configured subgraph (instead of introspected) had the same issue

# supergraph.yaml
federation_version: =2.7.0
subgraphs:
  api:
    schema:
      subgraph_url: http://localhost:8080/graphql
  dummy:
    routing_url: http://localhost:9999/dummy
    schema:
      file: ./dummy.graphqls

# dummy.graphqls
type Query {
    dummy: String
}