Duplicate series error even when LHS has no data.

[xujiaxj]

When we have a on(sth) operator query, and the right side has duplicate series, promxy returns the "found duplicate series..." message even when the left side has on data. The correct behavior, as I expect, should instead be returning empty data, like Prometheus does.

This is super easy to reproduce. Setup a Prometheus in a way we have two jobs scraping the same endpoint, for example

scrape_configs:
  - job_name: 'prometheus'
    static_configs:
      - targets: ['localhost:9090']

  - job_name: 'prometheus-another'
    static_configs:
      - targets: ['localhost:9090']

This will produce two up series with the same instance label but different job label:

up{instance="localhost:9090", job="prometheus"} 1
up{instance="localhost:9090", job="prometheus-another"} 1

If you run this query a * on(instance) up in Prometheus, it returns empty result as a does not exist.

Now start a local promxy pointing to the local Prometheus, and run the same query through promxy (I am running Grafana UI). See that you now have "found duplicate series..." error.

The same error happens if you run a * on() up.

If you just run a * up, it does not error out, but returns empty data.

[xujiaxj]

This looks like a bug to me.

[xujiaxj]

This probably can be fixed if we sync to the latest Prometheus code. This PR https://github.com/prometheus/prometheus/pull/9362/files would have fixed it. See the short-circuit change in engine.go, if either of the left or right is empty, it will just return nil instead of error out.