Discrepancy between seeded and unseeded `Testnet` nodes

[ok300]

It appears that Testnet::new(n) actually spins up n + 1 nodes.

Consider this test

    #[tokio::test]
    async fn single_key_republish_insufficient() {
        let dht = pkarr::mainline::Testnet::new(3).unwrap();
        let mut pkarr_builder = pkarr::ClientBuilder::default();
        pkarr_builder.bootstrap(&dht.bootstrap).no_relays();
        let pkarr_client = pkarr_builder.clone().build().unwrap();

        let public_keys = publish_sample_packets(&pkarr_client, 1).await;
        let public_key = public_keys.first().unwrap().clone();

        let mut settings = RepublisherSettings::default();
        settings
            .pkarr_client(pkarr_client)
            .min_sufficient_node_publish_count(NonZeroU8::new(4).unwrap());
        let publisher = MultiRepublisher::new_with_settings(settings, Some(pkarr_builder));
        let results = publisher.run_serially(public_keys).await.unwrap();
        let result = results.get(&public_key).unwrap();
        assert!(result.is_err());
    }

This test used to work before pkarr and mainline were upgraded in #315 .

There are two ways this test can be made to work:

• by using Testnet::new_unseeded(3) in the first line, to instantiate the dht
• by bumping the argument of min_sufficient_node_publish_count to 5
  • since the test expects an err, the fact a 5 here succeeds and 4 fails hints at the fact that the test published to 4 nodes

Both suggest that Testnet::new(n), which internally uses Testnet::build_seeded, actually spins up 4 nodes.

Please investigate and re-align the behavior of Testnet::new(n) with Testnet::new_unseeded(n) in terms of created nodes.

[SeverinAlexB]

The issue may not necessarily be in Testnet spinning up more nodes but could also be that republisher overcounts or the pkarr relay is counted as an additional node. So whoever tackles this issue, be open minded.

[ok300]

Possibly related to the fix in #320

In that PR, the tests that this issue describes (e.g. those needing a switch to new_unseeded) , appear to work fine¹ with a seeded variant.

Footnotes

1. They actually work better, because the fixes in that PR also addresses failing tests on macOS. ↩