I have finally been able to work around the issue with the tests that appeared in CI. The work-around is quite brittle, so it might become necessary to find an actual fix soon.

As far as I understood, the issue arises as a consequence of three things:

• tower_batch::Batch::new spawns a task in the Tokio runtime;
• the primitives::*::Verifiers are stored in shared global variables, properly handled by Lazy wrappers;
• #[tokio::test] creates separate runtimes for separate tests

This means that shared batch verifiers will spawn a task on the runtime that first uses it, but that task will not survive for the next runtime (i.e., the next test execution), causing it to fail.

The workaround included here was to merge the two tests into a single test, so that they share the same runtime. This allows the spawned task to live enough for both calls to the verifier.

This workaround is brittle because it means that as more tests are added, more "test merging" will have to be done to avoid the underlying issue. I opened #2390 to track a more permanent solution.

This workaround is brittle because it means that as more tests are added, more "test merging" will have to be done to avoid the underlying issue. I opened #2390 to track a more permanent solution.

I also added another idea to this ticket: create a shared test verifier tokio runtime.

We already have once-off logging init in the zebra_test crate. It would be easy to add another function for a shared tokio runtime.