Feature request: Process ID (count) added as an environment variable

[sondrelg]

Motivation

I would like to be able to run tests concurrently across multiple Redis databases.

While nextest allows me to run tests concurrently, I'm constrained to a single process by the fact that I have a shared storage layer with global queues, making it hard to isolate tests unless I run on separate databases.

Proposal

Add a NEXTEST_PROCESS_ID which would count 1, 2, 3, 4 for each process spawned.

This way I'd be able to deterministically do this in the app:

async fn get_test_connection() -> Connection {
    let db_index = 1 + std::env::var("NEXTEST_PROCESS_ID").expect("...");
    assert!(db_index <= 15, "The current implementation limits us to 14 processes");
    let redis_url = format!("redis://127.0.0.1:6379/{db_index}");
    ...
}

Alternatives

No response

Additional context

As an example, pytest-xdist from the python ecosystem does something similar: https://pytest-xdist.readthedocs.io/en/stable/how-to.html#envvar-PYTEST_XDIST_WORKER

They call their environment variables:

• PYTEST_XDIST_WORKER (current value)
• PYTEST_XDIST_WORKER_COUNT (total value)

[sunshowers]

Thanks -- this is an interesting idea. So in effect, set a number that increases from 1 and is unique for the duration the test is running.

I think one challenge here is that you can configure the number of slots for a test, setting it anywhere from 0 to the N in -jN. So we'll have to define the semantics carefully. But I think it's a reasonable thing to do.

[sunshowers]

Did this in #2145 since I believe we'll be able to use this at my workplace as well. Since nextest has test groups, with that PR we'll expose NEXTEST_TEST_GLOBAL_SLOT and _GROUP_SLOT environment variables. The numbers are unique for the lifetime of the test (and stable across retries), and also compact in the sense that the number assigned is the smallest one possible.

[sondrelg]

That was quick! The implementation sounds really nice, and consistency across runs sounds like an added bonus 👏 I'll take a stab at implementing the PR in my project to see how it runs!

[sunshowers]

(Note: stable across retries in the same run, if you configure nextest's retries functionality — not necessarily stable across multiple invocations of cargo nextest run.)

[sunshowers]

This is now out in cargo-nextest 0.9.90. Documentation here: https://nexte.st/docs/configuration/test-groups/#slot-numbers

[sondrelg]

Thanks a lot @sunshowers!