Partitioning test runs in CI

For CI scenarios where test runs take too long on a single machine, nextest supports automatically partitioning or sharding tests into buckets.

cargo-nextest supports both hash-based and count-based partitioning. Here’s an example of count-based partitioning, running tests in bucket 1 of 2.

Tests not in the current bucket are marked skipped.

Count-based partitioning is done per test binary. This means that the tests in one binary do not influence counting for other binaries.

Hash-based partitioning is similar, except buckets are specified in the format hash:m/n, where m is the current bucket and n is the number of buckets. The hash is completely deterministic, and is based on a combination of the binary and test names. For sufficiently large test runs, hash-based partitioning produces roughly the same number of tests per bucket.