[onedr0p]

What

Runs the e2e harness (crates/e2e run task) through cargo-nextest with a dedicated e2e profile that sets retries = 2, so a transiently-failing test is retried in a fresh process inside the same check run instead of failing the whole CI shard.

• .config/nextest.toml: new e2e profile - retries = 2, test-threads = 1 (the suite is stateful against one shared kind cluster; matches the old libtest --test-threads=1).
• crates/e2e/mise.toml: cargo test → cargo nextest run --profile e2e; flag mapping is 1:1 (--include-ignored → --run-ignored all, --nocapture → --no-capture, --no-fail-fast and --test <bin> sharding unchanged).
• .mise/config.toml + lockfile: pin cargo-nextest 0.9.140 via the aqua backend, locked for all 7 platforms.

The hermetic unit suite stays on plain cargo test - its second invocation compile-checks feature-gated targets, which nextest does not replace.

Why

On #307 two consecutive merge-queue runs failed on different e2e shards (apiserver-flap + cli-multicluster, then restore) with the same PR head, and the third unchanged attempt passed. Each flake evicts the queue entry (~30 min) and disarms auto-merge, requiring a manual re-queue. Retrying at the test level absorbs the flake where it happens; a retried-then-passed test reports FLAKY in the job log, so flakes stay visible instead of silently swallowed.

Verified

• cargo nextest list -P e2e -p kopiur-e2e --features e2e --run-ignored all compiles and enumerates every test binary with the new profile, no config warnings.
• The exact task invocation (incl. --test sharding and a name filter) parses and selects correctly.
• Behavior note: a name filter matching zero tests now fails (error: no tests to run) where libtest exited 0 - CI shards always match tests, so this only surfaces a typo'd KOPIUR_E2E_TESTFILTER locally, which failing loudly seems right for.

[greptile-apps]

Greptile Summary

This PR moves the stateful end-to-end suite to cargo-nextest with serial execution and per-test retries.

• Adds a dedicated e2e nextest profile with two retries and one test thread.
• Preserves ignored-test execution, output streaming, sharding, filtering, and no-fail-fast behavior.
• Pins cargo-nextest 0.9.140 through mise and locks its supported platform artifacts.
• Leaves the unit-test suite on standard cargo test.

Confidence Score: 5/5

The PR appears safe to merge with no actionable defects identified.

The e2e task runs from the workspace root, loads the requested nextest profile, preserves existing test-selection semantics, and receives cargo-nextest through the repository’s established mise provisioning.

T-Rex Logs

What T-Rex did

• The validation summary log consolidates version, warning-free enumeration, exact test selection, runtime execution, and effective profile assertions.
• The before-and-after e2e profile logs capture the requested profile parsing/listing, showing the state before and after.
• The profile consumption logs show the before/after values for NEXTEST_TEST_THREADS and NEXTEST_TOTAL_ATTEMPTS, changing from NEXTEST_TEST_THREADS=4 and NEXTEST_TOTAL_ATTEMPTS=1 to NEXTEST_TEST_THREADS=1 and NEXTEST_TOTAL_ATTEMPTS=3.
• The retries-after log captures the actual execution of the selected ignored e2e test under the changed command shape.

_{Ran code and verified through T-Rex}

Important Files Changed

Filename	Overview
.config/nextest.toml	Defines the dedicated e2e profile with the intended retry and serial-execution settings.
crates/e2e/mise.toml	Correctly migrates the e2e task from libtest to equivalent nextest arguments while preserving CI sharding and filtering.
.mise/config.toml	Adds the pinned cargo-nextest tool to the root mise configuration used by CI and nested tasks.
.mise/mise.lock	Locks cargo-nextest 0.9.140 artifacts and checksums for the relevant supported platforms.

_{Reviews (1): Last reviewed commit: "test(e2e): retry flaky tests with cargo-..." | Re-trigger Greptile}