- Replaced the e2e_tests nox session runner from sequential robot to the parallel pabot runner to shrink CI push times.
- Added _split_pabot_args and _pabot_parallel_args support to the e2e_tests session, mirroring the pattern used by integration_tests for consistent parallel execution control.
- CI workflow: added TEST_PROCESSES: "4" environment variable to the e2e_tests job to run four parallel workers. Rationale: E2E suites are IO-bound (LLM API calls) rather than CPU-bound, so increasing parallelism reduces wall-clock time without CPU contention.
- Updated the e2e_tests docstring to document parallelism control via the TEST_PROCESSES environment variable or the --processes positional argument.
- Pre-compiled bytecode comment updated to explain thundering-herd prevention for parallel workers.
- Template DB comment updated to explain the criticality of proper initialization for parallel pabot execution.
Key design decisions:
- Used pabot (robotframework-pabot>=4.0.0), a project-provided dependency, instead of introducing new tooling.
- Set TEST_PROCESSES=4 in CI (instead of the default min(cpu,2)) because E2E tasks are IO-bound and benefit from higher concurrency without CPU contention.
- Suite-level isolation via E2E Suite Setup (separate CLEVERAGENTS_HOME per suite) ensures workers do not share database state, enabling safe parallelism.
- Followed the exact same pattern as integration_tests for consistency and predictability.
Modules/components affected:
- e2e_tests nox session (parallelization logic and arg parsing)
- CI workflow (TEST_PROCESSES environment variable)
- E2E session docstring and related comments (documentation of parallelism and initialization)
- Inline comments for pre-compiled bytecode and template database initialization to reflect parallel execution considerations
ISSUES CLOSED: #1860
freemo