[wangyinz]

Related to #807

Summary

• initialize explicit/auto/mixed host and every process-handle state
• execute exact tunnel/container/worker argv without shell metacharacters or shell=True
• probe the configured MongoDB and Dask endpoints before workers/frontends
• detect every owned-child early exit and startup timeout
• terminate and wait for already-started owned children before reporting failure
• retain a failed process handle when cleanup itself fails so shutdown can be retried
• keep frontend launch nonblocking and retain its process handle

Renewed error-contract review

The issue text changed all startup failures to MsPASSError(Invalid), but this class is an infrastructure launcher and its baseline public behavior uses ordinary Python runtime/configuration exceptions. Commit 60be0795 restores that boundary:

• malformed timeout/poll settings raise ValueError before any child starts
• owned-child early exit/readiness timeout raises RuntimeError
• process-creation/system exceptions are re-raised with their original identity after cleanup
• if cleanup fails, that failure is explicit and the failed handle is retained rather than silently discarded

The long launch and shutdown docstrings now document these boundaries.

Validation

• focused launcher lifecycle contract: 29/29 passed
• added original-exception identity and cleanup-retry-handle tests
• Python 3.12/3.13 py_compile and git diff --check: passed
• the new test file is Black 26.5.1 clean; the pre-existing launcher file has legacy whole-file formatting, so unrelated reformatting was deliberately excluded

This remains draft pending fresh CI and human infrastructure/API review.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 74.35%. Comparing base (482218e) to head (fbe477a).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #951   +/-   ##
=======================================
  Coverage   74.35%   74.35%           
=======================================
  Files         183      183           
  Lines       36516    36516           
  Branches     2540     2540           
=======================================
  Hits        27151    27151           
  Misses       8455     8455           
  Partials      910      910           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[wangyinz]

Renewed review found three blockers before merge:

1. Drop the test_workflow.py commit; that lifecycle fix was already superseded by Isolate the workflow test Dask client lifecycle #1004/Fix sliding-window pipeline result and resource contracts #885 on master.
2. Normalize an auto-discovered Dask host to the same complete endpoint used by workers (default tcp://HOST:8786) before the readiness probe. A bare host can hang beyond the advertised 2-second boundary.
3. Do not pipe long-lived child stdout/stderr unless they are drained. Inheriting the batch job output is the smallest safe default.

Rebase and merge this PR before #952 because both edit launcher.py. No maintainer API decision remains for #951.

[wangyinz]

Implemented the renewed review fixes on current master:

• dropped the superseded test_workflow.py lifecycle commit
• normalized auto Dask readiness addresses to tcp://HOST:8786 (or DASK_SCHEDULER_PORT)
• removed undrained pipes from long-lived owned children
• parsed configured runtime arguments with shlex

Final local validation: 30 focused launcher tests, py_compile, and git diff --check passed. The PR remains Draft until this final head passes CI.