1. 1. Summary

The `resmoke_tests` commit-queue task can fail in its post-test step even when all tests pass, because the result-task-group activation hits a transient Evergreen API error. Example: a `503 Service Unavailable` on `POST /rest/v2/versions/

/activate_tasks` failed an otherwise-green task.

`buildscripts/evergreen_activate_result_tasks.py` already tried to harden against this by bumping the HTTP retry count, but the override was a *no-op*:

```python
evg_api = RetryingEvergreenApi.get_api(...)
evg_api._http_retry = Retry(total=DEFAULT_HTTP_RETRY_ATTEMPTS + 10, ...) # never takes effect
```

`RetryingEvergreenApi.get_api()` builds a sticky `requests.Session` at construction time with an `HTTPAdapter(max_retries=...)` already mounted using the *default* `Retry`. The `session` property returns that sticky session and `_create_session()` is never called again, so reassigning `evg_api._http_retry` mutates an attribute nothing re-reads — the mounted adapter keeps the original 10-attempt policy.

1. 1. Change

• Move the API construction into `get_evergreen_api()` which, after building the client, *re-mounts an `HTTPAdapter` with the new `Retry`* so the more aggressive policy is actually applied to outgoing requests.
• Add `EXTRA_HTTP_RETRY_ATTEMPTS` constant to make the intent explicit.
• Add `buildscripts/tests/test_evergreen_activate_result_tasks.py` plus the `py_library`/`py_test` bazel targets (the script had no test coverage before).

1. 1. Testing

• New unit test asserts the adapter that actually serves requests uses `DEFAULT_HTTP_RETRY_ATTEMPTS + EXTRA_HTTP_RETRY_ATTEMPTS` retries. It is a genuine regression guard: I verified that the previous "reassign `_http_retry` only" form leaves the mounted adapter at the default 10 attempts (test fails), while the fix raises it to 20 (test passes).

```
$ python -m unittest buildscripts.tests.test_evergreen_activate_result_tasks
Ran 2 tests in 0.001s
OK
```

1. 1. Notes

• This does not address the root cause (Evergreen API returning 503s); it makes the client resilient to short outages. The underlying 503 should be reported to DevProd separately.
• `SERVER-XXXX` placeholder in the title/commit — fill in before enqueueing.