T3348-Fix-failing-quality_test_Global_Childpool

[loris-fab]

Goal

Clicking the "365 children" button on the Global Childpool search page appeared to hang indefinitely (frozen screen, endless spinner), reported as a functional bug.

The feature wasn't actually broken: it ran 365 sequential HTTP requests to the GMC API (one per day of the year), taking ~1m30. Making them concurrent then exposed two latent defects in the shared GMC connector, fixed
here as well.

Technical aspect

• do_365_mix() (wizards/global_child_search.py) now fires the 365 (366 on leap years) requests concurrently via a ThreadPoolExecutor (15 workers), then re-sorts the results chronologically before creating the
  compassion.global.child records.
• Two correctness bugs found in the original loop:
  • "Second child" pagination: add_search() did self.skip += self.take (0 + 1), so every date from Jan 2nd onwards asked for the second best match. Dates with a single matching child returned nothing and
    were wrongly listed in missing_dates. skip is now 0 for all requests.
  • Misleading nb_found: it was overwritten on each iteration, so it only reflected December 31st. It is now the sum over all successful daily responses.
• Error handling: the old loop caught UserError without distinguishing "no child for that date" from an API error, so GMC failures were silently shown as "Missing birthdates". Only empty 200 responses are listed
  as missing now; real errors surface the GMC message.
• OnrampConnector thread-safety (message_center_compassion, shared by every GMC call in the repo):
  • It is a singleton whose construction and token refresh read res.config.settings through the request's DB cursor. Building it inside the workers used that cursor concurrently. It is now instantiated once on the
    request thread; workers only call send_message(), which never touches the ORM.
  • After a ConnError, send_message() rebuilt its requests.Session restoring only the params, dropping the bearer token. As the session is process-wide, one network drop left every later GMC call
    unauthenticated. The new session now inherits the headers too.
• Added a confirmation popup on the button (views/global_childpool_view.xml) to set expectations on the search duration. Now it take 10 s instead of 1m30.

Misc

• This PR also touches message_center_compassion: the connector changes are corrective (no API change) but affect all outbound GMC calls. ta

Outcome

The end result is a 365 children button that works 10 times faster.

[Copilot]

Pull request overview

This PR addresses a perceived “infinite loading” issue on the Global Childpool “365 children” action by parallelizing daily GMC API calls and improving UX with a confirmation dialog, while also fixing correctness issues in the prior sequential logic.

Changes:

• Reworked do_365_mix() to prepare per-day requests and execute them concurrently via ThreadPoolExecutor, then re-sort results chronologically.
• Fixed correctness issues around pagination (skip) and the nb_found counter by resetting skip and summing NumberOfBeneficiaries.
• Added a confirmation popup on the “365 children” button to set user expectations before starting the long-running search.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
child_compassion/wizards/global_child_search.py	Parallelizes the 365/366-day GMC search and adjusts result aggregation/state updates.
child_compassion/views/global_childpool_view.xml	Adds a user confirmation prompt before launching the “365 children” search.

Suppressed comments (3)

child_compassion/wizards/global_child_search.py:368

• future.result() will raise if any HTTP call errors, which will abort the entire 365-day search and leave the wizard in a partially-reset state. It’s safer to handle exceptions per-future and record the date as missing (or surface a summarized error) rather than crashing the whole run.

        with concurrent.futures.ThreadPoolExecutor(max_workers=15) as executor:
            futures = [executor.submit(fetch_http, req) for req in prepared_requests]
            for future in concurrent.futures.as_completed(futures):
                results.append(future.result())

child_compassion/wizards/global_child_search.py:380

• OnrampConnector.send_message() sets result["content"] to a string when the response isn’t valid JSON. In that case, result.get("content", {}).get(...) will raise because strings don’t have .get(), breaking the whole loop. Guard the type before treating content as a dict.

        for c_date, result in results:
            if result.get("code") == 200 and result.get("content", {}).get(result_name):
                total_matching_found += result["content"].get(
                    "NumberOfBeneficiaries", 0
                )
                children_data = result["content"][result_name]
                for child_data in children_data:

child_compassion/wizards/global_child_search.py:317

• missing_dates is only written at the very end of the method. If an exception occurs before the final write() (for example during concurrent requests), the wizard can keep stale missing_dates from a previous run while already having deleted global_child_ids. Reset it up-front to keep the wizard state consistent even on failure.

        self.global_child_ids.unlink()

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[greptile-apps]

Confidence Score: 5/5

T-Rex Logs

What T-Rex did

• Ran an isolated no-network harness with 15 concurrent workers to exercise the OnrampConnector retry path, where each worker first raised a ConnectionError before retrying.
• Observed the pre-fix behavior in the before-capture run: retry sessions kept parameters but did not retain Authorization, and all retries returned 401.
• Observed the post-fix behavior in the after-capture run: retry sessions retained both Authorization and parameters, and all retries returned 200.
• Uploaded the onramp-session-retry harness source and the before/after capture logs to document the comparison.
• Concluded that the current header-copying implementation preserves authentication during concurrent retries, as shown by the post-fix results.

_{Ran code and verified through T-Rex}

_{Reviews (4): Last reviewed commit: "[T3348] FIX: keep the auth headers when ..." | Re-trigger Greptile}