perf: avoid duplicate IA loan API call in get_loan

[lokesh]

Closes #

Performance fix (with a correctness fix included): removes a redundant blocking HTTP POST to the archive.org loan API from every edition-page loan check.

Estimated performance gain

~30–150 ms off server render time (p50–p90) for every anonymous edition-page loan check, and worst-case stalls halved from 2× to 1× the API timeout. The loan check is a synchronous cross-service POST on the render path; this PR halves the number of calls, so the saving is one full loan-API round-trip. Estimates assume typical same-datacenter latency to the IA loan service; measure with the existing stats timing on get_loan callers to confirm.

xychart-beta
    title "Blocking loan-API calls per anonymous loan check"
    x-axis ["Before", "After"]
    y-axis "sequential HTTP POSTs" 0 --> 2
    bar [2, 1]

Loading

For user-scoped lookups the second call still happens, but only when it can matter (linked @itemname and the first lookup found nothing) — previously it ran unconditionally.

Technical

Problem. get_loan() in openlibrary/core/lending.py made two sequential, unconditional calls to _get_ia_loan(), each a blocking requests.post to the IA loan API, and the second call unconditionally overwrote the first's result:

try:
    _loan = _get_ia_loan(identifier, account and userkey2userid(account.username))
except Exception: ...
try:
    _loan = _get_ia_loan(identifier, account and account.itemname)  # clobbers!
except Exception: ...

This had three concrete consequences:

1. Duplicate identical call for anonymous lookups. When user_key is None (the common case — get_edition_loans() in plugins/upstream/borrow.py calls get_loan(ocaid) with no user on edition-page renders, and is_loaned_out_on_ol() does the same), account is None, so both calls were byte-for-byte identical POSTs. Every loan-status check paid for two round-trips to the loan API where one suffices.
2. A found loan could be clobbered. If the first (ol:username) lookup found a loan and the second (itemname) lookup returned None, the function returned None — losing a real loan.
3. Unfiltered fallback for accounts without a linked IA item. When account.itemname is None, the second call was made with userid=None, i.e. an unfiltered query that returns any loan on the identifier — so a user-scoped get_loan(identifier, user_key) could return another patron's loan.

Fix. The second lookup now runs only when it can produce new, correctly-scoped information: if account and account.itemname and not _loan. Anonymous lookups make exactly one API call; user-scoped lookups fall back to the @itemname query only when the ol:username query found nothing.

Out of scope (pre-existing, noted for reviewers): the store-loan block just above (d = site.get().store.get("loan-" + identifier)) constructs a Loan(d) that is never returned when unexpired; that dead code predates this PR and is untouched.

Testing

New TestGetLoan in openlibrary/tests/core/test_lending.py (mocks lending.site and lending._get_ia_loan):

• anonymous lookup → exactly one API call, result propagated
• account lookup that finds a loan under ol:username → no second call, loan not clobbered
• account lookup that finds nothing → falls back to exactly one @itemname call
• account with no linked itemname → exactly one call

Reproduce: make test-py-uv PYTEST_ARGS="openlibrary/tests/core/test_lending.py" — 8 passed. pre-commit run --files openlibrary/core/lending.py openlibrary/tests/core/test_lending.py — all hooks pass (incl. mypy, ruff).

Screenshot

N/A — no UI change.

Stakeholders

Part of a performance-review series; see the review report branch claude/performance-review-top-changes-5wcs7g.

🤖 Generated with Claude Code

https://claude.ai/code/session_015ky5Sc19MGPkDgbgTWWWET