Problem

In __wt_btcur_skip_page, WiredTiger acquires a ref lock via WT_REF_LOCK to safely read aggregate timestamp information before deciding whether to skip an entire page. WT_REF_LOCK spins until it obtains the lock, which is wasteful because skip_page is a pure read-path optimization — if the lock is contended, the correct behavior is simply to not skip the page and let the cursor proceed normally.

Fix

Add a new WT_REF_TRYLOCK macro backed by _ref_try_lock(), a non-spinning CAS that returns EBUSY immediately if the ref is already locked. Use WT_REF_TRYLOCK in _wt_btcur_skip_page and return 0 (do not skip) when the lock is unavailable.

Validation

Submitted a sys-perf --alias perf-required patch (31 HVW tasks) against the MongoDB tree with this change vendored in. SPS Performance Comparison showed 0 regressions across 201 data points, with 2 minor throughput improvements (Z-score ≈ 2.0–2.1 on mixed_workloads_locust). No performance regression introduced.

Related support ticket: HELP-94648