WiredTiger has a number of prefetch statistics that are redundant and make diagnostics harder to interpret. Specifically, prefetch_skipped ("pre-fetch not triggered by page read") is incremented alongside a more specific stat at every call site, making the aggregate counter carry no additional information.

Context

In src/session/session_prefetch.c, every early-return path increments both a reason-specific stat and the generic prefetch_skipped:

if (F_ISSET(session, WT_SESSION_INTERNAL)) {
    WT_STAT_CONN_INCR(session, prefetch_skipped_internal_session);
    WT_STAT_CONN_INCR(session, prefetch_skipped);
    return (false);
}

if (F_ISSET(ref, WT_REF_FLAG_INTERNAL)) {
    WT_STAT_CONN_INCR(session, prefetch_skipped_internal_page);
    WT_STAT_CONN_INCR(session, prefetch_skipped);
    return (false);
}

if (F_ISSET(S2BT(session), WT_BTREE_SPECIAL_FLAGS) &&
  !F_ISSET(S2BT(session), WT_BTREE_VERIFY)) {
    WT_STAT_CONN_INCR(session, prefetch_skipped_special_handle);
    WT_STAT_CONN_INCR(session, prefetch_skipped);
    return (false);
}

if (session->pf.prefetch_disk_read_count < 2) {
    WT_STAT_CONN_INCR(session, prefetch_skipped_disk_read_count);
    WT_STAT_CONN_INCR(session, prefetch_skipped);
    return (false);
}

Because prefetch_skipped == sum of all per-reason skip stats, it provides no diagnostic value beyond what summing the specific counters would give. This pattern likely extends to other prefetch aggregate stats.

• Redundant stats make dashboards and diagnostic scripts harder to interpret.
• Operators and developers must cross-reference multiple counters to understand prefetch skip rates by cause.
• Removing or restructuring the duplicate counter would simplify analysis without loss of information.

Proposed Solution

1. Audit all prefetch stats in src/stat/stat_data.py and their call sites to identify which stats are always co-incremented with a more specific companion.
2. Determine whether prefetch_skipped (or similar aggregate stats) adds value as a roll-up, or whether it can be removed in favour of summing the per-reason counters in tooling/monitoring.
3. Propose and implement a cleaner stat design: either remove the redundant aggregate, rename stats for clarity, or add documentation strings that make the stat relationships explicit.
4. Update any downstream tooling or dashboards that rely on removed/renamed stats.

Definition of Done

• All redundant prefetch stats are either removed or documented with clear semantics.
• No stat is silently co-incremented with another without a clear reason.
• The change is reflected in stat_data.py and all relevant call sites.