Problem

test_schema_disagg_abort hangs intermittently on longer runs.

Root cause

The leader child runs 12 schema-worker threads and 1 checkpoint thread, coordinated by state->lock, a pthread_rwlock initialized with default attributes (workload.c:363):

pthread_rwlock_init(&state.lock, NULL);   // NULL attr => glibc default = PREFER_READER

The 12 schema workers hold the read lock in a tight loop (create/drop + publish, workload.c:181, 197). The single checkpoint thread needs the write lock (workload.c:270). glibc's default rwlock is reader-preferring: new readers cut ahead of a waiting writer, so with 12 continuous readers the write lock never becomes available.

Chain to the observed symptom:

1. Checkpoint thread starves in pthread_rwlock_wrlock – first checkpoint never completes.
2. READY_FILE sentinel never written (workload.c:283).
3. Parent busy-waits forever in while (!testutil_exists(...READY_FILE)) (main.c:119).
4. Log stops right after the CONFIG: line. Hang.

SIGCHLD does not rescue it – the child is alive (starving), not exiting.

Amplifier

Every layered create/drop calls __wt_meta_track_off(need_sync=true) – a synchronous metadata checkpoint + fdatasync – while holding the read lock. This stretches hold times so the lock is essentially always held once all 12 threads warm up, making starvation near-certain. This explains the intermittency: if the checkpoint grabs the write lock before the workers saturate it succeeds; otherwise it starves permanently.