A DROP only clears once the stable timestamp covers the table's newest insert. The test pins stable to the minimum completed timestamp over all workers, so a worker with nothing left to apply freezes stable for everyone and the drop can never clear. Checkpoints keep completing throughout the 30 second wait and none of them can help, because a use_timestamp=true checkpoint cannot clean a tree whose data is above stable.

Reproduces deterministically with -r l -t 10 -T 2 -u 4 -PSD13570525,E6333586: the stuck worker needed 179 and had reached 180, another worker sat at 177 with an empty queue, and every timestamp the run allocated had in fact been applied.

Fix: track the frontier exactly - the highest timestamp with nothing unfinished at or below it - instead of approximating it with the per-worker minimum. A full event queue is one way to starve a worker, not the only one, and needs no separate fix.