Unbounded disk growth on fullnodes since v1.76.0

With rpc.enable-indexing: true and num-epochs-to-retain: 2, the pruner maxes out at ~2-3 checkpoints/s while the chain produces ~4.5/s. The watermark falls behind permanently and store/perpetual grows ~8 GB/day per chain. We see this behavior across all of our nodes (mainnet + testnet, v1.76.1 and v1.78.0).

The pruner thread is pegged at 100% of a single core, almost all of it in prune_objects_and_indexes.

Is this a known limitation, and is there any mitigation? Right now I'm "solving" the issue by expanding the disks every week, but that's not a sustainable solution.

Appreciate any help!