rm_stm returning invalid_lso as last stable offset after node restart

[graphcareful]

Version & Environment

Found on dev while testing locally

What went wrong?

After a restart I noticed that the rm_stm is returning 0 for max_collectible_offset. When i look closer i see this condition is getting hit.

    auto last_applied = last_applied_offset();
    if (unlikely(
          !_bootstrap_committed_offset
          || last_applied < _bootstrap_committed_offset.value())) {
        // To preserve the monotonicity of LSO from a client perspective,
        // we return this unknown offset marker that is translated to
        // an appropriate retry-able Kafka error code for clients.
        vlog(
          _ctx_log.info,
          "Returning invalid_lso: {}",
          _bootstrap_committed_offset.has_value());
        return model::invalid_lso;
    }

It seems like this condition only occurs if the stm has consumed to the true end before the node restart. In the above snippet bootstrap_committed_offset is null because apply() hasn't yet been called due to the fact that raft::state_machine_next is already at the end of the log because the stm sucessfully snapshotted the final offset before it crashed.

Therefore when the last_stable_offset is queried 0 is returned instead of the actual value.

What should happen?

The actual last_stable_offset the stm has processed should be returned instead of 0.

How to reproduce the issue?

Shutdown a node after the rm_stm is up to date, then query last_stable_offset.