Summary
The --metrics flag only exposes a small subset of metrics unless --tracing is also turned on. In our production deployment, with metrics enabled, we only see:
rollup_boost_execution_mode
rollup_boost_rpc_blocks_created
The docs suggest that enabling --metrics should also expose span duration histograms such as:
fork_choice_updated_v3_duration
get_payload_v3_duration
new_payload_v3_duration
However, those histogram metrics appear to be gated behind --tracing, not --metrics (code ref).
Suggested fix
I think the cleanest fix would be to decouple Prometheus span metrics from OTLP tracing export.