Logging verbosity tax
What does raising process log level from warn to debug cost under forward_fast?
Numbers are same-host comparisons on a single reference host and are not service-level objectives. See the performance hub disclaimer.
When this matters
Other feature_tax cells run with logging.level: warn.
Operators sometimes raise verbosity while diagnosing production issues. This
study pairs a dedicated warn pole with the same fixture at debug under
forward_fast (metrics disabled in
both). Pipeline traces use the separate
tracing: block — see
Pipeline tracing tax.
What we varied
- Varied: logging level
(
logging_warnvslogging_debug) - Held fixed: metrics disabled, no dnstap sinks,
sync,forward_fast
Evidence
Feature tax — logging warn vs debug (forward_fast)
| Posture | Runtime | Achieved QPS | Avg latency (ms) | Sent | Completed | Lost | Workers |
|---|---|---|---|---|---|---|---|
| logging_warn | sync | 76457.0 | 26.1 | 767234 | 767234 | 0 | ingress=2 |
| logging_debug | sync | 64029.4 | 31.1 | 642890 | 642890 | 0 | ingress=2 |
At a glance
- logging warn vs debug (forward_fast):
logging_debugcosts about 16% QPS versuslogging_warn(~64k vs ~76k).
Takeaway
Debug logging carries a clear standing tax on this median. Debug costs about 16% QPS versus warn (~64k vs ~76k).
What to do: keep production at warn (or quieter). Use debug for diagnosis windows and remeasure on your hardware if you need a standing debug posture.