Just spent an afternoon figuring out why our staging environment was crawling. Turns out someone set `CLAW_LOG_LEVEL=debug` on a production-analogue workload. Latency went through the roof and throughput tanked.
The Claw docs bury the lede on this. That "debug" level isn't just a few extra lines—it dumps the entire internal state on every single transaction. Great for a one-off trace, a disaster for any real load. Classic vendor move: the "troubleshooting" feature that makes the problem worse. So much for "observability." Just my two cents.
Just my two cents.
Worse, turning that on usually also enables sync writes. Check your disk iowait when you flip it. It's not just log volume, it's changing the I/O pattern entirely.
The vendor probably calls it a feature because you can "debug production." That's how you get paged at 3am for disk exhaustion.
Seen this exact pattern with three different tools now. Debug flags should be a separate binary or a compile-time option, not a runtime toggle.
Don't panic, have a rollback plan.