Currently wikifunctions pods log ~2500 lines per hour in production (in one DC). The absolute majority of those logs are from "levelPath: trace/req" for the /metrics and /_info endpoints, which could probably suppressed. There are also a bunch of "levelPath: debug" logs logging memory and cpu usage, lacking any other identifier (like a request-id). Same goes for "levelPath: info" which seems to print the function call (plus arguments?) and some "levelPath: notice".
But especially the amount of trace/req logs probably made it hard to spot the problem that lead to T344998 right away, see https://logstash.wikimedia.org/goto/aa3497c585530d231916e0d2b479c026 vs. https://logstash.wikimedia.org/goto/d9f2a3f0cab9474b5d48bed41cf7b8c1