Some LD_BUG logs don't come with any warn/err logs, leading to silent metrics port bursts of 2000+ bug counts

I've been working with @toralf who reported surprising counts on the METRICS_NAME(bug_reached_count) metrics port counter:

<toralf> and I observed 2000 incremnents within 3 min in the past [...]

They were a mystery because the warn-level logs remained empty.

I looked into it some more, and it looks like this counter increments every time there is a log entry of domain LD_BUG.

For the most part we only log with LD_BUG at severity warn or err, but there are some exceptions, e.g.

• conflux_log_set() uses LD_BUG but here gets told to use LOG_INFO by default:

     log_fn(LOG_PROTOCOL_WARN, LD_CIRC,
            "Conflux set has too many legs to link. "
            "Rejecting this circuit.");
     conflux_log_set(LOG_PROTOCOL_WARN, unlinked->cfx, unlinked->is_client);

• In or/circuitstats.c we have some lines of the form log_debug(LD_BUG,

• On the client side,

       if (endreason != END_STREAM_REASON_RESOLVEFAILED) {
         log_info(LD_BUG,
                  "No origin circuit for successful SOCKS stream %"PRIu64

So... there appear to be quite a few edge cases like this.

I guess the hope was that we have an invariant where you only use LD_BUG when you are also (loudly) logging details about a real bug?

The robust simple fix would be to no longer increment the counter for LD_BUG.

The more pervasive and more brittle fix would be to audit all uses of LD_BUG and make sure we keep to our invariant.

Calling @dgoulet's name since maybe he can help us with what the original intent was for this metrics port counter.