Add debuglevel JSON-RPC

[jrick]

The debuglevel method allows the debug level of all loggers, or only a particular subsystem's logger, to be modified to a specified level. It is currently unimplemented by dcrwallet, so dcrwallet instead tries to perform RPC passthrough to dcrd.

The implementation can't be simply a copy of dcrd's due to dcrwallet's more modular structure. In dcrd, the JSON-RPC server is implemented as part of the main package, while in dcrwallet it is implemented in the legacyrpc package. Therefore, in order to provide access to the logger variables (in the main package), a mechanism similar to the RequestProcessShutdown server method should be used to indicate the request to the main package. In this case, I feel it makes more sense to invoke a callback set by main rather than using a channel.

[ggoranov]

I'm busy with this issue.

[ggoranov]

The proposal with the RequestProcessShutdown mechanism works pretty well. I'm also considering to split parseAndSetDebugLevels into parseDebugLevels and setDebugLevels so that parseDebugLevels can be reused on returning meaningful error when sending invalid pair subsystem_id - debuglevel.

E.g

dcrctl --wallet debuglevel "DCRW1=debug"
The specified subsystem [DCRW1] is invalid -- supported subsytems [CMGR DCRW GRPC LODR RPCS SYNC TKBY WLLT]

[jrick]

sounds reasonable

[ggoranov]

As parseAndSetDebugLevels and subsystemLoggers are part of main package it would have needed too many code changes to make it possible for dcrctl to return full error description.
With the last commit, we show error in Stderr.

[itswisdomagain]

Is this done? I'd like to take it up.

[jrick]

there is very likely little to be gained by being able to configure the verbosity levels anymore. All important messages are (or should be) always logged, and the Debug and Trace levels are very infrequently used, and where they are, only in very old code. We have also begun using runtime/trace for trace messages, which provides context-scoped logs. These have proved to be far more valuable than subsystem-level control, as they are easier to search through and grok compared to the unreasonable verbosity that you'd otherwise get with our standard logger.

We can probably skip supporting this but I'm open to feedback on that decision.

[jrick]

Closing per previous comment.

[jrick]

Reopening with an implementation here #2449.

There's lots of useful information currently hidden behind MIXP and MIXC debug levels, and it restarting to turn these on or off is less than ideal.