[FEA] Investigate C++ Logging libraries

[jrhemstad]

Is your feature request related to a problem? Please describe.

It would be nice to use a dedicated logging library for RMM logging.

Describe the solution you'd like

Investigate C++ logging library solutions.

[harrism]

Some links:

Some analysis:
https://kjellkod.wordpress.com/2015/06/30/the-worlds-fastest-logger-vs-g3log/

g3log: https://github.com/KjellKod/g3log
spdlog: https://github.com/gabime/spdlog
glog: https://github.com/google/glog

I think that we should discuss requirements / desires. Speed is important, and also simplicity, and very important is impact on installation of RMM (I think we would like RMM to become header-only). I think a logger that is header-only and mature, so we could vendor it directly into RMM rather than submoduling it.

I haven't thought about whether overall latency is important, as discussed in the first link above.

[jrhemstad]

I think a logger that is header-only and mature, so we could vendor it directly into RMM rather than submoduling it.

Agreed. I generated table to summarize these things

	Header only?	C++ Version	GitHub stars (maturity)	License	Notes
spdlog	Yes (static lib reccomended)	C++11	7800	MIT	More features than we really need.
g3log	No	C++14	550	Public Domain
glog	No	?	3600
loguru	Yes	C++11	831	Public Domain
plog	Yes	C++03	819	Mozilla Public 2.0	Very simple

Some requirements I've been thinking about:

• Simple
• Header-only
• Output to CSV
• Info we care about logging:
  • Allloc/Free + Size
  • Timestamp
  • Pool state (total free, in used)
  • Call stack?

[harrism]

Info: also # active allocations and allocation strategy-specific data like fragmentation metrics could come in handy.

[jakirkham]

I believe Glog is BSD-3-Clause. It included some third-party BSD-3-Clause code as well.

As to minimum C++ version it may C++03 or C++98, but wasn't sure either. Raised issue ( google/glog#527 ) to see if they could provide some guidance 🙂

[teju85]

I've seen some logging implementations need access to argc and argv for initializing the underlying "state" of the logger. IMO, that's particularly not great for libraries, and also especially when trying to access those libraries via python layer. Probably, this would also be something that we'd need to consider.

[jrhemstad]

I've seen some logging implementations need access to argc and argv for initializing the underlying "state" of the logger. IMO, that's particularly not great for libraries, and also especially when trying to access those libraries via python layer. Probably, this would also be something that we'd need to consider.

Yikes, yeah, that would definitely disqualify any logging library that required something like this.

[harrism]

spdlog chosen and implemented in #297. Closing.