Logging framework for OSv

[tzach]

Logging framework should have the following properties:

• Support local (file) and/or remote logging
• Can write to more than one output channels at the same time
• Easy to add underline protocols (e.g. ZeroMQ, REST, Amazon Kinesis, DB, more) as plug-ins
• Low footprint
• High throughput (latency not important)
• Log levels: fatal, error, warning, info, debug, trace (from Log4J)
• Log level can be update at run-time (via API), for each output channel
• Log sources can be update on run time (e.g. trun on debug level for http server only)
• Will not fail the main process if fail (sever not available, server too slow, local disk full etc)
• Log can be encrypted

Each logs line should include:

• Timestamp
• Log-level
• Log type (origin code)
• Channel: either STDOUT or STDERR
• Message

[raphaelcohn]

Please use syslog - both the POSIX API and the network protocol. This does all that you seek*; one can then use a second OSv instance as an aggregator, consumer, distributor, plug in, converter or whatever else... or not at all, end go direct to loggly, say. That's OSv's great strength - it allows the complete disaggregation of monolithic infrastructure into tiny, co-operating microservices, each of which can have a different purpose, upgrade path, version of software, and so on... allowing A+B testing, staggered deployment, partial rollback and lots of other good strategies besides.

• All these features are then not in core OSv, but could be done trivially (eg using rsyslogd).

[wkozaczuk]

Given limited resources we have I suggest we close it.