Ability to adjust log levels?

[osyoyu]

I'm working on a webapp which utilizes garage_client to communicate with another backend server, and would like to see garage_client requests in the logs.

GarageClient::Client#new takes the :verbose option, which is passed down to the underlying Faraday object, but its output is a bit noisy for the above usage.

So, my question is: could garage_client have an option to adjust the log level?

I'm thinking of something like aws-sdk-ruby, which has a configurable logger (e.g. AWS.config(:log_level => :debug)), which could even take a format string like this (docs):

pattern = '[REQUEST :http_status_code] :service :operation :duration'
formatter = AWS::Core::LogFormatter.new(pattern)

@cookpad/dev-infra What do you think?

[osyoyu]

Newer versions of Faraday::Response::Logger accept Logger instances (https://github.com/lostisland/faraday/blob/c80cf86c5a09be0f0cdc3dd00d1c85390fe55d5b/lib/faraday/response/logger.rb#L9). Maybe this issue could be partially solved by just allowing GarageClient::Client#new to accept logger instances (which have logger levels set), and just pass it down to Faraday::Response::Logger (which is already used in verbose mode).

---

Printing bodies of POST requests could also be useful, and Faraday::Response::Logger provides that when {bodies: true} is passed to its options. Although, accepting that option hash through GarageClient::Client#new would be bit complex, and destroy encapsulation. No good ideas for that in the current implementation.

[taiki45]

GarageClient::Client#new takes the :verbose option, which is passed down to the underlying Faraday object, but its output is a bit noisy for the above usage.

Could you give us more detail of the trouble point? I'm not sure why that verbose option doesn't work for your requirement.

[osyoyu]

The verbose option prints something like this:

[INFO] get https://example.com/v1/users?fields=id
[DEBUG] request: Accept: "application/json"
User-Agent: "garage_client 2.4.2 myapp"
[INFO] Status: 200
[DEBUG] response: date: "Mon, 28 Jan 2019 23:40:27 GMT"
connection: "close"
server: "h2o/2.0.6"
content-type: "application/json; charset=utf-8"
vary: "Accept-Encoding"
status: "200 OK"
link: "</v1/users?fields=id&page=2&per_page=20>; rel=\"next\"; page=\"2\""
cache-control: "public, max-age=0, s-maxage=600"
x-unicorn-pid: "29900"
x-amzn-trace-id: "Root=1-xxxxxxx-xxxxxxxxxx;Sampled=1"
x-chst: "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
etag: "W/\"xxxxxxxxxxxxxxxxxxxxxxxxxxx\""
x-request-id: "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
x-runtime: "0.068630"
transfer-encoding: "chunked"

While this information is "verbose", we don't need all of this in our logs; so, I'd like to have control over this.

We have patched GarageClient in our application to make it show compact logs like this:

[GarageClient 200] POST https://example.com/v1/users

[taiki45]

Got it, I also think current "verbose option" is too much for most of usec ases, so cloud we just change current GarageClient's "verbose option" behavior instead of adding more extensible API above? I don't think current "verbose option" is not good default behavior, so we can change this (IMO).

If we face additional more verbose requirement later, we can add that with extensible API. I think default behaviors should be simple and fit to most of use cases or basic requirements.