@OvermindDL1 great idea, I have updated the wiki!
swelham
AstonJ
swelham
Eiji
I got an interesting idea!
How about catch messages, save them to database and files?
The table structure:
- id
Integer - env
Atom - type
Atom - source
Atom - file_info
Map - caller_info
Map - stacktrace
ListofMaps - log_fiile_path
String - inserted_at - datae and time
- updated_at - datae and time
File default structure:
/project_directory
/logs
/env
/type
timeline.log
/descriptions
id.log
Where:
idis a database id.envis our environmenttypeis one of:[:error, :warning, :debug]- of course in production configuration developer could set only:errorreportingsourceis an info about from what it comes, for example::otpmessage or:ectomessage or:loggercallfile_infofor example:%{module: "MyProject.MyModule", function: %{name: "my_function", callback: "&my_function/2"}, file_path: "/lib/my_project/my_module.ex", line: "53"}caller_infofor example:%{module: "IO", function: %{name: "puts", callback: "&puts/1", args: "OK"}or:%{module: "Kernel", function: "raise", data: %{message: "Argument first is not an integer", type: "ArgumentError"}}stacktracefor example:[%{from: "stdlib", file: "lists.erl", line: 1338:, callback: "lists.foreach/2"}]log_fiile_pathis the path generated using config (for the directory structure)inserted_atis date of first time found message (error, warning, message) - sometimes useful if we know what update applied in specified day and timeupdated_atis last date of found message
Example format of timeline.log:
At
date_and_timefound #idtype. Details stored in: “logs/env/type/descriptions/23.log”.
Example format of id.log:
At: “2016-01-01 00:00:00.0000 UTC”
First reported at: “2015-12-12 00:00:00.0000 UTC”
In environment: “production”
Found #23 error
That was called from module: “MyProject.MyModule”, function “&my_function/2”, in file: “/lib/my_project/my_module.ex” at line: 23.
Caller is: module “IO” and function: “&puts/1” with data:
[
“OK”
]
Stacktrace:
(stdlib) lists.erl:1338: :lists.foreach/2
…
Logs descriptions could be optionally to catch them from logs based program like fail2ban.
With following structures we can easily format that message for example for special markup.
This file can be send by mail to service account or automatically add (if inserted) a new bug to for example Github (by public HTTP API) from production server, so it’s possible to get fast help directly after user gets “500 Internal Server Error” message without fill a bug report.
Phone call:
John: “Hey, your site is excellent, but one page is not working. Can you see what’s wrong?”
Support: “Please reload page - it’s already solved :-).”
John: “Wow! Thank you very much!”
Is is possible to catch all Elixir messages (runtime errors, deprecation warnings, IO operations and more) like I described?
What do you think about it?
2
OvermindDL1
You can create your own error logger to plug into Elixir Log, just like the file_logger plugin, but log to a DB. Of course you’d want to throttle it, cache recent ones and prevent duplicates and such or you can get overwhelmed with messages if something bad suddenly happens. 
A library for that would be cool though…
2
swelham
OvermindDL1
Well the others follow logger_file_backend and logger_json_file_backend and logger_papertrail_backend and logger_lager_backend and many others that follow logger_*_backend, so probably logger_ecto_backend or so? Or something more generic like logger_cache_callback_backend if it handles the deduplication and caching and such for you, just passed to a callback to then do whatever you want, maybe even to another log channel so something like logger_ecto_backend or logger_file_backend or so could log them as normal? Although maybe that would not be a backend but a filter or so? ^.^
1
swelham
I agree something more generic like a logger_filter which can handle the caching and deduplication and then have plugable backends such as logger_ecto_backend, logger_file_backend, etc…
This would allow it to be flexible and easily extendable.
2
Eiji
If I good understand how Logger works it will give me only data from: Logger.* methods like: Logger.info, right?
What about catching:
raise ArgumentError, message: "Bad argument!"
and deprecated warnings?
Of course I mean to only catch and log that situations (not rescue).
1