Lager (as in the beer) is a logging framework for Erlang. Its purpose is to provide a more traditional way to perform logging in an erlang application that plays nicely with traditional UNIX logging tools like logrotate and syslog. You can view the Lager EDocs here.
Travis-CI ::
- Finer grained log levels (debug, info, notice, warning, error, critical, alert, emergency)
- Logger calls are transformed using a parse transform to allow capturing Module/Function/Line/Pid information
- When no handler is consuming a log level (eg. debug) no event is even sent to the log handler
- Supports multiple backends, including console and file. More are planned.
To use lager in your application, you need to define it as a rebar dep or have some other way of including it in erlang’s path. You can then add the following option to the erlang compiler flags
{parse_transform, lager_transform}
Alternately, you can add it to the module you wish to compile with logging enabled:
-compile([{parse_transform, lager_transform}]).
Before logging any messages, you’ll need to start the lager application. The lager module’s start function takes care of loading and starting any dependencies lager requires.
lager:start().
Once you have built your code with lager and started the lager application, you can then generate log messages by doing the following:
lager:error("Some message")
Or:
lager:warning("Some message with a term: ~p", [Term])
The general form is lager:Severity() where Severity is one of the log levels mentioned above.
To configure lager’s backends, you use an application variable (probably in your app.config):
{lager, [ {handlers, [ {lager_console_backend, info}, {lager_file_backend, [ {"error.log", error, 10485760, "$D0", 5}, {"console.log", info, 10485760, "$D0", 5} ]} ]} ]}.
The available configuration options for each backend are listed in their module’s documentation.
All loggers have a default formatting that can be overriden. A formatter is any module that exports format(#lager_log_message{},Config#any()). It is specified as part of the configuration for the backend:
{lager, [ {handlers, [ {lager_console_backend, [info, {lager_default_formatter, [time," [",severity,"] ", message, "\n"]}}, {lager_file_backend, [ [{"error.log", error, 10485760, "$D0", 5,{lager_default_formatter,[date, " ", time," [",severity,"] ",pid, " ", message, "\n"]}], {"console.log", info, 10485760, "$D0", 5} ]} ]} ]}.
Included is lager_default_formatter. This provides a generic, default formatting for log messages using a “semi-iolist” as configuration. Any iolist allowed elements in the configuration are printed verbatim. Atoms in the configuration are treated as metadata properties and extracted from the log message. The metadata properties date,time, message, and severity will always exist. The properties pid, file, line, module, function, and node will always exist if the parser transform is used.
["Foo"] -> "Foo", regardless of message content. [message] -> The content of the logged message, alone. [{pid,"Unknown Pid"}] -> "<?.?.?>" if pid is in the metadata, "Unknown Pid" if not. [{pid, ["My pid is ", pid], "Unknown Pid"}] -> if pid is in the metadata print "My pid is <?.?.?>", otherwise print "Unknown Pid"
Optionally, a tuple of {atom(),semi-iolist()} can be used. The atom will look up the property, but if not found it will use the semi-iolist() instead. These fallbacks can be nested or refer to other properties.
[{pid,"Unknown Pid"}] -> "<?.?.?>" if pid is in the metadata, "Unknown Pid" if not. [{server,[$(,{pid,"Unknown Server"},$)]}}] -> user provided server metadata, otherwise "(<?.?.?>)", otherwise "(Unknown Server)"
Lager is also supplied with a error_logger handler module that translates traditional erlang error messages into a friendlier format and sends them into lager itself to be treated like a regular lager log call. To disable this, set the lager application variable `error_logger_redirect’ to `false’.
The error_logger handler will also log more complete error messages (protected with use of trunc_io) to a “crash log” which can be referred to for further information. The location of the crash log can be specified by the crash_log application variable. If undefined it is not written at all.
Messages in the crash log are subject to a maximum message size which can be specified via the crash_log_msg_size application variable.
You can change the log level of any lager backend at runtime by doing the following:
lager:set_loglevel(lager_console_backend, debug).
Or, for the backend with multiple handles (files, mainly):
lager:set_loglevel(lager_file_backend, "console.log", debug).
Lager keeps track of the minium log level being used by any backend and supresses generation of messages lower than that level. This means that debug log messages, when no backend is consuming debug messages, are effectively free. A simple benchmark of doing 1 million debug log messages while the minimum threshold was above that takes less than half a second.
Lager can rotate its own logs or have it done via an external process. To use internal rotation, use the last 3 values in the file backend’s configuration tuple. For example
{"error.log", error, 10485760, "$D0", 5}
This tells lager to log error and above messages to “error.log” and to rotate the file at midnight or when it reaches 10mb, whichever comes first and to keep 5 rotated logs, in addition to the current one. Setting the count to 0 does not disable rotation, it instead rotates the file and keeps no previous versions around. To disable rotation set the size to 0 and the date to “”.
The “$D0” syntax is taken from the syntax newsyslog uses in newsyslog.conf. The relevant extract follows:
Day, week and month time format: The lead-in character for day, week and month specification is a `$'-sign. The particular format of day, week and month specification is: [Dhh], [Ww[Dhh]] and [Mdd[Dhh]], respectively. Optional time fields default to midnight. The ranges for day and hour specifications are: hh hours, range 0 ... 23 w day of week, range 0 ... 6, 0 = Sunday dd day of month, range 1 ... 31, or the letter L or l to specify the last day of the month. Some examples: $D0 rotate every night at midnight $D23 rotate every day at 23:00 hr $W0D23 rotate every week on Sunday at 23:00 hr $W5D16 rotate every week on Friday at 16:00 hr $M1D0 rotate on the first day of every month at midnight (i.e., the start of the day) $M5D6 rotate on every 5th day of the month at 6:00 hr
To configure the crash log rotation, the following application variables are used:
- crash_log_size
- crash_log_date
- crash_log_count
See the .app.src file for further details.
Lager syslog output is provided as a separate application; lager_syslog. It is packaged as a separate application so Lager itself doesn’t have an indirect dependancy on a port driver. Please see the lager_syslog README for configuration information.
Jon Brisbin has written a lager backend to send lager messages into AMQP, so you can aggregate logs from a cluster into a central point. You can find it under the lager_amqp_backend project on github.
The team at KIVRA has written a lager backend to send lager messages into Loggly. You can find it under the lager_loggly project on github.
Ivan Blinkov has written a lager backend to send lager messages to email via SMTP, so you can get informed about critical errors in production immediately. You can find it under the lager_smtp project on github.
Lager supports basic support for redirecting log messages based on log message attributes. Lager automatically captures the pid, module, function and line at the log message callsite. However, you can add any additional attributes you wish:
lager:warning([{request, RequestID},{vhost, Vhost}], "Permission denied to ~s", [User])
Then, in addition to the default trace attributes, you’ll be able to trace based on request or vhost:
lager:trace_file("logs/example.com.error", [{vhost, "example.com"}], error)
You can also omit the final argument, and the loglevel will default to ‘debug’.
Tracing to the console is similar:
lager:trace_console([{request, 117}])
In the above example, the loglevel is omitted, but it can be specified as the second argument if desired.
You can also specify multiple expressions in a filter, or use the ‘*’ atom as a wildcard to match any message that has that attribute, regardless of its value.
Tracing to an existing logfile is also supported, if you wanted to log warnings from a particular module to the default error.log:
lager:trace_file("log/error.log", [{module, mymodule}], warning)
To view the active log backends and traces, you can use the lager:status() function. To clear all active traces, you can use lager:clear_all_traces().
To delete a specific trace, store a handle for the trace when you create it, that you later pass to lager:stop_trace/1:
{ok, Trace} = lager:trace_file("log/error.log", [{module, mymodule}]), ... lager:stop_trace(Trace)
Tracing to a pid is somewhat of a special case, since a pid is not a data-type that serializes well. To trace by pid, use the pid as a string:
lager:trace_console([{pid, "<0.410.0>"}])
Lager defaults to truncating messages at 4096 bytes, you can alter this by using the {lager_truncation_size, X} option. In rebar, you can add it to erl_opts:
{erl_opts, [{parse_transform, lager_transform}, {lager_truncation_size, 1024}]}.
You can also pass it to erlc, if you prefer:
erlc -pa lager/ebin +'{parse_transform, lager_transform}' +'{lager_truncation_size, 1024}' file.erl