Asynchronous log configuration problem log4j2 official documents and clarifications

Configuration and demo

A fully open method

Add startup parameters -DLog4jContextSelector = org.apache.logging.log4j.core.async.AsyncLoggerContextSelector
start the key parameters of a little saying:
the official documentation http://logging.apache.org/log4j/2.x/manual/async # AllAsync .html Key is -Dlog4j2.contextSelector = ...
but if you personally do experiments, or personally used, you may notice the problem, simply use with your version of:
-DLog4jContextSelector = org.apache.logging .log4j.core.async.AsyncLoggerContextSelector // == == full version supports
version supports -Dlog4j2.contextSelector = org.apache.logging.log4j.core.async.AsyncLoggerContextSelector // == 2.10.0 == later

Related code (2.10.0 later):
See org.apache.logging.log4j.util.PropertiesPropertySource and
org.apache.logging.log4j.util.PropertiesUtil.Environment
Environment There are three sets of configuration data, a different key, the same service semantics:

literal
{java.runtime.name=Java(TM) SE Runtime Environment}


normalized
{log4j2.javaRuntimeName=Java(TM) SE Runtime Environment} key 是log4j2.前缀 加上驼峰形式变量

tokenized
{[java, runtime, name]=Java(TM) SE Runtime Environment}  key是list

The method of mixing two synchronous asynchronous

<!-- 使用异步日志时，启用如下配置，异步配置默认不打印location。此种方法无需配置JVM启动参数。 -->
<asyncRoot level="debug">
    <AppenderRef ref="Console" />
</asyncRoot>

Some notes

The default location information is not printed, such as line numbers, classes, methods and the like. We need to add if needed includeLocation="true", but according to the official statement will be 30-100 times slower.
Asynchronous time, if the print log in a real field value of the object referenced in the log information changed in such "hello {}", user.name this. It will not affect the accuracy of the print log. The answer is no, because log4j made a snapshot mechanism.

Malpractice asynchronous log

Asynchronous log not only the advantages of no drawbacks, the official gives some scenes (Drawbacks) not used:

Exception handling. Even with exception handler can not cover all cases.
In some uses, such as MapMessage and StructuredDataMessage this problem by not logging time snapshot avoid asynchronous print information may change
CPU in a few circumstances, such as to a CPU
Print log much faster than appenderthroughput. At this time, asynchronous and can not solve the problem, need to use a faster appender.