LogstashLayout is the fastest Log4j 2 JSON layout allowing schema customization and Logstash-friendly output.

By default, LogstashLayout ships the official JSONEventLayoutV1 stated by log4j-jsonevent-layout Log4j 1.x plugin. Compared to JSONLayout included in Log4j 2 and log4j-jsonevent-layout, LogstashLayout provides the following additional features:

• Superior performance
• Customizable JSON schema (see eventTemplate[Uri] and stackTraceElementTemplate[Uri] parameters)
• Customizable timestamp formatting (see dateTimeFormatPattern and timeZoneId parameters)

• Usage
• FAT JAR
• Appender Support
• Performance
• Contributors
• License

Add the log4j2-logstash-layout dependency to your POM file

<dependency>
    <groupId>com.vlkan.log4j2</groupId>
    <artifactId>log4j2-logstash-layout</artifactId>
    <version>${log4j2-logstash-layout.version}</version>
</dependency>

together with a valid log4j-core dependency:

<dependency>
    <groupId>org.apache.logging.log4j</groupId>
    <artifactId>log4j-core</artifactId>
    <version>${log4j2.version}</version>
</dependency>

Below you can find a sample log4j2.xml snippet employing LogstashLayout.

<?xml version="1.0" encoding="UTF-8"?>
<Configuration status="warn">
    <Appenders>
        <Console name="CONSOLE" target="SYSTEM_OUT">
            <LogstashLayout dateTimeFormatPattern="yyyy-MM-dd'T'HH:mm:ss.SSSZZZ"
                            eventTemplateUri="classpath:LogstashJsonEventLayoutV1.json"
                            prettyPrintEnabled="true"
                            stackTraceEnabled="true"/>
        </Console>
    </Appenders>
    <Loggers>
        <Root level="info">
            <AppenderRef ref="CONSOLE"/>
        </Root>
    </Loggers>
</Configuration>

Or using the log4j2.properties file instead:

status = warn

appender.console.name = CONSOLE
appender.console.type = CONSOLE
appender.console.target = SYSTEM_OUT

appender.console.logstash.type = LogstashLayout
appender.console.logstash.dateTimeFormatPattern = yyyy-MM-dd'T'HH:mm:ss.SSSZZZ
appender.console.logstash.eventTemplateUri = classpath:LogstashJsonEventLayoutV1.json
appender.console.logstash.prettyPrintEnabled = true
appender.console.logstash.stackTraceEnabled = true

rootLogger.level = info
rootLogger.appenderRef.stdout.ref = CONSOLE

This generates an output as follows:

{
  "exception": {
    "exception_class": "java.lang.RuntimeException",
    "exception_message": "test",
    "stacktrace": "java.lang.RuntimeException: test\n\tat com.vlkan.log4j2.logstash.layout.demo.LogstashLayoutDemo.main(LogstashLayoutDemo.java:11)\n"
  },
  "line_number": 12,
  "class": "com.vlkan.log4j2.logstash.layout.demo.LogstashLayoutDemo",
  "@version": 1,
  "source_host": "varlik",
  "message": "Hello, error!",
  "thread_name": "main",
  "@timestamp": "2017-05-25T19:56:23.370+02:00",
  "level": "ERROR",
  "file": "LogstashLayoutDemo.java",
  "method": "main",
  "logger_name": "com.vlkan.log4j2.logstash.layout.demo.LogstashLayoutDemo"
}

LogstashLayout is configured with the following parameters:

^† Note that string value truncation via maxStringLength can take place both in object keys and values, and this operation does not leave any trace behind. maxStringLength is intended as a soft protection against bogus input and one should always rely on maxByteCount for a hard limit.

eventTemplateUri denotes the URI pointing to the JSON template that will be used while formatting the LogEvents. By default, LogstashLayout ships LogstashJsonEventLayoutV1.json providing the official Logstash JSONEventLayoutV1.

{
  "mdc": "${json:mdc}",
  "ndc": "${json:ndc}",
  "exception": {
    "exception_class": "${json:exception:className}",
    "exception_message": "${json:exception:message}",
    "stacktrace": "${json:exception:stackTrace:text}"
  },
  "line_number": "${json:source:lineNumber}",
  "class": "${json:source:className}",
  "@version": 1,
  "source_host": "${hostName}",
  "message": "${json:message}",
  "thread_name": "${json:thread:name}",
  "@timestamp": "${json:timestamp}",
  "level": "${json:level}",
  "file": "${json:source:fileName}",
  "method": "${json:source:methodName}",
  "logger_name": "${json:logger:name}"
}

Similarly, stackTraceElementUri denotes the URI pointing to the JSON template that will be used while formatting the StackTraceElements. By default, LogstashLayout ships classpath:Log4j2StackTraceElementLayout.json providing an identical stack trace structure produced by Log4j 2 JSONLayout.

{
  "class": "${json:stackTraceElement:className}",
  "method": "${json:stackTraceElement:methodName}",
  "file": "${json:stackTraceElement:fileName}",
  "line": "${json:stackTraceElement:lineNumber}"
}

In case of need, you can create your own templates with a structure tailored to your needs. That is, you can add new fields, remove or rename existing ones, change the structure, etc. Please note that eventTemplateUri parameter only supports file and classpath URI schemes.

Below is the list of allowed LogEvent template variables that will be replaced while rendering the JSON output.

JSON field lookups are performed using the ${json:<variable-name>} scheme where <variable-name> is defined as <resolver-name>[:<resolver-key>]. Characters following colon (:) are treated as the resolver-key.

Log4j 2 Lookups (e.g., ${java:version}, ${env:USER}, ${date:MM-dd-yyyy}) are supported in templates too. Though note that while ${json:...} template variables are expected to occupy an entire field, that is, "level": "${json:level}", a lookup can be mixed within a regular text: "myCustomField": "Hello, ${env:USER}!".

Similarly, below is the list of allowed StackTraceElement template variables:

As in LogEvent templates, StackTraceElement templates support Log4j 2 lookups too.

See layout-demo directory for a sample application demonstrating the usage of LogstashLayout.

Project also contains a log4j2-logstash-layout-fatjar artifact which includes all its transitive dependencies in a separate shaded package (to avoid the JAR Hell) with the exception of log4j-core, that you need to include separately.

This might come handy if you want to use this plugin along with already compiled applications, e.g., Elasticsearch 5.x and 6.x versions, which requires Log4j 2.

log4j2-logstash-layout is all about providing a highly customizable JSON schema for your logs. Though this does not necessarily mean that all of its features are expected to be supported by every appender in the market. For instance, while prettyPrintEnabled=true works fine with log4j2-redis-appender, it should be turned off for Logstash's log4j-json file input type. (See Pretty printing in Logstash issue.) Make sure you configure log4j2-logstash-layout properly in a way that is aligned with your appender of preference.

The source code ships a LogstashLayout-vs-JSONLayout (the one shipped by default in Log4j 2) JMH benchmark assessing the rendering performance of both plugins. There two different LogEvent profiles are employed:

• full: LogEvent contains MDC, NDC, and an exception.
• lite: LogEvent has no MDC, NDC, or exception attachment.

To give an idea, we ran the benchmark with the following settings:

• CPU: Intel i7 2.70GHz (x86-64, confined java process to a single core using taskset -c 0)
• JVM: Java HotSpot 1.8.0_161 (-XX:+TieredCompilation, -XX:+AggressiveOpts)
• OS: Xubuntu 18.04.1 (4.15.0-34-generic, x86-64)
• LogstashLayout: used default settings with the following exceptions:
  • stackTraceEnabled: true
• JSONLayout: used in two different flavors
  • DefaultJsonLayout: default settings
  • CustomJsonLayout: default settings with an additional "@version": 1 field (this forces instantiation of a wrapper class to obtain the necessary Jackson view)

The results are as follows. (See layout-benchmark directory for the full report.)

Let us try to answer some common questions:

• How come log4j2-logstash-layout can yield superior performance compared to Log4j 2 JSONLayout? Log4j 2 JSONLayout employs a single Jackson view to generate JSON, XML, and YAML outputs. For this purpose, it uses Jackson ObjectMapper, which needs to walk over the class fields via reflection and perform heavy branching and intermediate object instantiation. On the contrary, log4j2-logstash-layout parses the given template once and compiles an (almost) garbage- and (to a certain extent) branching-free JSON generator employing Jackson JsonGenerator.

• Why is log4j2-logstash-layout is not totally garbage-free?

  • Reusing Jackson JsonGenerators necessitate state reset when it gets corrupted. Though that is easier said than done, prone to bugs, and does not bring much performance improvement. Hence we instantiate a new JsonGenerator for each serialization to be on the safe (but still fast) side.

  • Since Throwable#getStackTrace() clones the original StackTraceElement[], accesses to (and hence rendering) stack traces can never be garbage-free.

  • Rendering stack traces to text (that is, exception[rootCause]StackTrace:text) allocates a new ByteArrayOutputStream-backed PrintStream (used by Throwable#printStackTrace()) each time.

  • Rendering of context data (that is, MDC) field values is garbage-free if the value is either null, or of type String, Short, Integer, Long, or byte[].

• How can one run the benchmark on his/her machine? After a fresh mvn clean package within the source directory, run layout-benchmark/benchmark.py.

• bakomchik
• chrissydee
• Eric Schwartz
• justinsaliba
• Michael K. Edwards
• Mikael Strand
• Rafa Gómez
• Yaroslav Skopets

Copyright © 2017-2019 Volkan Yazıcı

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.