micrometer-jvm-extras

A set of additional JVM process metrics for micrometer.io.

Motivation

• get "real" memory usage of the JVM beyond its managed parts
• get ahold of that info from within the JVM in environments where you can't instrument from the outside (e.g. PaaS)

Usage

<dependency>
    <groupId>io.github.mweirauch</groupId>
    <artifactId>micrometer-jvm-extras</artifactId>
    <version>x.y.z</version>
</dependency>

    /* Plain Java */
    final MeterRegistry registry = new SimpleMeterRegistry();
    new ProcessMemoryMetrics().bindTo(registry);
    new ProcessThreadMetrics().bindTo(registry);

    /* With Spring */
    @Bean
    public MeterBinder processMemoryMetrics() {
        return new ProcessMemoryMetrics();
    }

    @Bean
    public MeterBinder processThreadMetrics() {
        return new ProcessThreadMetrics();
    }

Available Metrics

ProcessMemoryMetrics

ProcessMemoryMetrics reads process-level memory information from /proc/self/smaps. All Meters are reporting in bytes. Please note that procfs is only available on Linux-based systems.

• process.memory.vss: Virtual set size. The amount of virtual memory the process can access. Mostly useles, but included for completeness sake.
• process.memory.rss: Resident set size. The amount of process memory currently in RAM.
• process.memory.pss: Proportional set size. The amount of process memory currently in RAM, accounting for shared pages among processes. This metric is the most accurate in terms of "real" memory usage.
• process.memory.swap: The amount of process memory paged out to swap.
• process.memory.swappss: The amount of process memory paged out to swap accounting for shared memory among processes. Since Linux 4.3. Will return -1 if your kernel is older. As with pss, the most accurate metric to watch.

ProcessThreadMetrics

ProcessThreadMetrics reads process-level thread information from /proc/self/status. Please note that procfs is only available on Linux-based systems.

• process.threads: The number of process threads as seen by the operating system.

Notes

• procfs data is cached for 100ms in order to relief the filesystem pressure when Meters based on this data are queried by the registry one after another on collection run.