hsperfdata

What’s this?

This is a golang parser for the newest V2 java HotSpot virtual machine performance data, support all platform theoretically.

What’s hsperfdata file?

It is a log directory created by JVM while running your code. By default it is created inside the tmp folder of the operating system that you are using! This directory is a part of Java Performance counter. And the file in this directory is named by the pid number of java process.

For example, if you are running a java process which pid is 1111, the hsperfdata file is %TEMP%/hsperfdata_<yourusername>/1111 on windows, /tmp/hsperfdata_<yourusername>/1111 on linux.

You can disable creating this directory by using -XX:-UsePerfData or -XX:+PerfDisableSharedMem which is not recommended.

Used as a library

First, you should get the file path string of the hsperfdata file that you want to parser, then use ReadPerfData function to parser the hsperfdata file

There are several functions to get the path, include PerfDataPath(pid string), PerfDataPaths(pids []string), UserPerfDataPaths(user string), CurrentUserPerfDataPaths(), AllPerfDataPaths(), DataPathsByProcessName(processName string).

For how to use these functions, look at hsperfdata function documentation or you can just read the hsperfdata.go to figure out how to use them, the source code is easy to read through.

There is a demo using this project as a go library:

package main

import (
	"fmt"
	"log"
	"sort"

	"github.com/xin053/hsperfdata"
)

func main() {
	filePaths, err := hsperfdata.DataPathsByProcessName("java")
	if err != nil {
		log.Fatal(err)
	}

	for pid := range filePaths {
		entryMap, err := hsperfdata.ReadPerfData(filePaths[pid], true)
		if err != nil {
			log.Fatal("open fail", err)
		}

		var keys []string
		for k := range entryMap {
			keys = append(keys, k)
		}

		sort.Strings(keys)

		for _, key := range keys {
			fmt.Printf("%s=%v\n", key, entryMap[key])
		}
	}
}

Used as a command

There is a demo hstat.go

build hstat yourself

# go1.12+
go build .\cmd\hstat.go

# Usage: hstat pid
# if you have a java process which pid is 1111, then run this
hstat 1111

use the release package

download the executable from the release page, then here you go!

Want deeper?

java HotSpot virtual machine performance data structures

// perfdataHeader http://openjdk.java.net/groups/serviceability/jvmstat/sun/jvmstat/perfdata/monitor/AbstractPerfDataBufferPrologue.html
// source code https://github.com/dmlloyd/openjdk/blob/jdk/jdk/src/jdk.internal.jvmstat/share/classes/sun/jvmstat/perfdata/monitor/AbstractPerfDataBufferPrologue.java
type perfdataHeader struct {
    Magic     uint32 // magic number - 0xcafec0c0
    ByteOrder byte   // big_endian == 0, little_endian == 1
    Major     byte   // major version numbers
    Minor     byte   // minor version numbers
    // ReservedByte byte   // used as Accessible flag at performance data V2
}

// prologue http://openjdk.java.net/groups/serviceability/jvmstat/sun/jvmstat/perfdata/monitor/v2_0/PerfDataBufferPrologue.html
// source code https://github.com/dmlloyd/openjdk/blob/jdk/jdk/src/jdk.internal.jvmstat/share/classes/sun/jvmstat/perfdata/monitor/v2_0/PerfDataBufferPrologue.java
type bufferPrologueV2 struct {
    Accessible   byte  // Accessible flag at performance data V2
    Used         int32 // number of PerfData memory bytes used
    Overflow     int32 // number of bytes of overflow
    ModTimestamp int64 // time stamp of the last structural modification
    EntryOffset  int32 // offset of the first PerfDataEntry
    NumEntries   int32 // number of allocated PerfData entries
}

// entryHeader http://openjdk.java.net/groups/serviceability/jvmstat/sun/jvmstat/perfdata/monitor/v2_0/PerfDataBuffer.html
// source code https://github.com/dmlloyd/openjdk/blob/jdk/jdk/src/jdk.internal.jvmstat/share/classes/sun/jvmstat/perfdata/monitor/v2_0/PerfDataBuffer.java
type entryHeader struct {
    EntryLength  int32 // entry length in bytes
    NameOffset   int32 // offset to entry name, relative to start of entry
    VectorLength int32 // length of the vector. If 0, then scalar.
    DataType     byte  // JNI field descriptor type
    Flags        byte  // miscellaneous attribute flags 0x01 - supported
    DataUnits    byte  // unit of measure attribute
    DataVar      byte  // variability attribute
    DataOffset   int32 // offset to data item, relative to start of entry.
}

You can read the source code hsperfdata.go to get more information, have a good time!

You can open a issue if you have any questions.

Attention

The newest java HotSpot virtual machine performance data structures was V2 when I wrote this code, so these data structures may change from release to release, so this parser code only support JVM performance data V2. If there is new version, please open a issue or pull request, thx.

Reference link

1. http://openjdk.java.net/groups/serviceability/jvmstat/index.html
2. https://github.com/tokuhirom/go-hsperfdata
3. https://github.com/njwhite/telegraf/blob/master/plugins/inputs/hsperfdata/hsperfdata.go
4. https://github.com/twitter/commons/blob/master/src/python/twitter/common/java/perfdata/bin/jammystat.py
5. https://github.com/YaSuenag/hsbeat/blob/master/module/hotspot/hsperfdata/parser.go

License