README
¶
诸葛
救命用的profile。
在程序CPU、内存等异常的情况下，自动生成Profile并提交到Profiler以便分析。
安装
首先，你需要部署一个 pyroscope server
安装方式很简单，官方提供各种操作系统的安装命令，并且你还可以通过Docker安装，或者部署在Kubernates环境中。
建议你配置身份认证功能，这样就避免别人恶意往你的服务器中灌数据。并且设置数据保留的期限，节省成功。
我通过rpm安装在Linux操作系统中,所以修改配置/etc/pyroscope/server.yml 如下：
retention: 24h
auth:
# Specifies which role will be granted to a newly signed up user.
# Supported roles: Admin, ReadOnly. Defaults to ReadOnly.
# The parameter affects all authentication providers.
signup-default-role: ReadOnly
# Internal authentication provider configuration.
internal:
enabled: true
signup-enabled: false
在你的应用中发送Profile
官方的pyroscope agent会定时每10秒发送profile到pyroscope server。
正如官方所说，采集profile会带来2%~5%性能的损耗，不止于此，每10秒的采样数据也会带来磁盘的占用，所以我们想在必要、紧急的情况下才上传profile。
这正是这个项目的目的。
你可以通过下面的initPyroscope方法，在你的应用程序中配置Pyroscope,注意修改你的Pyroscope server地址和token。
package main
import (
"math/rand"
"os"
"time"
"github.com/smallnest/zhuge"
"github.com/smallnest/zhuge/examples/btree"
)
func init() {
rand.Seed(time.Now().UnixNano())
}
var profiler *zhuge.SampleProfiler
func main() {
initPyroscope()
tick := time.NewTicker(10 * time.Second)
for range tick.C {
btree.Run(16)
}
}
func initPyroscope() {
sc := zhuge.Config{
ApplicationName: "zhuge.test_client",
ServerAddress: "http://rpcx.io:4040",
Tags: map[string]string{"host": "192.168.3.1"},
Logger: zhuge.StandardLogger,
AuthToken: os.Getenv("PYROSCOPE_AUTH_TOKEN"),
ProfileTypes: []zhuge.ProfileType{
zhuge.ProfileCPU,
zhuge.ProfileAllocObjects,
zhuge.ProfileAllocSpace,
zhuge.ProfileInuseObjects,
zhuge.ProfileInuseSpace,
},
}
var err error
profiler, err = zhuge.NewSampleProfiler(sc)
if err != nil {
panic(err)
}
}
然后你可以在想采样的时候执行下面的语句就可以自动采样一次了:
go profiler.SampleNow()
TODO
下一步计划是参考mosn/holmes处理逻辑，由用户设定采样的触发条件，比如CPU占用高或者内存占用高的时候自动采样上传。
欢迎一起开发它的功能。
Credits
- pyroscope-io/client: the initial codes is cloned from pyroscope-io/client.
Documentation
¶
Index ¶
Constants ¶
This section is empty.
Variables ¶
var DefaultProfileTypes = []ProfileType{ ProfileCPU, ProfileAllocObjects, ProfileAllocSpace, ProfileInuseObjects, ProfileInuseSpace, }
var Labels = pprof.Labels
var StandardLogger = &standardLoggerImpl{}
Functions ¶
Types ¶
type Config ¶
type Config struct {
ApplicationName string // e.g backend.purchases
Tags map[string]string
ServerAddress string // e.g http://pyroscope.services.internal:4040
AuthToken string // specify this token when using pyroscope cloud
SampleRate uint32
Logger Logger
ProfileTypes []ProfileType
DisableGCRuns bool // this will disable automatic runtime.GC runs between getting the heap profiles
}
type Logger ¶
type Logger interface {
Infof(_ string, _ ...interface{})
Debugf(_ string, _ ...interface{})
Errorf(_ string, _ ...interface{})
}
Logger is an interface that library users can use It is based on logrus, but much smaller — That's because we don't want library users to have to implement all of the logrus's methods
type ProfileType ¶
type ProfileType string
const ( ProfileCPU ProfileType = "cpu" ProfileInuseObjects ProfileType = "inuse_objects" ProfileAllocObjects ProfileType = "alloc_objects" ProfileInuseSpace ProfileType = "inuse_space" ProfileAllocSpace ProfileType = "alloc_space" DefaultSampleRate = 100 )
type SampleProfiler ¶
type SampleProfiler struct {
// contains filtered or unexported fields
}
func NewSampleProfiler ¶
func NewSampleProfiler(cfg Config) (*SampleProfiler, error)
SampleProfiler creates a priofiler profiling go code
func (*SampleProfiler) SampleNow ¶
func (sp *SampleProfiler) SampleNow()
type SampleSession ¶
type SampleSession Session
func NewSampleSession ¶
func NewSampleSession(c SessionConfig) (*SampleSession, error)
func (*SampleSession) SampleNow ¶
func (ps *SampleSession) SampleNow()
type Session ¶
type Session struct {
// contains filtered or unexported fields
}
func NewSession ¶
func NewSession(c SessionConfig) (*Session, error)
Source Files
Directories