arthas常用命令速记

arthas常用命令速记

https://github.com/alibaba/arthas

thread

thread -n 3
thread 16

jad 反编译

jad org.slf4j.Logger
jad org.slf4j.Logger -c 61bbe9ba

jad com.taobao.tddl.common.IdGenerator
jad --source-only com.taobao.tddl.common.IdGenerator
jad --source-only com.taobao.tddl.common.IdGenerator > /tmp/IdGenerator.java

反编译生成java代码

mc 编译生成新的class

将修改后的java代码编译成class(因为依赖的关系可能失败)

mc /tmp/IdGenerator.java -d /tmp

redefine 加载新的class

将修改后的class代码热加载

redefine /tmp/IdGenerator.class
redefine -c 1e80bfe8 /tmp/com/alibaba/middleware/drds/worker/task/RegisterTask.class

可以再次jad 反编译确认class中是修改后的代码:

jad --source-only com.alibaba.cobar.server.ServerConnection > /tmp/SC.java

有时候 redefine 看到成功,可是实际并不一定,最好再次 jad 确认一下。

线上环境快速修改代码验证三部曲:jad反编译得到源代码、修改后mc编译成class、redefine替换新的class。

classload

classloader -l
classloader -c 1e80bfe8 -r com/alibaba/middleware/drds/worker/task/RegisterTask.class
classload -t
classload -c 6e0be858
classloader ch.qos.logback.core.AppenderBase

sc

sc -d com.taobao.tddl.common.IdGenerator
sc -df ch.qos.logback.core.AppenderBase

sm

列出class的方法

sm ch.qos.logback.core.AppenderBase -d

getstatic 查看静态成员

通过getstatic查看静态成员,来追踪一个logger没有设置level的话他的输出级别到底是什么?

先 sc 获取classloader的hash

sc -df io.netty.channel.nio.NioEventLoop

getstatic -c 1e80bfe8 io.netty.channel.nio.NioEventLoop logger 'getClass().getName()'
field: logger
@String[io.netty.util.internal.logging.Slf4JLogger]
Affect(row-cnt:1) cost in 5 ms.

然后查看 logger的具体内容,可以看到level等,level为null的话会从父logger继承:

getstatic -c 1e80bfe8 io.netty.channel.nio.NioEventLoop logger 'logger'
field: logger
@Logger[
    serialVersionUID=@Long[5454405123156820674],
    FQCN=@String[ch.qos.logback.classic.Logger],
    name=@String[io.netty.channel.nio.NioEventLoop],
    level=null,
    effectiveLevelInt=@Integer[20000],
    parent=@Logger[Logger[io.netty.channel.nio]],
    childrenList=null,
    aai=null,
    additive=@Boolean[true],
    loggerContext=@LoggerContext[ch.qos.logback.classic.LoggerContext[default]],
]

再次用getstatic命令来确定jar包的location:

getstatic -c 1e80bfe8 io.netty.channel.nio.NioEventLoop logger 'logger.getClass().getProtectionDomain().getCodeSource().getLocation()'
field: logger
@URL[
    BUILTIN_HANDLERS_PREFIX=@String[sun.net.www.protocol],
    serialVersionUID=@Long[-7627629688361524110],
    protocolPathProp=@String[java.protocol.handler.pkgs],
    protocol=@String[file],
    host=@String[],
    port=@Integer[-1],
    file=@String[/home/admin/drds-worker/lib/logback-classic-1.1.8.jar],
    query=null,
    authority=@String[],
    path=@String[/home/admin/drds-worker/lib/logback-classic-1.1.8.jar],
    userInfo=null,
    ref=null,
    hostAddress=null,
    handler=@Handler[sun.net.www.protocol.file.Handler@5a98007],
    hashCode=@Integer[-1217964899],
    tempState=null,
    factory=null,
    handlers=@Hashtable[isEmpty=false;size=3],
    streamHandlerLock=@Object[java.lang.Object@3bf379e9],
    serialPersistentFields=@ObjectStreamField[][isEmpty=false;size=7],
]

然后通过getstatic来获取到这个parent属性的内容。然后通过多个parent操作,可以发现level都是INFO,最终发现ROOT level是INFO:

getstatic -c 1e80bfe8 io.netty.channel.nio.NioEventLoop logger 'logger.parent.parent.parent.parent.parent'
field: logger
@Logger[
    serialVersionUID=@Long[5454405123156820674],
    FQCN=@String[ch.qos.logback.classic.Logger],
    name=@String[ROOT],
    level=@Level[INFO],
    effectiveLevelInt=@Integer[20000],
    parent=null,
    childrenList=@CopyOnWriteArrayList[isEmpty=false;size=4],
    aai=@AppenderAttachableImpl[ch.qos.logback.core.spi.AppenderAttachableImpl@3f0908e1],
    additive=@Boolean[true],
    loggerContext=@LoggerContext[ch.qos.logback.classic.LoggerContext[default]],
]

logger 查看logger配置

列出所有logger,然后修改logger的level

classloader -l
logger -c 27bc2616
ognl -c 6e0be858 '@com.alibaba.cobar.server.ServerConnection@logger'
ognl -c 6e0be858 '@org.slf4j.LoggerFactory@getLogger("root").setLevel(@ch.qos.logback.classic.Level@DEBUG)'

或者

logger --name ROOT --level debug

trace 耗时超过10ms的方法堆栈

查看调用耗时超过 10ms的函数堆栈

stack ch.qos.logback.core.AppenderBase doAppend
trace -j ch.qos.logback.core.AppenderBase doAppend '#cost > 10'

image.png

截图中红框的数字表示代码行号

ongl 调用static函数并查看返回值

ognl '#value1=@com.alibaba.middleware.drds.manager.common.utils.AddressUtil@getHostIp(), {#value1}'
@ArrayList[
    @String[10.0.174.135],
]


 ognl '#value1=@com.alibaba.middleware.drds.worker.Config@getInstance(), {#value1}'
    @ArrayList[
@Config[Config(receivedManagerInfo=true, registeredToManager=true, workerRpcPort=8188, managerIp=10.0.171.193, managerPort=8080, drdsServerPort=3306, drdsManagerPort=3406, host=10.0.118.18, vpcId=vpc-bp1tsocjn451k7ur52vwl, urlToGetVpcId=http://100.100.100.200/latest/meta-data/vpc-id, heartBeatIntervalSeconds=180, registerInveralSeconds=2, manageDrdsIntervalSeconds=60, miniVersion=1, version=0.0.0.41, registerUrl=http://hostPlaceHolder:portPlaceHolder/v1/worker/register, heartBeatUrl=http://hostPlaceHolder:portPlaceHolder/v1/worker/heartBeat, manageDrdsServerUrl=http://hostPlaceHolder:portPlaceHolder/v1/worker/manageDrdsServer, gotVpcId=true, nodeType=drds-server, watcher=null, scheduledThreadPoolExecutor=java.util.concurrent.ScheduledThreadPoolExecutor@3aa3f85f[Running, pool size = 1, active threads = 1, queued tasks = 0, completed tasks = 0])],
]

#Netty的SelectorProvider.provider()创建Selector驱动的时候通过JDK create到的Selector驱动
#如果是windows平台:WindowsSelectorProvider(); macos
#下面是Linux平台的默认Selector驱动:
$ options unsafe true
$ ognl  '#value1=@sun.nio.ch.DefaultSelectorProvider@create(), {#value1}'
@ArrayList[
    @EPollSelectorProvider[sun.nio.ch.EPollSelectorProvider@5bf6cb51],
]
#或者
$  ognl  '#value1=@java.nio.channels.spi.SelectorProvider@provider(), {#value1}'
@ArrayList[
    @EPollSelectorProvider[sun.nio.ch.EPollSelectorProvider@74c4ede7],
]

tt 观察函数调用和回放

先通过tt观察某个函数的调用,然后再用 tt -i 回放这个调用并查看返回值等

tt -t com.alibaba.middleware.drds.manager.common.utils.AddressUtil getHostIp
tt -t com.alibaba.middleware.drds.worker.task.RegisterTask getHostInfoIfNeeded
tt -i 1000
tt -i 1000 -p
tt -n 3 -t com.alibaba.middleware.drds.worker.task.RegisterTask getHostInfoIfNeeded
tt -n 3 -t com.alibaba.middleware.drds.manager.common.utils.AddressUtil getHostIp

 tt -i 1010 -p
     RE-INDEX      1010
     GMT-REPLAY    2019-09-27 12:59:05
     OBJECT        NULL
     CLASS         com.alibaba.middleware.drds.manager.common.utils.AddressUtil
     METHOD        getHostIp
     IS-RETURN     true
     IS-EXCEPTION  false
     COST(ms)      0.577817
     RETURN-OBJ    @String[10.0.118.18]

watch 查看函数调用的参数内容和返回值

指定输出结果的属性遍历深度,默认为 1:

watch  com.alibaba.middleware.drds.manager.common.utils.AddressUtil getHostIp "{params,returnObj}" -x 2

watch com.alibaba.middleware.drds.worker.task.RegisterTask getHostInfoIfNeeded "{params,returnObj}" -x 2
    Press Q or Ctrl+C to abort.
    Affect(class-cnt:1 , method-cnt:1) cost in 56 ms.
    ts=2019-09-27 13:24:00; [cost=0.2698ms] result=@ArrayList[
        @Object[][isEmpty=true;size=0],
        @Boolean[true],
    ]
    ts=2019-09-27 13:24:02; [cost=0.030039ms] result=@ArrayList[
        @Object[][isEmpty=true;size=0],
        @Boolean[true],
    ]

可以看到处理请求的handler是 om.example.demo.arthas.user.UserController.findUserById:

$ watch org.springframework.web.servlet.DispatcherServlet getHandler returnObj
Press Q or Ctrl+C to abort.
Affect(class-cnt:1 , method-cnt:1) cost in 332 ms.
ts=2019-06-04 11:38:06; [cost=2.75218ms] result=@HandlerExecutionChain[
    logger=@SLF4JLocationAwareLog[org.apache.commons.logging.impl.SLF4JLocationAwareLog@665c08a],
    handler=@HandlerMethod[public com.example.demo.arthas.user.User com.example.demo.arthas.user.UserController.findUserById(java.lang.Integer)],
    interceptors=null,
    interceptorList=@ArrayList[isEmpty=false;size=2],
    interceptorIndex=@Integer[-1],
]
  • watch 命令定义了4个观察事件点,即 -b 方法调用前,-e 方法异常后,-s 方法返回后,-f 方法结束后
  • 4个观察事件点 -b、-e、-s 默认关闭,-f 默认打开,当指定观察点被打开后,在相应事件点会对观察表达式进行求值并输出
  • 这里要注意方法入参和方法出参的区别,有可能在中间被修改导致前后不一致,除了 -b 事件点 params 代表方法入参外,其余事件都代表方法出参
  • 当使用 -b 时,由于观察事件点是在方法调用前,此时返回值或异常均不存在

参考资料

官方文档