Summary of Java problem location and solutions-javaTutorial-php.cn

Background
"There is no problem offline", "There cannot be problems with the code because of the system", "Can I remotely debug online"
Online problems are different from bugs during development and runtime. Environment, pressure, concurrency situations, specific business related. For online problems, it is important to use the tools available in the online environment to collect necessary information to locate the problem.
It is difficult to intuitively obtain data on bugs and resource bottlenecks that cause problems. It is necessary to infer the root cause of the problem based on resource usage data, logs and other information. And the location of difficult problems usually requires the use of different methods to trace the source.
In this wiki, I have compiled the tools I have used and shared some cases.

1. Frequently Asked Questions
1.1 Availability
Here are a few common situations that lead to service availability:
a) 502 Bad Gateway
The most common problem for application systems, especially http-based applications Nothing is more serious than "502 Bad Gateway", which means that the back-end service is completely unavailable. Possible reasons
Insufficient resources 1: caused by garbage collection, which will cause serious application suspension when the CMS leaks application memory or lacks memory. .
Insufficient resources 2: Insufficient number of server threads. Common web servers such as tomcat and jetty are configured with a maximum working thread.
Insufficient resources 3: Insufficient database resources. The database usually uses a connection pool configuration. The maxConnection configuration is too low and too high. How slow a query will block the working thread of the web server
Insufficient resources 4: IO resource bottleneck, online environment IO is shared, especially for mixed environments (fortunately, CRM does not have this situation, but there are many agents), we The commonly used log4j logging tool is also an exclusive resource for each recorded log file. The thread must first obtain the lock before it can record data to the log.
... ...
Various OOM
b) Socket exceptions
Common Connection reset by peer, Broken Pipe, EOFException
Network problems: in the case of cross-operator and computer room access You may encounter
Program bug: socket is closed abnormally
1.2 Average response time
The most intuitive symptom when a problem occurs in the system. This parameter can provide early warning before the situation worsens and infects other services, causing the entire system to become unavailable. Possible reasons:
Resource competition 1: CPU
Resource competition 2: IO
Resource competition 3: network IO
Resource competition 4: Database
Resource competition 5: solr, medis
Downstream Interface: Exception causes response delay
1.3 Machine alarm
Compared with application service unavailability, this type of error will not directly lead to service unavailability, and if there is confusion, multiple services deployed on the machine may interfere with each other:
CPU
Disk
fd
IO (Network Disk)
1.4 Summary
It took me a long time to write, many cases are mentioned repeatedly, usually the cause of online problems is nothing more than system resources. , applications, and mastering the tools to monitor and view these resources and data will make it easier to locate online problems.
2Commonly used tools
2.1 Linux tools
a) sysstat:
iostat: View read and write pressure

    [sankuai@cos-mop01 logs]$ iostat
    Linux 2.6.32-20131120.mt (cos-mop01.lf.sankuai.com)     2015年10月21日     _x86_64_    (4 CPU)
    avg-cpu:  %user   %nice %system %iowait  %steal   %idle
               1.88    0.00    0.87    0.12    0.05   97.07
    Device:            tps   Blk_read/s   Blk_wrtn/s   Blk_read   Blk_wrtn
    vda               1.88        57.90        12.11 2451731906  512911328
    vdb               0.01         0.40         1.41   17023940   59522616
    vdc               1.14        28.88        36.63 1223046988 1551394969

Copy after login

sar: View CPU network IO IO, enable parameters to view historical data

    /etc/sysconfig/sysstat
    HISTORY=7
    /etc/cron.d/sysstat
    */10 * * * * root /usr/lib/sa/sa1 1 1
       
    sar -u/-r/-B/-b/-q/-P/-n -f /var/log/sa/sa09

Copy after login

b) top
Pay attention to load, cpu, mem, swap
You can view resource information according to threads (version is greater than 3.2.7)

top - 19:33:00 up 490 days,  4:33,  2 users,  load average: 0.13, 0.39, 0.42
Tasks: 157 total,   1 running, 156 sleeping,   0 stopped,   0 zombie
Cpu(s):  4.9%us,  2.7%sy,  0.0%ni, 92.2%id,  0.0%wa,  0.0%hi,  0.0%si,  0.3%st
Mem:   5991140k total,  5788884k used,   202256k free,     4040k buffers
Swap:  2096440k total,   447332k used,  1649108k free,   232884k cached
  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  SWAP CODE DATA COMMAND
18720 sankuai   20   0 8955m 4.3g 6744 S 22.6 74.5 174:30.73    0    4 8.6g java
27794 sankuai   20   0 5715m 489m 2116 S 11.6  8.4   3922:43 121m    4 3.9g java
13233 root      20   0  420m 205m 2528 S  0.0  3.5   1885:15  91m    4 304m puppetd
21526 sankuai   20   0 2513m  69m 4484 S  0.0  1.2  45:56.28  37m    4 2.4g java

Copy after login

c) vmstat

[sankuai@cos-mop01 logs]$ vmstat
procs -----------memory---------- ---swap-- -----io---- --system-- -----cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 0  0 447332 200456   4160 234512    0    0    11     6    0    0  2  1 97  0  0

Copy after login

d) tcpdump
A tool for locating network problems. You can see the details of TCPIP messages. You need to be familiar with the TCPIP protocol at the same time. It can be used in conjunction with wireshark.
Common scenario analysis of network delay, network packet loss, and network problem analysis in complex environments.

#!/bin/bash
tcpdump -i eth0 -s 0 -l -w - dst port 3306 | strings | perl -e '
while() { chomp; next if /^[^ ]+[ ]*$/;
    if(/^(SELECT|UPDATE|DELETE|INSERT|SET|COMMIT|ROLLBACK|CREATE|DROP|ALTER|CALL)/i)
    {
        if (defined $q) { print "$q\n"; }
        $q=$_;
    } else {
        $_ =~ s/^[ \t]+//; $q.=" $_";
    }
}'

Copy after login

3.2 java tools
a) jstat

[sankuai@cos-mop01 logs]$ jstat -gc 18704
 S0C    S1C    S0U    S1U      EC       EU        OC         OU       MC     MU    CCSC   CCSU   YGC     YGCT    FGC    FGCT     GCT
3584.0 3584.0  0.0    0.0   24064.0  13779.7   62976.0      0.0     4480.0 677.9  384.0   66.6       0    0.000   0      0.000    0.000

Copy after login

b) jmap

jmap -dump:format=b,file=heap.bin $pid

Copy after login

c) jstack or kill -3
View deadlocks and thread waits.
Thread status:
Running
TIMED_WAITING(on object monitor)
TIMED_WAITING(sleeping)
TIMED_WAITING(parking)
WAINTING(on object monitor)
d) jhat jconsole
jhat is difficult to use jconsole to obtain information through jmx, which has an impact on performance
e) gc log
-XX:+UseParallelOld
-XX:+ConcurrentMultiSweep
3.3 Third-party tools
a) mat
Object details

Summary of Java problem location and solutions

inboud/outbound

Summary of Java problem location and solutions

thread overview

Summary of Java problem location and solutions

配置项

Summary of Java problem location and solutions

./MemoryAnalyzer -keep_unreachable_objects heap_file
4. 案例分析
4.1 cpu高

现象：CPU报警
定位问题：
查看CPU占用高的线程

sankuai@sin2:~$ ps H -eo user,pid,ppid,tid,time,%cpu|sort -rnk6 |head -10
sankuai  13808 13807 13808 00:00:00  8.4
sankuai  29153     1 29211 00:21:13  0.9
sankuai  29153     1 29213 00:20:01  0.8
sankuai  29153     1 29205 00:17:35  0.7
sankuai  29153     1 29210 00:11:50  0.5
sankuai  29153     1  1323 00:08:37  0.5
sankuai  29153     1 29207 00:10:02  0.4
sankuai  29153     1 29206 00:07:10  0.3
sankuai  29153     1 29208 00:06:44  0.2

Copy after login

thread dump

    jstack $pid > a.txt
    printf %x $tid
        $xTID

Copy after login

查找线程执行的代码

"main-SendThread(cos-zk13.lf.sankuai.com:9331)" #25 daemon prio=5 os_prio=0 tid=0x00007f78fc350000 nid=$TIDx runnable [0x00007f79c4d09000]
java.lang.Thread.State: RUNNABLE
    at org.apache.zookeeper.ClientCnxn$SendThread.run(ClientCnxn.java:1035)
    at java.util.concurrent.FutureTask.run(FutureTask.java:266)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
    at java.lang.Thread.run(Thread.java:745)

Copy after login

4.2 io高

现象：磁盘IO报警
环境：需要安装sysstat工具
定位问题：
a) 查看CPU占用高的线程

pidstat -d -t -p $pid

b) 其他同4.1

4.3 资源
a) 数据库

"DB-Processor-13" daemon prio=5 tid=0x003edf98 nid=0xca waiting for monitor entry [0x000000000825f000]
java.lang.Thread.State: BLOCKED (on object monitor)
                at ConnectionPool.getConnection(ConnectionPool.java:102)
                - waiting to lock <0xe0375410> (a beans.ConnectionPool)
                at Service.getCount(ServiceCnt.java:111)
                at Service.insert(ServiceCnt.java:43)
"DB-Processor-14" daemon prio=5 tid=0x003edf98 nid=0xca waiting for monitor entry [0x000000000825f020]
java.lang.Thread.State: BLOCKED (on object monitor)
                at ConnectionPool.getConnection(ConnectionPool.java:102)
                - waiting to lock <0xe0375410> (a beans.ConnectionPool)
                at Service.getCount(ServiceCnt.java:111)
                at Service.insertCount(ServiceCnt.java:43)

Copy after login

b) log

"RMI TCP Connection(267865)-172.16.5.25" daemon prio=10
tid=0x00007fd508371000 nid=0x55ae  waiting for monitor entry
[0x00007fd4f8684000]   java.lang.Thread.State:  BLOCKED (on object monitor)
at org.apache.log4j.Category.callAppenders(Category.java:201)
-  waiting to lock <0x00000000acf4d0c0> (a org.apache.log4j.Logger)
at org.apache.log4j.Category.forcedLog(Category.java:388)
at org.apache.log4j.Category.log(Category.java:853)
at org.apache.commons.logging.impl.Log4JLogger.warn(Log4JLogger.java:234)
at com.xxx.core.common.lang.cache.remote.MemcachedClient.get(MemcachedClient.java:110)

Copy after login

c) web server

有两个非常重要的系统参数：

maxThread: 工作线程数

backlog：TCP连接缓存数，Jetty(ServerConnector.acceptQueueSize) Tomcat(Connector.acceptCount)，高并发下设置过小会有502

4.4 gc

a) CMS fail
promotion failed

    172966 2015-09-18T03:47:33.108+0800: 627188.183: [GC 627188.183: [ParNew (promotion failed)
     172967 Desired survivor size 17432576 bytes, new threshold 1 (max 6)
     172968 - age   1:   34865032 bytes,   34865032 total
     172969 : 306688K->306688K(306688K), 161.1284530 secs]627349.311: [CMS CMS: abort preclean due to time 2015-09-18T03:50:14.743+0800: 627349.818:
    [CMS-concurrent-abortable-preclean: 1.597/162.729 secs] [Times: user=174.58 sys=84.57, real=162.71 secs]
     172970  (concurrent mode failure): 1550703K->592286K(1756416K), 2.9879760 secs]
    1755158K->592286K(2063104K), [CMS Perm : 67701K->67695K(112900K)], 164.1167250 secs] [Times: user=175.61 sys=84.57, real=164.09 secs]

Copy after login

concurrent fail

    [CMS2015-09-18T07:07:27.132+0800: 639182.207: [CMS-concurrent-sweep: 1.704/13.116 secs] [Times: user=17.16 sys=5.20,real=13.12 secs]
    443222 (concurrent mode failure): 1546078K->682301K(1756416K), 4.0745320 secs] 1630977K->682301K(2063104K), [CMS Perm :67700K->67693K(112900K)], 15.4860730 secs] [Times: user=19.40 sys=5.20, real=15.48 secs]

Copy after login

b) 连续Full GC

应用存在内存泄漏，垃圾收集会占用系统大量cpu时间，极端情况下可能发生90%以上时间在做GC的情况。
在系统使用http访问check alive或者使用了Zookeeper这种通过心跳保证存活性的应用中，会可用性异常或者被zk的master剔除。

5. 注意
保留现场：threaddump top heapdump
注意日志记录：文件数据库

The above is the detailed content of Summary of Java problem location and solutions. For more information, please follow other related articles on the PHP Chinese website!