1. Tomcat Expert Series
Troubleshooting in Production
Filip Hanik
SpringSource
2009
Copyright 2007 SpringSource. Copying, publishing or distributing without express written permission is prohibited.
2. Topics in this Session
• Brief overview of JVM memory layout
• Understanding Out Of Memory Errors
– Causes
– Solutions
• Error logs and stack traces
• The tale of the thread dump
2
3. Topics in this Session
• Using OS utilities to narrow down the
problem
• JMX – what it can do for you
3
4. The JVM process heap
OS Memory (RAM)
Process Heap (java/java.exe)
Java Object Heap
Everything else…
4
5. Storing data in memory
• JVM manages the process heap (in most cases)
– JNI managed memory would be an exception, and there
are others
• No shared memory between processes
– At least not available through the Java API
• JVM creates a Java Heap
– Part of the process heap
• Configured through –Xmx and –Xms settings
5
6. JVM Process Heap
• Maximum size is limited
– 32 bit size, roughly 2GB
– 32 bit JVM on 64 bit OS, roughly 3.7GB
– 64 bit, much much larger ?
• If 2GB is the max for the process
– -Xmx1800m –Xms1800m – not very good
– Leaves no room for anything else
6
7. Gotcha #1
• -Xmx and -Xms
– Only controls the Java Object Heap
• Often misunderstood to control the process heap
• Confusion leads to incorrect tuning
– And in some cases, the situation worsens
7
8. Java Object Heap Allocation
• Aggressive Heap Allocation
• -XX:MinHeapFreeRatio=
– Default is 40 (40%)
– When the JVM allocates memory, it allocates enough to
get 40% free
• -XX:MaxHeapFreeRatio=
– Default 70%
– To give back memory when a majority of it is not used
• Not important when –Xms == -Xmx
8
9. Java Object Heap
Java Object Heap (-Xmx/-Xms)
Young Generation
Old Generation
A good size for the YG is 33% of the total heap
9
10. Java Object Heap
• Young Generation
– All new objects are created here
– Only moved to Old Gen if they survive one or more minor
GC
• Sized using
– -Xmn – fixed value
– -XX:NewRatio=<value> - dynamic sizing
– -XX:MaxNewSize/-XX:NewSize – similar to -Xmx/-Xms
• Survivor Spaces
– 2, used during the GC algorithm (minor collections)
– Mainly to alleviate fragmentation
10
11. Young Generation
Young size(-XX:NewRatio )
Survivor Ratio(-XX:SurvivorRatio )
Eden Space Survivor Space
New Objects
2Mb default
To From
64Kb default
11
12. Old Generation
Tenured Space
5Mb min 44Mb max (default)
Garbage collection section will
explain in detail how these spaces
are used during the GC process.
12
13. Java Heap Space
• java.lang.OutOfMemoryError: Java heap space
– Most common out of memory error
– Caused by too many objects in the Java heap
• Solution
– Increase -Xmx if possible
– Add -XX:+HeapDumpOnOutOfMemoryError
– Fix memory leak if application is consuming more
memory than expected
• Side effects
– Increasing -Xmx can have other side effects
13
14. JVM Process Heap
• Yes, there is more...
– Permanent Space
– Code Generation
– Socket Buffers
– Thread stacks
– Direct Memory Space
– JNI Code
– Garbage Collection
– JNI allocated memory
14
15. Permanent Space
• Permanent Generation
– Permanent Space (name for it)
– 4Mb initial, 64Mb max
• Stores classes, methods and other meta data
– -XX:PermSize=<value> (initial)
– -XX:MaxPermSize=<value> (max)
• Common OOM for webapp reloads
– Separate space for pre-historic reasons
– Early days of Java, class GC was not common, reduces
size of the Java Heap
– Does not exist in IBM JVM
15
16. Permanent Space
• Permanent Space Memory Errors
– Too many classes loaded
– Classes not unloaded/not being GC:ed
– Unaffected by –Xmx flag
• java.lang.OutOfMemoryError: PermGen space
– Many situations, increasing max perm size will help
– i.e. no leak, but just not enough memory
– Others will require to fix the leak
16
17. Socket Buffers
• Each connection contains two buffers
– Receive buffer ~37k
– Send buffer ~25k
• Configured in Java code
– Default and Max limits set in kernel
• Very common tune parameter for content delivery
• Usually hit other limits than memory before you
run out of memory
– IOException: Too many open files (for example)
– SocketException: No buffer space available
17
18. Thread Stacks
• Each thread has a separate memory space called
“thread stack”
• Configured by –Xss
• Default value depends on OS/JVM
• As number of threads increase, memory usage
increases
18
19. Thread Stacks
• java.lang.OutOfMemoryError: unable to create new
native thread
• Solution
– Decrease –Xmx (or other space) and/or
– Decrease –Xss
– Or, you have a thread leak, fix the program
• Gotcha
– Increasing –Xmx (32bit systems) will leave less room for
threads if it is being used, hence the opposite of the
solution
– Too low –Xss value can cause
java.lang.StackOverflowError
• Thread dump will lead to instant answer
– Returns all the threads 19
20. Garbage Collection
• However, if there is excessive GC
• java.lang.OutOfMemoryError: GC overhead limit
exceeded
• 98% of the time is spent in GC
– less than 2% of the heap is recovered
• To disable
– -XX:-UseGCOverheadLimit
• To enable
– -XX:+UseGCOverheadLimit
20
21. GC: How It Works
Eden Space Survivor Space
From To
Tenured Space
4. Next time Eden is full
1. New object isis fullnd minor 1st Eden
3. Copy EDEN created
Copy nd fillsEdenobjects remain in
5. If 2 from and to 2 –
2. When surviving objects into collection or 1st
Copy from copied nd the tenured
survivor space 2 to
These get 1st to
21
23. Troubleshooting steps
• What are the symptoms
• How does the problem manifest itself
• How can I dig into the problem
23
24. Tomcat Logs
• Tomcat is really good at logging
• It doesn’t spit out info you don’t need know
• When an error happens it can, however, generate
tons of log entries
• Common IO exceptions are swallowed
– Normal TCP behavior
24
25. Tomcat Logs
• Unexpected errors are always logged
– Application and container errors
• Logs are always a resource
• Log entries are categorized
– DEBUG
– INFO
– WARNING
– SEVERE
– FATAL
25
26. Tomcat Logs
• INFO – no error, just information given to you
• WARNING – you might care a little bit
• SEVERE – yes, now you got an error
• FATAL – what ever this is, it cant be good!
• Always pay attention to the log level
26
27. Tomcat Logs
• So what do I need to look at
• SEVERE – yes, now you got an error
• It’s easy to ‘grep’ logs for these entries
27
28. Tomcat Logs
• Very basic error
• So Tomcat can’t log application errors
– Only if the application doesn’t ‘trap’ the error, tomcat
will catch it and spit out some info
// An uncaught application error could look like this
SEVERE: Servlet.service() for servlet jsp threw exception
java.lang.NullPointerException at org.apache.jsp.npe_jsp._jspService(npe_jsp.java:55)
• Found in catalina.2008-03-28.log
– Uncaught application exception
28
29. Java Stack Traces
• Shows the code execution path up until the error
happened
// An uncaught application error could look like this
SEVERE: Servlet.service() for servlet jsp threw exception
java.lang.NullPointerException
at org.apache.jsp.npe_jsp._jspService(npe_jsp.java:55)
at org.apache.jasper.runtime.HttpJspBase.service(HttpJspBase.java:70)
at javax.servlet.http.HttpServlet.service(HttpServlet.java:803)
at org.apache.jasper.servlet.JspServletWrapper.service(JspServletWrapper.java:374)
at org.apache.jasper.servlet.JspServlet.serviceJspFile(JspServlet.java:337)
at org.apache.jasper.servlet.JspServlet.service(JspServlet.java:266)
at javax.servlet.http.HttpServlet.service(HttpServlet.java:803)
at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:290)
at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:206)
at org.apache.catalina.core.StandardWrapperValve.invoke(StandardWrapperValve.java:233)
at org.apache.catalina.core.StandardContextValve.invoke(StandardContextValve.java:175)
at org.apache.catalina.core.StandardHostValve.invoke(StandardHostValve.java:128)
at org.apache.catalina.valves.ErrorReportValve.invoke(ErrorReportValve.java:102)
at org.apache.catalina.core.StandardEngineValve.invoke(StandardEngineValve.java:109)
at org.apache.catalina.valves.RequestDumperValve.invoke(RequestDumperValve.java:151)
at org.apache.catalina.connector.CoyoteAdapter.service(CoyoteAdapter.java:286)
at org.apache.coyote.http11.Http11Processor.process(Http11Processor.java:844)
at org.apache.coyote.http11.Http11Protocol$Http11ConnectionHandler.process(Http11Protocol.java:583)
at org.apache.tomcat.util.net.JIoEndpoint$Worker.run(JIoEndpoint.java:447)
at java.lang.Thread.run(Thread.java:595)
29
30. Java Stack Traces
• Traces can be chained, only the root cause is the
real error
// An uncaught application error could look like this
SEVERE: Servlet.service() for servlet jsp threw exception
java.lang.RuntimeException: java.lang.NullPointerException
at org.apache.jsp.npe_jsp._jspService(npe_jsp.java:59)
at org.apache.jasper.runtime.HttpJspBase.service(HttpJspBase.java:70)
…
Caused by: java.lang.NullPointerException
at org.apache.jsp.npe_jsp._jspService(npe_jsp.java:57)
... 19 more
• Always find the top most error and the bottom
most root cause
30
31. Tomcat Logs
• If the logs are not yielding any information
– Gather more information about the error
• Can you reproduce it
• Is it happening on your server
– Trace the request down
31
32. Viewing Requests
• Access logs can help
– They record every request and it’s response code
– You can print out headers, cookies, request and session
information
• Often very useful to see how traffic is flowing
• When using httpd in front
– httpd access log combined with Tomcat access log
– Excellent way to consolidate requests
32
33. Seeing traffic
• When using httpd in front
– httpd has a mod_dumpio module
– Print out everything one wants to know
– Useful when privileges for a sniffer are not present
• RequestDumper Valve
– Valve that spits out everything, similar to mod_dumpio
– Poorly designed, it breaks out the output into multiple
log statement
– Not recommended
33
34. Seeing traffic
• Network sniffers (client/server)
– Nothing compares to getting exact data
– Wireshark, ethereal, tcpdumper, etc
• Many choices, just pick one.
– Often one is already installed
– Requires root privileges
• Client side visualizer
– MS Fiddler is an excellent tool for Windows users
– Firefox firebug
34
35. Seeing traffic
• Ability to map
– Request to error
– Error to a time frame
– Error to a client
• Traffic pattern can
– Help you reproduce the error
– Resolve the issue faster
35
36. Thread dumps
• Displays the state of all threads in a virtual
machine
• Provides plenty of information about activity and
any dead locks
• Provides a trace where each thread started to
where its current point in execution
36
37. Thread dumps
• On Unix -> kill -3 <tomcat pid>
– jstack -l also works
– On Solaris, powerful pstack utility
• On Windows Ctrl+Break
– JDK 1.6+ you have jstack to help
• Tanuki Wrapper
– telnet <host:port> D
• Thread dump is printed to stdout
– Or wherever stdout is redirected to
– Don't send it to /dev/null !
37
38. The tale of Thread dumps
• Alternate way to dump
– jstack – tool that comes with the JDK
– Use -l option with jstack to get lock information
• More than just threads
Heap
def new generation total 9088K, used 5497K [0x04070000, 0x04a40000, 0x067d0000)
eden space 8128K, 60% used [0x04070000, 0x045402a0, 0x04860000)
from space 960K, 59% used [0x04950000, 0x049de428, 0x04a40000)
to space 960K, 0% used [0x04860000, 0x04860000, 0x04950000)
tenured generation total 121024K, used 1656K [0x067d0000, 0x0de00000, 0x24070000)
the space 121024K, 1% used [0x067d0000, 0x0696e068, 0x0696e200, 0x0de00000)
compacting perm gen total 12288K, used 4482K [0x24070000, 0x24c70000, 0x2c070000)
the space 12288K, 36% used [0x24070000, 0x244d0920, 0x244d0a00, 0x24c70000)
ro space 8192K, 66% used [0x2c070000, 0x2c5bd978, 0x2c5bda00, 0x2c870000)
rw space 12288K, 52% used [0x2c870000, 0x2ceb9cb8, 0x2ceb9e00, 0x2d470000)
• JDK 1.6+ also prints out memory stats
38
39. The tale of Thread dumps
• More than just threads
• Dead lock detection
Found one Java-level deadlock:
=============================
"pool-2-thread-6":
waiting for ownable synchronizer 0x482eaeb0,
(a java.util.concurrent.locks.ReentrantReadWriteLock$NonfairSync),
which is held by "http-8081-2"
"http-8081-2":
waiting to lock monitor 0x08143b14 (object 0x482eade0,
a org.apache.catalina.ha.session.DeltaRequest),
which is held by "pool-2-thread-6"
39
40. The tale of Thread dumps
• Thread information
• First line for each thread contains critical info
"http-8082-7" thread name
daemon type of thread [daemon, empty means non daemon]
prio=10 thread priority [1..10]
tid=0x09142000 C++ pointer to JVM OSThread object
nid=0x67c kernel thread identifier
in Object.wait() what the thread is doing
[0xba3ff000..0xba3ff4e0] address space
40
41. The tale of Thread dumps
• Thread information
– CPU usage very high, caused by a spinning thread, but
which one?
nid=0x67c kernel thread identifier
– On linux for example, one can get CPU usage per thread
ps -eL -o pid,%cpu,lwp | grep -i `ps -ef
|grep -v grep |grep java|awk '{print $2}'` |grep -v 0.0
• Lists all threads that run inside a java process with
a CPU usage higher than 0.0%
41
42. The tale of Thread dumps
• Threads dumps can help you identify
• Show what threads are waiting for a lock
• Dead locks
• Spinning threads, and what code it belongs
• Memory usage
• Cause of an unresponsive JVM
42
43. The tale of Thread dumps
• When taking thread dumps ALWAYS take two or
more dumps
• This will help you see if threads are changing
execution path
• Single thread dump can cause a lot of “false
positives”, where you think a thread is stuck but
it’s not
43
44. The tale of Thread dumps
• When taking thread dumps ALWAYS take two or
more dumps
• Sometimes when a OS is overloaded, maybe
swapping, threads move, but very slowly
• If you hadn’t had two dumps, you’d never know
that!
44
45. The tale of Thread dumps
• Examples of thread dumps and GC logs
– Stuck but not dead locked (jvm-locked.txt)
– What went wrong? (five-thread-dumps.txt)
– Memory leak confirmed (gc.log/gc-logs-explained.txt)
– Excessive GC (gc-excessive-object.creation.txt)
45
46. OS utilities
• Troubleshooting a Tomcat server
– often involves more than just Tomcat and JVM
troubleshooting
– OS utilities come very handy
• Tomcat logs internal errors
– If they spawned an exception
– Container bugs are hard to troubleshoot, but rare in
frequencies
46
47. OS utilities
• File descriptors are common problems
– Most OS have a limit on FD
– A FD can be an open file or a socket
• File descriptor leaks are also very common in web
applications
– java.io.IOException: Too many open files
• Use utilities to track it down
47
48. OS utilities
• File descriptors
– Sockets
– Open files
• On Linux – lsof –p <process id>
– Will list all open FD
– This will often put you on the right path on what is
going wrong
– Solaris – pfiles
48
49. OS utilities
• Active Connections
– Can be HTTP/AJP connections
– Useful to track database or other connections
• netstat is an excellent tool to view sockets and
their current state
• On Unix
– Able to track socket buffers and their usage
– Send buffer filling up, slow client or bad network
– Receive buffer filling up, application is not reading fast
enough
49
50. OS utilities
• Linux
– nmon
– Very nice system stats collector
– CPU,Memory,disk,network and more
• Windows
– Expand your task manager
– It reports as much as you need (thread counts, IO
activity, virtual vs residential memory
50
51. JMX
• Both Tomcat and the JVM
– Make information available through JMX
• jconsole
– Utility that comes with the JDK
– Lets you attach to a JVM and get information
• JVM or application inoperable
– JMX may not report accurately
– Don't rely on it exclusively
51
52. So far…
• We’ve covered tons of options to troubleshoot
your systems
– without actually adding any software
– This is usually a requirement for production
troubleshooting
• There are additional options available
• Such as profilers and monitoring applications
52
53. Profilers
• Comes in all shapes and sizes (and prices)
• My preferences
– www.yourkit.com
– Very biased, as a Tomcat developer, I get it for free
• But it has many advantages such as:
– Inexpensive
– Works fairly well in production environments
– Excellent support
– Great tool to use
53
54. Profilers
• Only last resort for production
– When you are unable to reproduce in UAT/QA/etc
– But sometimes required to solve the problem
54
55. Summary
• Gather all the information first
– Even if you think you don't need it
– Create a check list
• Thread dumps
• Logs
• Configuration files
• OS statistics
• etc
• Chasing a problem without all that, you can easily
miss something
– And you'll chase a needle in a haystack
55