You've heard of jstack and pstack, well this is a new tool called astack.
astack is a swiss army knife of easily getting thread dumps from JVMs and analyzing them from the command line.
- No complicated network setup (e.g. JMX)
- Designed for linux (sorry OS X)
- Single file script for ease of copying
There are two steps to invoking
astack: (1) select which process to inspect and
(2) specify what action to take. There are two ways to select which process:
-p PID- Select based on PID
-n NAME- Match based on case insensitive search of command line
Note that name searching does not work if more than one process match.
Once you have a process, there are a few options you can take to get different output:
-r- Just get the raw stacktrace as if you sent a SIGQUIT to the java process and captured the stdout with some extra info.
-a NLINES- Group and count the threads by
NLINESof stack and display them in order of occurrence with one representative thread.
-s NSAMPLES- Sample the threaddump a few times and display which ones are most active (most oftenly in RUNNABLE state).
Usage description from the process itself:
Usage: astack [options] Options: -h, --help show this help message and exit -p PID, --pid=PID process pid -n NAME, --process-name=NAME match name of process -r, --raw print the raw stacktrace and exit -u, --upgrade automatically upgrade -a AGG, --aggregate=AGG Aggregate stacktraces (specify the number of lines to aggregate) -s SAMPLE, --sample=SAMPLE Sample stacktraces to the most active ones (specify the number of seconds) -i INPUT, --input=INPUT read stacktrace from file (or - with stdin) --pretty, --pretty Force colors -g MATCH, --grep=MATCH Show only threads that match text
Theory of Operation
Rather than hooking into the JVM and asking it for a stack trace via instrumentation,
this script takes the approach of sending a
SIGQUIT signal and extracting the stacktrace
from the JVM while it's running. The way it does this is by using
gdb to temporarily
redirect stdout while the JVM is spouting out the thread dump, and switching it back when
it's done. This does mean that it can occasionally get some artifacts if your JVM is
rapidly sending output to stdout. In most typical scenarios (e.g. log lines) you wouldn't
see any interference.
The advantage of using this technique is that even when a JVM is under heavy load and cannot
fulfill a instrumentation approach, the low level response to a
SIGQUIT signal is still
functional. In most cases, thread dumps are most useful when the JVM is at its limit, so
this technique can get interesting results very easily.
$ sudo pip install astack
Ubuntu 10.04 and up support
On Ubuntu 10.04 and up you'll need to run this (as root):
# echo 0 > /proc/sys/kernel/yama/ptrace_scope
Bugs & Issues
Feel free to file any issues with github's issues page.
MIT License, copyright HubSpot 2012. See the bundled LICENSE file for details.