search

patric-r / jvmtop

Java monitoring for the command-line, profiler included
GNU General Public License v2.0
1.22k stars 252 forks source link

Additional command line switches to support scripting #70

Closed mskadu closed 7 years ago

mskadu commented 8 years ago

I am using your excellent script to do some diagnostics on a java-based legacy system. It would be dead useful to be able to provide additional switches that can provide the ability to

a) write/ append output from every cycle to file (--write ??) b) Produce only specific output c) Hide the first two line (--hide-header ??)

This will make the tool scripting-friendly. Thanks!

patric-r commented 8 years ago

a) jvmtop follows the common unix tradition to only write to stdout/stderr. If you need its output in a file, you're a free to redirect it. In order to output only once and exit which you most probably need for scripting, use the --once switch, e.g. ./jvmtop.sh --once > myfile

b) Can you clarify what kind of specific output you need by showing an example?

c) I'll think about it. Maybe a parser-friendly output like CSV (different mode) would help you even more?

mskadu commented 8 years ago

Hello there.

Thanks for taking the time to respond. Writing to a file is exactly that I am doing at the moment. However, the technique I am forced to use is akin to screen scraping and has a few challenges.

For instance --once > myfile produces only the header and not the top 10 list. When I manually run the command it takes a few seconds for the top 10 to show up (RHEL 6.x/ Sun JDK 1.6.x). I am in the process of figuring out how to deal with this.

As an afterthought, how about a flag switch to convert the output in machine readable format (--machine-readable or better) so that the output can be parsed with CLI tools like awk, grep or similar.

Using the typical output of ./jvmtop.sh 24227 as reference

  1. JvmTop 0.8.0 alpha - 13:20:26, amd64, 4 cpus, Linux 2.6.32-57, load avg 0.00
  2. http://code.google.com/p/jvmtop
  4. PID 24227: org.apache.catalina.startup.Bootstrap
  5. ARGS: start
  6. VMARGS: -Dcatalina.base=/usr/share/tomcat -Dcatalina.home=/usr/share/tomc[...]
  7. VM: Oracle Corporation Java HotSpot(TM) 64-Bit Server VM 1.8.0_60
  8. UP: 3:36m #THR: 17 #THRPEAK: 21 #THRCREATED: 48 USER: tomcat
  9. GC-Time: 0: 0m #GC-Runs: 45 #TotalLoadedClasses: 2466
  10. CPU: 0.48% GC: 0.00% HEAP: 8m / 418m NONHEAP: 22m / n/a
  12. TID NAME STATE CPU TOTALCPU BLOCKEDBY
  13. 54 RMI TCP Connection(26)-127.0.0 RUNNABLE 1.64% 0.19%
  14. 55 JMX server connection timeout RUNNABLE 0.12% 0.06%
  15. 17 ajp-bio-8009-AsyncTimeout TIMED_WAITING 0.02% 3.55%
  16. 15 http-bio-8080-AsyncTimeout TIMED_WAITING 0.01% 4.20%
  17. 53 JMX server connection timeout TIMED_WAITING 0.00% 0.04%
  18. 52 JMX server connection timeout TIMED_WAITING 0.00% 0.01%
  19. 22 RMI Scheduler(0) TIMED_WAITING 0.00% 0.08%
  20. 20 RMI TCP Accept-0 RUNNABLE 0.00% 0.01%
  21. 19 Attach Listener RUNNABLE 0.00% 0.09%
  22. 16 ajp-bio-8009-Acceptor-0 RUNNABLE 0.00% 0.00%
  23. Note: Only top 10 threads (according cpu load) are shown!

Possible "hide info" switches

That way, someone looking to use command line on jvmtop could pick and chose what info gets generated and how. And then easily use awk/grep/ etc to parse the output.

Just some thoughts. I am improving as I have more chance to think them through a bit better. Please do ask more questions :)

patric-r commented 8 years ago

Hi, the absence of the top 10 threads information when using --once is a known limitation which will be addressed in a future release. Please use issue #58 for tracking that one.

Your suggestion for a machine readable format corresponds somehow to the "c)" section in my previous comment. I am asking myself which format (CSV?) should be used for such a mode - what do you think?

I'll think about the additional --hide-xyz switches - thanks for clarifying this. This should be quite easy to implement.

BTW, just to gain some insights into the real-life usage scenarios of jvmtop ;) How do you actually use jvmtop's output after you parsed its output? Do you use any further tool chains like nagios?

mskadu commented 8 years ago

Re. which format - I would go with CSV for now. It's the simplest and can easily be converted to others downstream by command line utilities. Vagrant also appears to be doing a wait-and-watch, until which they've gone with CSV (see: https://docs.vagrantup.com/v2/cli/machine-readable.html)

Re. --hide-xyz - thanks, that would be very handy.

Re. usage scenario - in my case I am using jmvtop as a light-weight profiler of heap/non-heap and some CPU stats of processes within a JVM. My company has a Java based product that consumes considerable memory and CPU. I was looking for a simple profiler that does not require me to change any JAVA_OPT variables or make changes to my PROD code.

My script is basically a simple UNIX script running in cron and captures the relevant figures and writes them to a file. The data in this "capture" file can then be used for whatever purposes. This is just the first usage - the more info JVMtop can produce, the more useful my script can make of it.

On that topic, how possible is it to add memory related columns to the top-10 field list?

mskadu commented 8 years ago

Hello again - any luck with this?

mskadu commented 7 years ago

Closing this as I have not had a response to this in nearly a year now.