Re:DTrace Toolkit
by matt_heimer on Sunday March 03, 2013 @12:15AM
(#5312643)
Yeah, most of his stuff is pretty great but always read them before making the assumption that they are right. It is very easy to think everything is perfect if the output seems reasonable. Take Java method calls, the scripts that track method calls don't take into consideration method overloading at all. For an example of what you need to do take a look at http://www.osdev.org/dtrace/hotspot_extras/j_calltime2.d
[ Parent
| on HN ]
Re:DTrace Toolkit
by brendangregg on Sunday March 03, 2013 @01:25AM
(#5312757)
Thanks for fixing it, and sorry for the bug.
Your advice is good - and is the case for all observability tools. Bugs happen, even in older tools like vmstat(1), top(1), etc. I try to test every tool I use (including those I develop) with "known workloads", and make sure the numbers match, or understand why they don't.
An advantage of the DTrace-based tools is that they are usually short scripts that you can understand easily and also tweak. Don't like strace on Linux? That's non-trivial to understand and to then change yourself. Don't like dtruss on Mac OS X? It's a DTrace script you can make a copy of and edit (as people have done). dtruss originated from the DTraceToolkit...
Re:DTrace Toolkit
by matt_heimer on Sunday March 03, 2013 @02:43AM
(#5312887)
Less of a bug and more of an underestimation on how out of control method overloading is in Java :)
Thanks for the great tools!