Re: general questions on Frysk

["Naoya Maruyama" <naoya dot maruyama at gmail dot com>]

Hello,

Thank you for the reply. I'm especially interested in how Frysk will
be extended to monitoring of distributed systems. I'll come to this
project later.

Naoya

On 3/18/06, Andrew Cagney <cagney@redhat.com> wrote:
> Naoya Maruyama wrote:
>
> >Hello,
> >
> >I'm new to Frysk, and am wondering how it works. I've read the Frysk
> >web site, but yet
> >
> >
> >
> Naoya, good questions,
>
> >- How does Frysk obtain events from other processes? What does the
> >interface look like? Is it the same as the traditional debugger using
> >ptrace?
> >
> >
> At present frysk is using the kernel ptrace to obtain the event
> information; really, though, it isn't sufficent.  Some kernel people are
> also looking at providing a more modern interface and when ever that
> becomes available we'll be all over it :-)
>
> >- How does Frysk monitor distributed systems?
> >
> >
> That's unfortunatly a work not really in progress - development is
> immediatly focused on getting the local native case working.  Frysk's
> core architecture, however, is designed to allow distribution, see:
> http://sourceware.org/frysk/javadoc/public/frysk/model/doc-files/arch-distrib.jpg
> for a high level diagram.  It has been suggested that a prototype can be
> quickly brought up using Java's distributed object framework.
>
> >- How much would the overhead be?
> >
> >
> We'll need to find out, and explore the options.  The important thing is
> that the communication does not need to be low level level, but instead
> communicatioin can occure at a higher-level and involve less frequent
> exchanges exchanges.  For instance, limited to just specific events.
>
> >- Are there any documents describing the internals of Frysk?
> >
> The page:
>
> http://sourceware.org/frysk/javadoc/public/
>
> gives a very high level overview.  Bbeyond that when it comes to more
> specific details the best thing is to ask either on this mailing list,
> or the irc:gimp.net/frysk channel.
>
> A good, but very very low level, way to see the core in action is to run
> its JUnit tests with logging enabled (/usr/libexec/frysk/funit -l FINE
> or cd frysk-core && ./TestRunner -l FINE).  A file in ~/.frysk/logs will
> be created containing a log all the object|event interactions; for instance:
>
> {frysk.proc.LinuxTask@445370,id={TaskId,13106},state=attaching}
> receiveStoppedEvent
> {frysk.proc.LinuxTask@445370,id={TaskId,13106},state=attaching}
> processStoppedEvent
> {frysk.proc.LinuxProc@49bdc0,id={ProcId,13106},state=Attaching.ToMainTask}
> performTaskAttachCompleted
>
> is a very small part of the event sequence that occures when frysk is
> attaching to the Task (Linux LW) 13106.  The corresponding code is in
> frysk-core/frysk/proc/TaskState.java.  The details of the state machine
> pattern, on which that code is based, is found in the <<group of four>>
> patterns book.
>
> >Any help would be appreciated.
> >
> >
> Andrew
>
>