Re: [PATCH 17/17] libtracefs: Add man page for tracefs_sql()

[Steven Rostedt]

On Sun, 1 Aug 2021 15:39:25 +0200
"Ahmed S. Darwish" <a.darwish@xxxxxxxxxxxxx> wrote:

> On Fri, Jul 30, 2021, Steven Rostedt wrote:
> > +
> > +The SQL format is as follows:
> > +
> > +*SELECT* <fields> FROM <start-event> JOIN <end-event> ON <matching-fields> WHERE <filter>
> > +
> > +Note, although the examples show the SQL commands in uppercase, they are not required to
> > +be so. That is, you can use "SELECT" or "select" or "sElEct".
> > +  
> 
> Maybe it would be helpful to mention that, unlike normal SELECT queries,
> the JOIN and ON parts above are _not_ optional?
> 
> That is, generic "one event source" queries:
> 
>   SELECT common_pid,msr,val FROM write_msr WHERE msr=72
> 
> are not supported. (I wish they were though ;-))

Actually, the sql parser should support it, but it will fail on the
creation of events. That's because I started trying to make this create
normal histograms. The problem is, that it can't really do a 1 to 1 on
histograms and selects, so I gave up. But perhaps for the subset it can
create, maybe I can still have it do so. That may require changing the
API slightly. 

I'm not a big SQL person, so I don't know all the magic and I have no
idea how to add the "values" part of the hist trigger.

I could use SQL CAST to let people redefine the types that histograms
allow, like:

   SELECT CAST(common_pid AS execname), CAST(id AS syscall) FROM sys_enter

and have that produce:

 echo 'hist:keys=common_pid.execname,id.syscall' > events/raw_syscalls/sys_enter/trigger

Any ideas for the syntax to get values? Or do people care about the
value parameter any more than just "hitcount"? I don't think I have
ever used it for anything.

> 
> BTW, thanks a lot for this work. It will finally make synthetic events
> usable!
> 

Thanks! It's also something to make it easier for me to use. And
hopefully, allow for even more complex synthetic events.

-- Steve