Re: pmlogger -u questions

[Ken McDonell <kenj@xxxxxxxxxxxxxxxx>]

On 15/04/14 09:18, Frank Ch. Eigler wrote:
> Hi -
>
> On Tue, Apr 15, 2014 at 08:46:09AM +1000, Ken McDonell wrote:
> > [...]
> > Having actually made the change now, I like this even more ... in the
> > simplest implementation it is a one line change in __pmLogNewFile() to add
> >      setvbuf(f, NULL, _IONBF, 0);
> We should do a round of analysis (strace or stap or whatever) to see
> how the change would affect i/o syscall traffic, just to confirm that
> we write records in nice big chunks.  (If not, another possibility
> would be to adjust the stdio buffering dynamically, based on the
> actual size of the logger data records.)
Yep, already doing that.

Before with default config on my workstation ... every 60 secs I see 2 
writes of 4096 and/or 8192 bytes (can't explain why as yet)/
After with same config ... every 60 secs I see 2 writes of 11244 and 4 
bytes (the latter is the trailer that I am confident I can force into 
the first write ... but this is more than a 1 line change as the 
semantics of __pmLogPutResult have to change in a subtle way).