select() and pipes

select() and pipes

[William N. Zanatta]

   Hi,

   I'm writing a tiny program which gets its input from a pipe. It works 
with apache like in this example...

------ httpd.conf-------

   CustomLog "|/path/my/bin"

------ httpd.conf-------

   On apache2 it works pretty well but on 1.3 series I'm experiencing 
some problemas and one of them is the CPU Load. The program's main() 
used to live in a infinite loop making a fgets() and checking for data 
availability on the returned buffer.

   Well, guessing this is the main reason for the CPU Load, I started 
with select() to check for data availability and I've found that the 
couple select() and the pipe as stdin will always return a ready to read 
status even if there's no data to be read.

   How does it work internally? How could I minimize the cpu load? 
Should  I use named pipes instead and tell apache to throw logs there?

   Thanks,

william

Re: select() and pipes

[Glynn Clements]

William N. Zanatta wrote:

>    I'm writing a tiny program which gets its input from a pipe. It works 
> with apache like in this example...
> 
> ------ httpd.conf-------
> 
>    CustomLog "|/path/my/bin"
> 
> ------ httpd.conf-------
> 
>    On apache2 it works pretty well but on 1.3 series I'm experiencing 
> some problemas and one of them is the CPU Load. The program's main() 
> used to live in a infinite loop making a fgets() and checking for data 
> availability on the returned buffer.
> 
>    Well, guessing this is the main reason for the CPU Load, I started 
> with select() to check for data availability and I've found that the 
> couple select() and the pipe as stdin will always return a ready to read 
> status even if there's no data to be read.

How exactly are you using select()? Are you initialising the fd_set
correctly? Are you specifying a timeout?

Is there any reason why you can't just perform blocking reads?

-- 
Glynn Clements <glynn.clements@virgin.net>