Re: EAGAIN in "read jobs pipe"

[James Coleman]

Paul Smith wrote:
> On Tue, 2007-01-09 at 10:32 +0000, James Coleman wrote:
> > Ken Takusagawa wrote:
> > > For reasons I haven't completely triaged yet, I've been having my
> > > parallel makes die with
> > > "read jobs pipe: Resource temporarily unavailable.  Stop."
> > All memory on machine might be used up (real memory + all swap).
>
> I don't think this can be the case.  In a standard POSIX system a buffer
> is allocated to the pipe when it's created.  There's no need to get more
> memory when writing to or reading from the pipe.  Further, GNU make
> simply writes one byte to the pipe for each possible job, so as long as
> you're not using -j65536 or something silly like that it's unlikely to
> be a problem.


Yes I'm not sure about the read jobs pipe error.

The error I have seen is due to gcc running out of memory. I thought it might 
be somewhat related.
We have some .h and .cpp files generated for ASN1 encode/decode and they are 
huge and they result in
a very large gcc process.


:) -j65536

We have sun machines here with 4 multiple cores & 8 virtual cores per core 
(coolthreads). solaris 10
They have their own limitations but you can gmake -j<ridiculous> on them and 
build is much faster
and machine stays responsive (I presume solaris 10 kernel is managing things 
better than before).
http://www.sun.com/servers/coolthreads/t2000/

Thought you might like to know that.


Noticed something else.

Something else completely pointless!

<ridiculous> must not be > 16385

 gmake -j 16386; # or above hang
 gmake -j 16385; # or below fire along rapidly

GNU Make 3.80
SunOS xxxxxxxx 5.10 Generic_118833-03 sun4v sparc SUNW,Sun-Fire-T200
Sun Fire T200 (32 X SPARC 1000MHz)
slot 1 has a 8184MB (7.9921875GB) DIMM


I'll go back to my corner and be quiet now.

James.