Re: Feature requests for execline & s6

From: Olivier Brunel <jjk_at_jjacky.com>
Date: Tue, 27 Jan 2015 00:50:06 +0100

On 01/27/15 00:12, Laurent Bercot wrote:
> On 26/01/2015 23:52, Olivier Brunel wrote:
>> That looks good, except that I need a default value, in case the file
>> doesn't exist. If there's a way to work a default value in that case
>> then that'd be fine, yes.
>> ...is there a way to add a default value if FILE doesn't exist?
>
> I will add a "-D default" option to backtick, for when the subprocess
> exits nonzero. I've wanted this feature for some time, but only now
> realized that I wanted it. Thanks for helping me clarify my mind. :)

My pleasure :)


>> (re: misconfiguration, in what way though? It could lead to the program
>> being triggered for each new line, but that doesn't mean uncontrollably
>> spawning children?
>
> Well, some services log a lot. Spawning a process for every new line
> can be really, really heavy. (And I don't even want to imagine how heavy
> it would be on Solaris.)
> I really don't mind forking and execing, as can be proven by execline;
> but potentially forking every time a daemon outputs a newline is still
> really scary. I need to find the right safety guards before I implement
> that.

I see... so you're thinking e.g. a control directive to set the maximum
amount of time the action can be triggered per minute or something?


>> It might be a lot, but with a rotation every 4096 bytes one might
>> already have frequent calls to the processor, no?)
>
> Sure, with s4096, but who limits their log files to 4k ? It requires
> specific configuration to do that; the admin really wanted it to
> happen, it's controlled. A daemon's output matching a given regexp
> - which could be .* - is a lot less predictable. And if the daemon is
> spewing nothing but newlines, you could find yourself forking 4096 times
> more than you're already doing with the processor...
>
Received on Mon Jan 26 2015 - 23:50:06 UTC