Re: squid3-largeobj squid3/src HttpHdrRange.cc...

Hi,

Alex Rousskov wrote:
>
> If a unitless configuration is rejected, there is no problem. If it is
> accepted (with a warning) and then interpreted differently from Squid2,
> I think there is a problem because many admins will [continue to] ignore
> the warning.

The configuration line "quick_abort_max 16"
in squid2 will produce a warning like:
  WARNING: No units on 'quick_abort_max 16', assuming 16 KBytes

and in squid3 the warning will be:
  WARNING: No units on 'quick_abort_max 16', assuming 16 bytes

The administrator will see it.
Because the quick_abort_max/quick_abort_min are not critical I am not
believe that it is so important.

>
> Can you make the units required in this particular case?
No.. yes ... but I believe that it will not be good because squid3 must
have the same policy for all parameters expressed in bytes.

The original talk is about the need of eliminating uses of parameters in
the form of "Config.quickAbort.min << 10". To use
Config.quickAbort.min/max you must always remember to multiply it with
1024. The danger of overflow can fixed using 64bit integers in these
parameters.
Because the Config.quickAbort.max/min is used only in 1-2 places inside
the squid code is not big problem to keep the current behavior. But, I
agree that it can confuse developers....

>
> If there is consensus that units should be required everywhere, I am
> fine with that (but still worried).
>

Most of administrators tend to just uncomment the related lines in
configuration file and just change the default values. Because the
default config file always include units I believe that only a small
number of installation will have problem with it.

Regards,
     Christos
Received on Thu Aug 02 2007 - 01:37:32 MDT