Comment 28 for bug 701329

On 5 February 2011 10:18, John Arbash Meinel <email address hidden> wrote:
> Note that when I was testing it, you can get the failure while writing
> headers, regardless of rendering the rest of the page. At least:
>
>  echo "HEAD /... HTTP/1.0" | telnet localhost 8080
>
> can get the exception to trigger very early.
>
> As for why should we render the page even for just a HEAD request... If
> HEAD requests are only happening at the haproxy level, why not? Is it
> worth spending developer time just to handle this case?
>
> Arguably there is the proxy situation, where a given proxy could say
> "try HEAD and see if I can just use my cached value". I didn't think
> loggerhead was really ready for that, because of needing ETAGS support, etc.
>
> I don't doubt that it is a waste of CPU, but if <1% of our time is spent
> in HEAD requests, it doesn't seem worthwhile to spend developer time to
> fix it. (We may need to have better logs to determine what the specific
> overhead is.)

haproxy sends a request every 10s, so if they're doing unnecessary
work it's potentially going to create a noticeable load. It's a bit
surprising too. But it's not really essential to fixing this bug: we
could file a separate issue saying we noticed this, and it can wait
until we want to reduce the load on that machine.