Hello,
On 03/03/2017 07:20 AM, Antonio Diaz Diaz wrote: > Assaf Gordon wrote: >> Download statistics for gnu's ftp-mirror: >> https://download.savannah.gnu.org/ftpmirror-stats/ > > Does this work if the user goes to http://ftpmirror.gnu.org/ , is > redirected to a mirror, and then navigates the mirror to find the file > he wants to download? Or it only works if the file is accessed trough a > full link like http://ftpmirror.gnu.org/ed/ed-1.14.2.tar.lz ? > [...] > But the funniest thing is that it can apparently invent files that do no > exist (ddrescue is released in lz format only): Actually, it counts exactly one thing: HTTP 'GET' requests to 'ftpmirror.gnu.org'. The first subdirectory in the URL is taken as the 'package name'. It does not track any other mirror usage, nor direct access to ftp.gnu.org. It also does not verify the file actually exists. And neither does it separates bots from real people (based on user-agent or otherwise). Any request is counted. Over time, I assume incorrect requests will be drowned-out as noise compared to real requests. Though they can serve as useful flags to know if some runaway script somewhere on the internet has a wrong URL in it. > I have noticed that files can be counted in more than one place. > ed-1.14.2.tar.lz 1 > [...] > gnu > ed/ed-1.14.2.tar.lz 6 Again, this script counts requests, so these are two different request URLs. The confusion/duplication stems from the fact the for 'ftp.gnu.org', the "/gnu/" subdirectory is required, e.g.: http://ftp.gnu.org/gnu/ed/ed-1.14.2.tar.lz #works http://ftp.gnu.org/ed/ed-1.14.2.tar.lz #does not exist But mirrors do not always have the '/gnu/' sub directory (e.g. "http://gnumirror.nkn.in"; is one of the mirrors, no "/gnu/" sub directory). The correct mirroring URL should not have "/gnu/": http://ftpmirror.gnu.org/ed/ed-1.14.2.tar.lz However, I assume a long time ago someone realized this is too confusing, and added an apache rewrite rule to also support (and strip) a "/gnu/" subdirectory for ftpmirror.gnu.org . We kept this 'rewrite' rule when we moved to the new VMs recently ( in vcs0:/etc/nginx/sites-available/ftpmirror-common.inc , for future reference). So everything 'just works' with and without '/gnu/'. But this is an indication that somewhere, someone has constructed their own mirror URL, and it's incorrect. regards, - assaf
