CentOS mirrorlist service
Note
the mirrorlist.centos.org is crucial for all deployed CentOs instances all around the world as each deployed CentOS instance will query the mirrorlist webservice to get a list of validated and up2date mirrors to retrieve their content from. It's using GeoIP or checking if coming from a cloud provide (like EC2), in which case it would redirect to the nearest (GeoIP) or internal (Cloudfront setup for AWS/EC2) mirror
Overview
It contains the following kind of scripts:
- backend : so scripts used by our "crawler" node, validating in loop all the external mirrors through IPv4 and IPv6 and so producing the 'mirrorlists', each one per repo/arch/country
- frontend : python scripts used for :
Backend (crawler)
There are two Perl scripts for checking mirrors:
- makemirrorlists-combined.pl for creating files for mirrorlist.centos.org
- makeisolists-combined.pl for creating files for isoredirect.centos.org.
Both scripts can create lists for all CentOS supported released ,including SIG and AltArch content. makemirrorlists-combined.pl will test each mirror separately for IPv4 and IPv6.
mirrorlist.centos.org will then be able to present only IPv6-capable mirrors to the clients when mirrorlist.centos.org is accessed over IPv6. More details about the internals of these scripts can be found in backend/mirrorlist_crawler_deployment_notes.txt
Frontend
All scripts are located in the frontend folder. The following items are needed for the mirrorlist/isoredirect service:
- A http server (apache) using mod_proxy_balancer (see frontend/httpd/mirrorlist.conf vhost example)
- python-bottle to run the {ml,isoredirect}.py code for various instances
- Maxmind Geolite2 database : City version
- python-geoip2 pkg (to consume those Geolite2 DB)
- python-memcached (to cache results for GeoIP/Cloud providers)
- For each worker, a specific instance/port can be initialized and added to Apache config for the proxy-balancer (see frontend/systemd/centos-ml-worker@.service)
Those services (mirrorlist/isoredirect) just consume mirrorlist files, pushed to those nodes, and updated in loop by the Crawler process (see Backend section above)
When a request is made to the service, the python script :
- checks for IPv4 or IPv6 connectivity
- checks if IP is in memcached (for country/cloud provider)
- searches if IP is from cloud provider
- computes Geolocation based on the origin IP
- searches for validated mirrors in the same country/state for the request arch/repo/release
- returns such list