- From: Peter Linss via GitHub <sysbot+gh@w3.org>
- Date: Sat, 03 May 2025 01:47:58 +0000
- To: public-css-archive@w3.org
Sorry, been busy. First some stats, there are currently *two* servers, a modern 'main' server with 4 CPUs, 8GB RAM, 160GB SSD, hosting cost $58/month; and a 'legacy' server with 6 CPUs, 16GB RAM, 320GB SSD, hosting cost $116/month. The legacy server is still around because it's running a bunch of Python 2 and PHP5 custom code that needs to be ported to Python 3 and PHP8 to migrate to a modern server. I currently do not have the time to port this code. The main server is running everything else (including the DBs) and proxying the legacy server. The main server served 3.59TB of data in March, and 4.75TB of data in April. This has been a noticeable increase over the ~2TB/month it averaged in 2024. Most of the traffic is crawlers. The legacy server is typically the one that gets overwhelmed by bots these days due to PHP CPU load. These are not cheap 'toy' severs. The servers are hosting: * Draft servers for CSSWG, FXTF, and Houdini * The hosted Bikhshed service. * The IRC logbot (log.csswg.org) * An IRC bouncer, not clear how many people are still using this. * The CSSWG Wiki (a Dokuwiki instance) * An email server hosting an archive of www-style available via IMAP. * A proxy to the W3CTAG's Nextcloud instance used for calendars (not sure if this has transitioned to the W3C calendar, but I still see updates). If this continues to be used it should probably get its own Nextcloud instance. * Shepherd, which includes the spec parsers that provide (provided?) anchor data to Bikeshed. * The CSSWG test harness and it's API for spec annotations. * The CSS2.1 test suites * Active defenses against bots (and other cybersecurity threats). The hosting company does provide DDOS protection, but **many** bots get past that. * Various server infra, like DNS, LDAP, email, monitoring, admin pages, etc. * Probably something else I'm forgetting The wiki isn't under particularly heavy load, but it's relying on Shepherd's authentication system since Dokuwiki's registration is easily abused by spambots. If Shepherd goes away, this could be transitioned to the draft server's auth system as it's the same as Shepherd's. If we dump the Wiki, someone needs to migrate the data (and it's not going to be me). If we keep the Wiki, I'd like to put [Anubis](https://anubis.techaro.lol/) in front of it (and maybe everything else as well). The draft servers maintain a log of each push to the GH repo (and yes, the Mercurial repo is still there too but no longer web-facing as it was getting crawled a lot too). The last built drafts for each push are available by dated URL and via a list on the index page. When there's a push, it regenerates everything modified (specs, issues lists, and markdown files), emails authors about any Bikeshed error output, then re-parses any updated specs, updates Bikehsed's anchor DB, then regenerates *all the other specs* so that all cross references are up to date. The server also makes the last Bikeshed error output available for each spec. The draft server for the CSSWG is proxying GH pages for the current built drafts. Anything not available on GH (like the index page, historical drafts and Bikeshed output) gets served by the legacy server (via a caching proxy on the main server). For FXTF and Houdini everything is served by the legacy server (via caching proxy on the main server) as there aren't GH pages versions of those draft repos (this could be fixed). -- GitHub Notification of comment by plinss Please view or discuss this issue at https://github.com/w3c/csswg-drafts/issues/12054#issuecomment-2848356189 using your GitHub account -- Sent via github-notify-ml as configured in https://github.com/w3c/github-notify-ml-config
Received on Saturday, 3 May 2025 01:47:59 UTC