27 lines
1.4 KiB
Ruby
27 lines
1.4 KiB
Ruby
module DataUpdateScripts
|
|
class GradualArticleCacheBust
|
|
def run
|
|
# This script is designed to bust the cache on article pages sufficient such that significant changes to
|
|
# CSS will not create meaningful visual regressions for most users.
|
|
# All articles have their caches naturally recycle once per day to finalize the transition.
|
|
|
|
# This uses exponential backoff in order to bust the caches of more recent articles first, but not
|
|
# bust everything in too short a period, so that the edge cache can gradually re-heat without everything
|
|
# going cold at once. Articles with high "hotness score" (e.g. more recent) are the ones most in need of busting.
|
|
|
|
# It only busts the "?i=i" variant of the path which is the "internal nav" version, aka the page when swapped
|
|
# with internal navigation, not the one landed on directly (which wouldn't have cache mismatches)
|
|
relation = Article.published.order(hotness_score: :desc).select(:path)
|
|
|
|
relation.limit(1500).each_with_index do |article, index|
|
|
n = index + 300 # + 300 gives the server time to boot up
|
|
BustCachePathWorker.set(queue: :high_priority).perform_in(n.seconds, "#{article.path}?i=i")
|
|
end
|
|
|
|
relation.offset(1500).limit(3000).each_with_index do |article, index|
|
|
n = (index * 3) + 450
|
|
BustCachePathWorker.set(queue: :medium_priority).perform_in(n.seconds, "#{article.path}?i=i")
|
|
end
|
|
end
|
|
end
|
|
end
|