Repository navigation
Cache aware verification
The most expensive sentence in this whole pattern is: "I deployed the fix and nothing changed." Nine times out of ten, the fix is live and four cache layers are each showing you a different decade.
- Your own browser. Hard-refresh (Cmd/Ctrl+Shift+R) or test in a private window before you form any hypothesis at all.
-
Server page cache (LiteSpeed, FastCGI cache, WP caches). Full-page HTML is
cached for minutes by default. Cache-bust every check: append
?cb=1783123456(any changing value) to the URL. -
opcache / bytecode cache. PHP files you just wrote may keep serving the old
bytecode until revalidation kicks in.
opcache.revalidate_freqof 60 means up to a minute of stale logic on a perfect deploy. If you control it,opcache_reset()via a one-shot worker; if not, wait out the TTL before declaring failure. - CDN / edge. Purge it, or verify against the origin directly.
Verify the new behavior, not the absence of the old one. "The banner is gone"
fails when the banner was already gone for other reasons. "The page contains
data-djc-fixbanner1" cannot pass on a stale file. Your worker should already be
printing byte counts and marker checks - read them:
OK fixbanner1: .legacy-banner hidden
bytes: 269602 -> 269692
done - worker self-deleted
Then confirm the marker survives one more cache-busted fetch from the public URL. If the worker says patched but the public page disagrees, you have a cache problem, not a code problem - go find which layer, and do not write a second patch while the first one is still in flight to production.
Front-end bugs deserve a clean reproduction before a patch: fresh headless browser, no cache, no extensions. A bug that refuses to reproduce headlessly is usually stale cache, a logged-in state, or an extension - not your code. Patching a ghost is how you get two bugs.
Guarded Deploy Workers
Repo