The scraper cached a single RedisCache (one multiplexed connection) in a
OnceCell forever. When that one connection broke (Redis restart, idle
timeout, network blip), every cache op failed with 'cache io: broken pipe',
and because Cache::get_or_set propagated the post-compute write error,
EVERY API (anime, anime2, komik) returned 500 until a process restart.
Fix:
- Build a fresh RedisCache from a freshly checked-out deadpool connection
per call, so a broken connection self-heals without a process restart
(deadpool recycles/drops dead connections and reconnects on checkout).
- Make Cache::get_or_set treat a cache-write failure as non-fatal: return
the freshly computed value (cache is best-effort), so a transient Redis
outage degrades to cache-less instead of 500.
Pre-existing clippy warnings (repositories/parsers) untouched — out of scope.
mytheclipse-queue and mytheclipse-tracing are now published on crates.io
(v1.21.2). Switch both from local path deps to version deps so the whole
mytheclipse family resolves from crates.io — no more path deps in the
dependency graph.
The user published all mytheclipse crates to crates.io at 1.21.2
(including the mytheclipse-cache redis-0.32 fix). Switch the scraper to
version deps for the released crates (mytheclipse, -cache, -config,
-event) so it tracks the published releases instead of local path deps.
mytheclipse-queue + mytheclipse-tracing are not yet on crates.io, so they
remain path deps to the local workspace.
asepharyana
asepharyana