deb

authdb (20260712+5) unstable; urgency=high

  * Cluster::fetchSessionBySid(): pass an explicit 5s deadline to

    paritypp::client::retrieve() instead of the default (none) — omitting

    it gives every stripe its own independent ~30s retry budget with no

    overall cap, the same unbounded-retry shape that caused mediadb's

    20260711+32 incident. Session-cache misses go through this on any

    authenticated request across every service using authdb, and unlike

    mediadb's HTTP server (server.cpp boosts worker threads to

    max(cores*4, 16)), authdb never boosted its own worker-thread count —

    so a handful of concurrent misses during a degraded-but-not-yet-

    critical window was enough to tie up its entire pool. Session lookups

    now fail closed within 5s instead of hanging.

  * main(): boost authdb.threads to max(cores*4, 16), matching mediadb's

    server.cpp margin, instead of netplus::event's un-overridden default

    of just the core count. Directly implicated in a production incident

    where authdb stopped responding alongside mediadb — a service that

    authenticates requests for other services shouldn't run with less

    request-handling headroom than the busiest of them.

 -- Jan Koester <jan.koester@tuxist.de>  Sat, 18 Jul 2026 06:53:29 +0200

authdb (20260712+4) unstable; urgency=high

  * ClusterBackend::fetchFromCluster(): only skip the real network fetch