test

        authdb::AuthDB authdb(backend,httpd->getServerSockets());

        // netplus::event's default is just the core count — mediadb boosts

        // the same base class to max(cores*4, 16) (server.cpp) since request

        // handlers can block on cluster round-trips; authdb's session-lookup

        // path (Cluster::fetchSessionBySid) has the same shape, so give it

        // the same margin instead of running with far fewer worker threads

        // than the service it authenticates for.

        authdb.threads = std::max(authdb.threads * 4, 16);

        // Pre-load all domain backends into cache so API requests

        // don't need to lock the admin backend for domain resolution.

        if (backend.end() > sizeof(authdb::AuthHeader))

        if (session_read_client_) {

            try {

                auto data = session_read_client_->retrieve(sid_gid);

                // Bound the whole call — omitting the deadline gives every

                // stripe its own independent ~30s retry budget (see

                // paritypp::client::retrieve()'s doc comment), which can hang

                // an authdb worker thread for a long time while the session

                // cluster is degraded. authdb's HTTP server doesn't boost its

                // worker-thread count the way mediadb's does, so a handful of

                // concurrent misses like that is enough to exhaust the whole

                // pool. Session lookups fail closed (return false below) well

                // before the cluster's own health_monitor_loop would flag it

                // critical and short-circuit this call entirely.

                auto deadline = std::chrono::steady_clock::now() + std::chrono::seconds(5);

                auto data = session_read_client_->retrieve(sid_gid, deadline);

                if (!data.empty()) {

                    int64_t created_at;

                    if (SessionBlock::deserialize(data, sid, uid, did, members, username, gpo_results, created_at)) {