Each stack gets its own ClickHouse (2G limit, system logs off, low-memory config) and the schedule worker consumes scheduler_recorder. Next keeps 30 days of history, production 365. Goes live with the move to the bigger instance.
- Report the front node server to its own Sentry project (SENTRY_SERVER_DSN)
- Replace trace/session rates with session and error replay rates
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Split cojedzie_version into front/api/worker versions so CI can pin each by digest
- Replace the cron service with a worker consuming the scheduler queues
- Pull images from registry.kadet.net, fall back to latest on full runs
- Drop pinned serverVersion from DATABASE_URL
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
