2026.10.61 — 2026-10-27

Fixed

• Start on an empty database without an error in either log (7d8cd24d)

  On a first start Rahdar read its deploy-time properties from rahdar_config_property before the migrations had created it. Nothing failed, but PostgreSQL logged ERROR: relation "rahdar_config_property" does not exist and Rahdar logged a warning, on every replica's first boot, which reads like a broken installation. Rahdar now looks the table up first and, when it is not there yet, says so at info level and goes on with the packaged defaults. An unreachable database is still a warning.

• Start the stacks from a clone made under a strict umask (86df9db6)

  Postgres, HAProxy and the brokers read the files the compose stacks mount (postgresql.conf, the pg_stat_statements script, haproxy.cfg, broker.xml) as their own users, not as whoever cloned the repository. A clone made under umask 077 leaves those files owner-only: Postgres stops with "could not open configuration file ... Permission denied", HAProxy refuses its configuration, and every replica's database calls fail while nothing answers on port 8080.

  just up, up-single, dev-up, demo, qa-up, load and ha-failover now make exactly those files world-readable before they start a stack; deploy/compose/.env, which holds secrets, is left alone. Running compose by hand from such a clone, run chmod a+r deploy/postgres/* deploy/compose/ha/haproxy.cfg first (the quickstart and the high-availability guide say so).