The published entrypoint downloads the 22MB clickhouse-backup binary from GitHub at container boot
with 'set -ex' + a single silenced no-retry wget to ephemeral /tmp. Any transient failure of that
download (rate-limit / network) exits the container BEFORE clickhouse-server starts, so swarm restarts
it, it re-downloads, and the throttle is amplified into a crash-loop (deploy timeout).
clickhouse-backup is the BACKUP tool (backupbot pre/post hooks), not required for clickhouse-server to
run. This hardening caches the binary on the persistent /var/lib/clickhouse volume (fetched at most
once, reused on restart), retries with backoff, never blocks the server start on a fetch failure, and
un-silences the wget for diagnosability. No behaviour change when the first download succeeds.
