Use $HOSTNAME in restore.sh instead of the hostname binary

[tamalsaha]

scripts/restore.sh is the only script in this repo that shells out to the hostname binary; every other one reads bash's $HOSTNAME.

hostname is not installed in the RHEL/CentOS Stream based images that run this script — neither ghcr.io/appscode-images/postgres-enterprise:16.9-stream9-ext nor the ghcr.io/kubedb/postgres-archiver:*_16.9-stream9* image that actually executes it (Walg.Image, wired up in kubedb/postgres pkg/controller/sidekick.go). There is no hostname package in either; command -v hostname comes back empty in both.

The failure

HOST=$(hostname)

The script has no set -e, so it carries on with HOST empty. Every later branch keys off it:

if [[ "$HOST" =~ -0$ ]]; then
  echo "recovery_target_action = 'promote'" >>/tmp/postgresql.conf
else
  echo "recovery_target_action = 'pause'" >>/tmp/postgresql.conf
fi

With HOST empty the test is false on every pod, so ordinal 0 is configured to pause instead of promote. The script then takes the replica path at the bottom and loops forever on pg_is_wal_replay_paused() waiting for a node that was supposed to promote — a PITR restore that never completes.

The fix

Use $HOSTNAME, which bash sets itself, matching what config_recovery.conf.sh, ha_backup_job.sh, warm_stanby.sh, start.sh and the rest of the repo already do. No os-release branching needed — this works identically on alpine, bookworm and stream9.

Verification

HOST="$HOSTNAME" in a #!/bin/bash script, run with --hostname my-pg-0 and --hostname my-pg-2:

=== ghcr.io/kubedb/postgres-archiver:v0.28.0-rc.2_16.9-stream9_linux_amd64 ===
HOST=[my-pg-0] -> primary path
HOST=[my-pg-2] -> replica path
=== ghcr.io/appscode-images/postgres-enterprise:16.9-stream9-ext ===
HOST=[my-pg-0] -> primary path
HOST=[my-pg-2] -> replica path

bash -n scripts/restore.sh is clean.