b. Docker Compose

#!/bin/bash
# Entrypoint of every node of the cluster.
#   PRIMARY_HOST unset -> primary: the official entrypoint creates the database
#                         and runs /docker-entrypoint-initdb.d (none here: see below).
#   PRIMARY_HOST set   -> replica: on the first start, clones the primary with
#                         pg_basebackup and starts as a hot standby.
set -euo pipefail

PGDATA=${PGDATA:-/var/lib/postgresql/data}

if [ -s "$PGDATA/PG_VERSION" ]; then
  echo "node: data directory already initialised, starting PostgreSQL"

elif [ -z "${PRIMARY_HOST:-}" ]; then
  echo "node: empty data directory, initialising a new primary"
  # Runs once, right after initdb: the replication role and its pg_hba rule.
  cp /scripts/init-primary.sh /docker-entrypoint-initdb.d/10-init-primary.sh

else
  echo "node: empty data directory, cloning $PRIMARY_HOST"
  until pg_isready -h "$PRIMARY_HOST" -U "$POSTGRES_USER" -q; do
    echo "node: waiting for $PRIMARY_HOST..."
    sleep 1
  done
  mkdir -p "$PGDATA"
  chown postgres:postgres "$PGDATA"
  chmod 700 "$PGDATA"
  # -X stream: also copy the WAL written during the backup
  # -R: write standby.signal + primary_conninfo, so the node starts as a standby
  PGPASSWORD="$REPLICATION_PASSWORD" gosu postgres \
    pg_basebackup -h "$PRIMARY_HOST" -U "$REPLICATION_USER" \
                  -D "$PGDATA" -X stream -R --progress
fi

exec docker-entrypoint.sh "$@"

#!/bin/bash
# Runs once, when the primary creates its database (docker-entrypoint-initdb.d).
# Everything here lives in the data directory, so pg_basebackup copies it to every replica.
set -euo pipefail

psql -v ON_ERROR_STOP=1 --username "$POSTGRES_USER" --dbname "$POSTGRES_DB" <<SQL
-- A role that may only stream WAL: replicas connect with it, never the application.
CREATE ROLE $REPLICATION_USER WITH REPLICATION LOGIN PASSWORD '$REPLICATION_PASSWORD';
-- WAL kept on disk for a replica that falls behind (or is stopped for a while).
ALTER SYSTEM SET wal_keep_size = '256MB';
-- Lets pg_rewind turn an old primary into a replica after a failover.
ALTER SYSTEM SET wal_log_hints = on;
SQL

# "all" in pg_hba.conf does not match replication connections: they need their own rule.
echo "host replication $REPLICATION_USER all scram-sha-256" >> "$PGDATA/pg_hba.conf"

#!/bin/bash
# Turns a stopped old primary into a replica of the new primary.
# Usage (from the folder of compose.yaml, with the node stopped):
#   docker compose run --rm --no-deps --entrypoint bash pg-primary \
#     /scripts/rejoin-as-replica.sh pg-replica1
set -euo pipefail

NEW_PRIMARY=${1:?usage: rejoin-as-replica.sh <new-primary-host>}
PGDATA=${PGDATA:-/var/lib/postgresql/data}

# pg_rewind copies back only the blocks that changed since the timelines split,
# discarding the writes the old primary made that never reached the new one.
gosu postgres pg_rewind --target-pgdata="$PGDATA" --progress \
  --source-server="host=$NEW_PRIMARY user=$POSTGRES_USER password=$POSTGRES_PASSWORD dbname=$POSTGRES_DB"

# Same as pg_basebackup -R: start as a standby, streaming from the new primary.
gosu postgres touch "$PGDATA/standby.signal"
echo "primary_conninfo = 'host=$NEW_PRIMARY user=$REPLICATION_USER password=$REPLICATION_PASSWORD'" \
  | gosu postgres tee -a "$PGDATA/postgresql.auto.conf" > /dev/null
echo "rejoin: $(hostname) will follow $NEW_PRIMARY on its next start"