fix: override bootstrap.post_init via SPILO_CONFIGURATION to create missing 'postgres' role before Spilo's real post_init.sh runs. Spilo hardcodes ALTER VIEW...OWNER TO postgres with no way to parameterize, which fails since our superuser is PGadmin not postgres.

postgresql/cutover/test/pgha-dryrun.yaml

@@ -16,8 +16,11 @@
 # NOTE: command: blocks use $$(...) not $(...) — Compose's own variable
 # interpolation parses $( as an attempted ${VAR} reference and fails with
 # "invalid interpolation format" / "you may need to escape any $ with
-# another $". $$ escapes to a literal $ for the shell at runtime. Same
+# another $". $$ escapes to a literal $ for the shell at runtime. This
-# reasoning applies to $$PGDATA below.
+# applies uniformly to EVERY literal $ character anywhere in a command:
+# block, including inside heredocs and nested quoting — Compose scans the
+# raw text before any shell/heredoc logic ever runs, so quoting context
+# doesn't exempt anything from this escaping requirement.
 #
 # NOTE: ETCD3_HOSTS not ETCD_HOSTS — confirmed against zalando/spilo
 # configure_spilo.py: PATRONI_DCS includes both "etcd" (legacy v2 API,
@@ -28,18 +31,28 @@
 # v3-API client.
 #
 # NOTE: "legacy" needs a pg_hba.conf rule permitting REPLICATION-type
-# connections, which is a distinct connection class from normal client
+# connections (distinct from normal client connections), fixed via a
-# connections in Postgres. The vanilla postgres:17 image's default
+# /docker-entrypoint-initdb.d/ hook script (official extension point).
-# pg_hba.conf allows normal client connections (pg_dumpall worked earlier)
+# "trust" is acceptable ONLY because this container is fully disposable.
-# but NOT replication connections, which pg_basebackup (used by Patroni's
+# See note "ADR-0001 Addendum — pg_hba.conf replication prerequisite
-# CLONE_WITH_BASEBACKUP) requires. Fixed via a /docker-entrypoint-initdb.d/
+# discovered in dry run" for the real production prerequisite this exposes.
-# hook script (the officially supported extension point, runs once right
+#
-# after initdb, before the server's final restart) that appends a
+# NOTE: patroni-0/patroni-1 override bootstrap.post_init via
-# permissive "trust" rule for replication connections. "trust" is
+# SPILO_CONFIGURATION (Spilo's documented, supported mechanism for
-# acceptable ONLY because this container is fully disposable and exists
+# overriding any generated Patroni config — configure_spilo.py deep-merges
-# solely to validate cutover mechanics, not auth. See the note
+# user-supplied SPILO_CONFIGURATION on top of its own generated config).
-# "ADR-0001 Addendum — pg_hba.conf replication prerequisite discovered in
+# This is needed because Spilo's own /scripts/post_init.sh hardcodes
-# dry run" for the real production prerequisite this exposes.
+# "ALTER VIEW ... OWNER TO postgres" with no way to parameterize that role
+# name via env vars. Since our superuser is PGUSER_SUPERUSER=PGadmin (to
+# match production), there is no role literally named "postgres" in the
+# cloned data, so Spilo's unmodified script fails with
+# 'ERROR: role "postgres" does not exist'. The override points
+# bootstrap.post_init at our own wrapper script instead, which creates a
+# harmless, idempotent, NOLOGIN "postgres" role first, then execs Spilo's
+# real, UNMODIFIED post_init.sh with all original arguments passed through
+# — Zalando's script itself is never patched or forked. This affects real
+# production too (superuser there is also PGadmin, not postgres) and must
+# be carried into the real staging/production compose files as well.
 # ─────────────────────────────────────────────────────────────────────────
 version: "3.6"
 
@@ -156,6 +169,14 @@ services:
         export PGPASSWORD_STANDBY="$$(cat /run/secrets/postgresql_replication_password)"
         export PATRONI_RESTAPI_PASSWORD="$$(cat /run/secrets/postgresql_patroni_password)"
         export CLONE_PASSWORD="$$(cat /run/secrets/postgresql_password)"
+        mkdir -p /scripts
+        cat > /scripts/post_init_wrapper.sh <<'WRAP'
+        #!/bin/bash
+        set -e
+        psql -d "$$2" -v ON_ERROR_STOP=1 -c 'DO $$do$$ BEGIN IF NOT EXISTS (SELECT 1 FROM pg_catalog.pg_roles WHERE rolname = $$x$$postgres$$x$$) THEN CREATE ROLE postgres; END IF; END $$do$$;'
+        exec /scripts/post_init.sh "$$@"
+        WRAP
+        chmod +x /scripts/post_init_wrapper.sh
         exec /bin/sh /launch.sh init
     environment:
       SCOPE: pgha-test
@@ -170,6 +191,9 @@ services:
       CLONE_HOST: postgresql
       CLONE_PORT: "5432"
       CLONE_USER: PGadmin
+      SPILO_CONFIGURATION: |
+        bootstrap:
+          post_init: /scripts/post_init_wrapper.sh "zalandos"
     secrets:
       - postgresql_password
       - postgresql_replication_password
@@ -194,6 +218,14 @@ services:
         export PGPASSWORD_STANDBY="$$(cat /run/secrets/postgresql_replication_password)"
         export PATRONI_RESTAPI_PASSWORD="$$(cat /run/secrets/postgresql_patroni_password)"
         export CLONE_PASSWORD="$$(cat /run/secrets/postgresql_password)"
+        mkdir -p /scripts
+        cat > /scripts/post_init_wrapper.sh <<'WRAP'
+        #!/bin/bash
+        set -e
+        psql -d "$$2" -v ON_ERROR_STOP=1 -c 'DO $$do$$ BEGIN IF NOT EXISTS (SELECT 1 FROM pg_catalog.pg_roles WHERE rolname = $$x$$postgres$$x$$) THEN CREATE ROLE postgres; END IF; END $$do$$;'
+        exec /scripts/post_init.sh "$$@"
+        WRAP
+        chmod +x /scripts/post_init_wrapper.sh
         exec /bin/sh /launch.sh init
     environment:
       SCOPE: pgha-test
@@ -208,6 +240,9 @@ services:
       CLONE_HOST: postgresql
       CLONE_PORT: "5432"
       CLONE_USER: PGadmin
+      SPILO_CONFIGURATION: |
+        bootstrap:
+          post_init: /scripts/post_init_wrapper.sh "zalandos"
     secrets:
       - postgresql_password
       - postgresql_replication_password