Repair monitor cluster upgrade without blocking K1 package configuration

tests/test_node_k1_package_lifecycle.py

@@ -9,9 +9,10 @@ import pytest
 PACKAGING = Path(__file__).resolve().parents[1] / "apps/node-agent/packaging"
 
 
+@pytest.mark.parametrize("monitor_fails", [False, True])
 @pytest.mark.parametrize("active,configured", [(True, True), (False, True), (True, False)])
 def test_node_update_restores_only_previously_running_configured_plugin(
-    tmp_path, active, configured,
+    tmp_path, active, configured, monitor_fails,
 ):
     binary = tmp_path / "bin"
     binary.mkdir()
@@ -34,7 +35,7 @@ case "$1" in
 esac
 ''')
     (binary / "getent").write_text("#!/bin/sh\nexit 0\n")
-    (binary / "setup-monitor").write_text("#!/bin/sh\nprintf '%s\\n' monitor-setup >> \"$TEST_EVENTS\"\n")
+    (binary / "setup-monitor").write_text("#!/bin/sh\nprintf '%s\\n' monitor-setup >> \"$TEST_EVENTS\"\nexit " + ("1" if monitor_fails else "0") + "\n")
     (binary / "dpkg-query").write_text(
         "#!/bin/sh\necho 'install ok " + ("installed" if configured else "unpacked") + "'\n"
     )
@@ -56,3 +57,5 @@ esac
     assert ("start mission-core-k1.service" in calls) is (active and configured)
     assert state.exists() is (active and configured)
     assert not (run / "mission-core-node-k1-upgrade-active").exists()
+
+    assert (run / "mission-core-monitor-setup-failed").exists() is monitor_fails

tests/test_node_monitor_package.py

@@ -0,0 +1,74 @@
+"""Exercise the monitor bootstrap against PostgreSQL's systemd path contract."""
+
+import os
+import subprocess
+from pathlib import Path
+
+import pytest
+
+SETUP = Path(__file__).resolve().parents[1] / "apps/node-agent/packaging/setup-monitor"
+
+
+@pytest.mark.parametrize("legacy,foreign", [(False, False), (True, False), (True, True)])
+def test_cluster_path_and_r18_migration_are_repeatable(tmp_path, legacy, foreign):
+    binary = tmp_path / "bin"
+    binary.mkdir()
+    conf = tmp_path / "etc/postgresql/16"
+    conf.mkdir(parents=True)
+    if legacy:
+        (conf / "ndc-monitor").mkdir()
+        (conf / "ndc-monitor/postgresql.conf").write_text("preserved R18 configuration")
+    driver = binary / "driver"
+    driver.write_text("""#!/usr/bin/env python3
+import os,sys,pathlib
+root=pathlib.Path(os.environ['TEST_ROOT']);name=pathlib.Path(sys.argv[0]).name;args=sys.argv[1:]
+with (root/'events').open('a') as log:log.write(name+' '+ ' '.join(args)+'\\n')
+conf=root/'etc/postgresql/16'
+if name=='id':print(0)
+elif name=='install':pathlib.Path(args[-1]).mkdir(parents=True,exist_ok=True)
+elif name=='pg_conftool':
+ print(os.environ.get('TEST_DATA_ROOT',str(root/'var/lib/mission-core-monitor-db')) if args[-1]=='data_directory' else '5433')
+elif name=='pg_createcluster':
+ (conf/args[1]).mkdir();(conf/args[1]/'postgresql.conf').write_text('new database')
+elif name=='pg_renamecluster':
+ (conf/args[1]).rename(conf/args[2])
+elif name=='systemctl' and args[0]=='restart' and args[1].startswith('postgresql@'):
+ # Exact installed Debian template: AssertPathExists=/etc/postgresql/%I/postgresql.conf.
+ instance=args[1].split('@')[1].removesuffix('.service')
+ assert (root/'etc/postgresql'/instance.replace('-','/')/'postgresql.conf').is_file()
+""")
+    driver.chmod(0o700)
+    for name in [
+        "id",
+        "getent",
+        "install",
+        "pg_conftool",
+        "pg_createcluster",
+        "pg_renamecluster",
+        "systemctl",
+        "runuser",
+    ]:
+        (binary / name).symlink_to(driver)
+    script = tmp_path / "setup"
+    text = SETUP.read_text()
+    for path in ["/etc/", "/var/lib/", "/run/"]:
+        text = text.replace(path, str(tmp_path / path.lstrip("/")) + "/")
+    script.write_text(text)
+    env = dict(
+        os.environ, PATH=str(binary) + os.pathsep + os.environ["PATH"], TEST_ROOT=str(tmp_path)
+    )
+    if foreign:
+        env["TEST_DATA_ROOT"] = "/unrelated/database"
+        result = subprocess.run(["/bin/sh", str(script)], env=env, capture_output=True)
+        assert result.returncode != 0
+        assert (conf / "ndc-monitor/postgresql.conf").is_file()
+        assert "pg_renamecluster" not in (tmp_path / "events").read_text()
+        return
+    for _ in range(2):
+        subprocess.run(["/bin/sh", str(script)], env=env, check=True, capture_output=True)
+    calls = (tmp_path / "events").read_text()
+    assert calls.count("pg_renamecluster 16 ndc-monitor ndcmonitor") == int(legacy)
+    assert calls.count("pg_createcluster 16 ndcmonitor") == int(not legacy)
+    assert not (conf / "ndc-monitor").exists()
+    if legacy:
+        assert (conf / "ndcmonitor/postgresql.conf").read_text() == "preserved R18 configuration"