fix: prevent data loss on restart after unclean shutdown

sdk/python/tests/test_pg0.py

@@ -1,5 +1,9 @@
 """Tests for pg0 Python client."""
 
+import os
+import signal
+import time
+
 import pytest
 import pg0
 from pg0 import Pg0, InstanceInfo, Pg0AlreadyRunningError, Pg0Error
@@ -147,6 +151,38 @@ def test_port_conflict_error(self, clean_instance):
             pg2.stop()
             pg0.drop(f"{TEST_NAME}-2")
 
+    def test_data_survives_crash(self, clean_instance):
+        """Test that data is preserved after an unclean shutdown (SIGKILL).
+
+        Regression test for https://github.com/vectorize-io/pg0/issues/6
+        Simulates a crash by sending SIGKILL to the PostgreSQL process,
+        which leaves a stale postmaster.pid. On restart, data must still exist.
+        """
+        pg = Pg0(name=TEST_NAME, port=TEST_PORT)
+        info = pg.start()
+
+        # Create a table and insert data
+        pg.execute("CREATE TABLE crash_test (id serial PRIMARY KEY, value text);")
+        pg.execute("INSERT INTO crash_test (value) VALUES ('survive_crash');")
+        result = pg.execute("SELECT value FROM crash_test;")
+        assert "survive_crash" in result
+
+        # Simulate a crash: SIGKILL the postgres process (leaves stale postmaster.pid)
+        pid = info.pid
+        assert pid is not None
+        os.kill(pid, signal.SIGKILL)
+        time.sleep(1)  # Wait for process to die
+
+        # Restart — this must NOT lose data
+        info = pg.start()
+        assert info.running is True
+
+        # Verify data survived the crash
+        result = pg.execute("SELECT value FROM crash_test;")
+        assert "survive_crash" in result
+
+        pg.stop()
+
 
 class TestConvenienceFunctions:
     """Tests for module-level convenience functions."""

src/main.rs

@@ -537,7 +537,13 @@ fn start(
         if is_process_running(info.pid) {
             return Err(CliError::AlreadyRunning(info.pid));
         }
-        // Stale instance, clean up
+        // Stale instance: clean up instance metadata but preserve data directory.
+        // Remove stale postmaster.pid so PostgreSQL can start with existing data.
+        let pid_file = info.data_dir.join("postmaster.pid");
+        if pid_file.exists() {
+            println!("Removing stale postmaster.pid (process {} no longer running)...", info.pid);
+            fs::remove_file(&pid_file)?;
+        }
         remove_instance(&name)?;
     }
 
@@ -611,6 +617,7 @@ fn start(
         installation_dir: version_install_dir,
         configuration,
         trust_installation_dir: true, // Use our extracted files
+        temporary: false, // Never delete data directory on drop - pg0 manages data lifecycle explicitly
         timeout: Some(std::time::Duration::from_secs(600)), // 10 minute timeout for slow systems (ARM64 emulation under QEMU)
         ..Default::default()
     };