SELECT 💣(); (#8270)

## Problem
We want to be able to test how our infrastructure reacts on segfaults in
Postgres (for example, we collect cores, and get some required
logs/metrics, etc)

## Summary of changes
- Add `trigger_segfauls` function to `neon_test_utils` to trigger a
segfault in Postgres
- Add `trigger_panic` function to `neon_test_utils` to trigger SIGABRT
(by using `elog(PANIC, ...))
- Fix cleanup logic in regression tests in endpoint crashed

pgxn/neon_test_utils/Makefile

@@ -7,7 +7,7 @@ OBJS = \
 	neontest.o
 
 EXTENSION = neon_test_utils
-DATA = neon_test_utils--1.2.sql
+DATA = neon_test_utils--1.3.sql
 PGFILEDESC = "neon_test_utils - helpers for neon testing and debugging"
 
 PG_CONFIG = pg_config

pgxn/neon_test_utils/neon_test_utils--1.3.sql

@@ -45,3 +45,21 @@ CREATE FUNCTION neon_xlogflush(lsn pg_lsn DEFAULT NULL)
 RETURNS VOID
 AS 'MODULE_PATHNAME', 'neon_xlogflush'
 LANGUAGE C PARALLEL UNSAFE;
+
+CREATE FUNCTION trigger_panic()
+RETURNS VOID
+AS 'MODULE_PATHNAME', 'trigger_panic'
+LANGUAGE C PARALLEL UNSAFE;
+
+CREATE FUNCTION trigger_segfault()
+RETURNS VOID
+AS 'MODULE_PATHNAME', 'trigger_segfault'
+LANGUAGE C PARALLEL UNSAFE;
+
+-- Alias for `trigger_segfault`, just because `SELECT 💣()` looks fun
+CREATE OR REPLACE FUNCTION 💣() RETURNS void
+LANGUAGE plpgsql AS $$
+BEGIN
+    PERFORM trigger_segfault();
+END;
+$$;

pgxn/neon_test_utils/neon_test_utils.control

@@ -1,6 +1,6 @@
 # neon_test_utils extension
 comment = 'helpers for neon testing and debugging'
-default_version = '1.2'
+default_version = '1.3'
 module_pathname = '$libdir/neon_test_utils'
 relocatable = true
 trusted = true

pgxn/neon_test_utils/neontest.c

@@ -42,6 +42,8 @@ PG_FUNCTION_INFO_V1(clear_buffer_cache);
 PG_FUNCTION_INFO_V1(get_raw_page_at_lsn);
 PG_FUNCTION_INFO_V1(get_raw_page_at_lsn_ex);
 PG_FUNCTION_INFO_V1(neon_xlogflush);
+PG_FUNCTION_INFO_V1(trigger_panic);
+PG_FUNCTION_INFO_V1(trigger_segfault);
 
 /*
  * Linkage to functions in neon module.
@@ -489,3 +491,24 @@ neon_xlogflush(PG_FUNCTION_ARGS)
 	XLogFlush(lsn);
 	PG_RETURN_VOID();
 }
+
+/*
+ * Function to trigger panic.
+ */
+Datum
+trigger_panic(PG_FUNCTION_ARGS)
+{
+    elog(PANIC, "neon_test_utils: panic");
+    PG_RETURN_VOID();
+}
+
+/*
+ * Function to trigger a segfault.
+ */
+Datum
+trigger_segfault(PG_FUNCTION_ARGS)
+{
+    int *ptr = NULL;
+    *ptr = 42;
+    PG_RETURN_VOID();
+}

test_runner/fixtures/neon_fixtures.py

@@ -943,6 +943,8 @@ class NeonEnvBuilder:
                 # if the test threw an exception, don't check for errors
                 # as a failing assertion would cause the cleanup below to fail
                 ps_assert_metric_no_errors=(exc_type is None),
+                # do not fail on endpoint errors to allow the rest of cleanup to proceed
+                fail_on_endpoint_errors=False,
             )
             cleanup_error = None
 
@@ -1214,11 +1216,11 @@ class NeonEnv:
         for f in futs:
             f.result()
 
-    def stop(self, immediate=False, ps_assert_metric_no_errors=False):
+    def stop(self, immediate=False, ps_assert_metric_no_errors=False, fail_on_endpoint_errors=True):
         """
         After this method returns, there should be no child processes running.
         """
-        self.endpoints.stop_all()
+        self.endpoints.stop_all(fail_on_endpoint_errors)
 
         # Stop storage controller before pageservers: we don't want it to spuriously
         # detect a pageserver "failure" during test teardown
@@ -3900,9 +3902,17 @@ class EndpointFactory:
             pageserver_id=pageserver_id,
         )
 
-    def stop_all(self) -> "EndpointFactory":
+    def stop_all(self, fail_on_error=True) -> "EndpointFactory":
+        exception = None
         for ep in self.endpoints:
-            ep.stop()
+            try:
+                ep.stop()
+            except Exception as e:
+                log.error(f"Failed to stop endpoint {ep.endpoint_id}: {e}")
+                exception = e
+
+        if fail_on_error and exception is not None:
+            raise exception
 
         return self
 

test_runner/regress/test_endpoint_crash.py

@@ -0,0 +1,23 @@
+import pytest
+from fixtures.neon_fixtures import NeonEnvBuilder
+
+
+@pytest.mark.parametrize(
+    "sql_func",
+    [
+        "trigger_panic",
+        "trigger_segfault",
+        "💣",  # calls `trigger_segfault` internally
+    ],
+)
+def test_endpoint_crash(neon_env_builder: NeonEnvBuilder, sql_func: str):
+    """
+    Test that triggering crash from neon_test_utils crashes the endpoint
+    """
+    env = neon_env_builder.init_start()
+    env.neon_cli.create_branch("test_endpoint_crash")
+    endpoint = env.endpoints.create_start("test_endpoint_crash")
+
+    endpoint.safe_psql("CREATE EXTENSION neon_test_utils;")
+    with pytest.raises(Exception, match="This probably means the server terminated abnormally"):
+        endpoint.safe_psql(f"SELECT {sql_func}();")