Enable sanitizers for postgres v17 (#10401)

Add a build with sanitizers (asan, ubsan) to the CI pipeline and run
tests on it.

See https://github.com/neondatabase/neon/issues/6053

---------

Co-authored-by: Alexander Bayandin <alexander@neon.tech>

test_runner/fixtures/parametrize.py

@@ -124,5 +124,8 @@ def pytest_runtest_makereport(*args, **kwargs):
     allure.dynamic.parameter(
         "__lfc", "with-lfc" if os.getenv("USE_LFC") != "false" else "without-lfc"
     )
+    allure.dynamic.parameter(
+        "__sanitizers", "enabled" if os.getenv("SANITIZERS") == "enabled" else "disabled"
+    )
 
     yield

test_runner/regress/test_compatibility.py

@@ -314,7 +314,10 @@ def test_forward_compatibility(
 
 
 def check_neon_works(env: NeonEnv, test_output_dir: Path, sql_dump_path: Path, repo_dir: Path):
-    ep = env.endpoints.create_start("main")
+    ep = env.endpoints.create("main")
+    ep_env = {"LD_LIBRARY_PATH": str(env.pg_distrib_dir / f"v{env.pg_version}/lib")}
+    ep.start(env=ep_env)
+
     connstr = ep.connstr()
 
     pg_bin = PgBin(test_output_dir, env.pg_distrib_dir, env.pg_version)
@@ -363,7 +366,7 @@ def check_neon_works(env: NeonEnv, test_output_dir: Path, sql_dump_path: Path, r
     )
 
     # Timeline exists again: restart the endpoint
-    ep.start()
+    ep.start(env=ep_env)
 
     pg_bin.run_capture(
         ["pg_dumpall", f"--dbname={connstr}", f"--file={test_output_dir / 'dump-from-wal.sql'}"]

test_runner/regress/test_pg_regress.py

@@ -120,7 +120,7 @@ def post_checks(env: NeonEnv, test_output_dir: Path, db_name: str, endpoint: End
 
 # Run the main PostgreSQL regression tests, in src/test/regress.
 #
-@pytest.mark.timeout(900)  # Contains many sub-tests, is slow in debug builds
+@pytest.mark.timeout(3000)  # Contains many sub-tests, is slow in debug builds
 @pytest.mark.parametrize("shard_count", [None, 4])
 def test_pg_regress(
     neon_env_builder: NeonEnvBuilder,
@@ -194,7 +194,7 @@ def test_pg_regress(
 
 # Run the PostgreSQL "isolation" tests, in src/test/isolation.
 #
-@pytest.mark.timeout(600)  # Contains many sub-tests, is slow in debug builds
+@pytest.mark.timeout(1500)  # Contains many sub-tests, is slow in debug builds
 @pytest.mark.parametrize("shard_count", [None, 4])
 def test_isolation(
     neon_env_builder: NeonEnvBuilder,
@@ -222,6 +222,8 @@ def test_isolation(
             "max_prepared_transactions=100",
             # Enable the test mode, so that we don't need to patch the test cases.
             "neon.regress_test_mode = true",
+            # Stack size should be increased for tests to pass with asan.
+            "max_stack_depth = 4MB",
         ],
     )
     endpoint.safe_psql(f"CREATE DATABASE {DBNAME}")
@@ -417,7 +419,7 @@ def test_tx_abort_with_many_relations(
         try:
             # Rollback phase should be fast: this is one WAL record that we should process efficiently
             fut = exec.submit(rollback_and_wait)
-            fut.result(timeout=5)
+            fut.result(timeout=15)
         except:
             exec.shutdown(wait=False, cancel_futures=True)
             raise

test_runner/regress/test_subscriber_restart.py

@@ -3,12 +3,14 @@ from __future__ import annotations
 import threading
 import time
 
+import pytest
 from fixtures.neon_fixtures import NeonEnv
 from fixtures.utils import wait_until
 
 
 # This test checks of logical replication subscriber is able to correctly restart replication without receiving duplicates.
 # It requires tracking information about replication origins at page server side
+@pytest.mark.timeout(900)  # This test is slow with sanitizers enabled, especially on ARM
 def test_subscriber_restart(neon_simple_env: NeonEnv):
     env = neon_simple_env
     env.create_branch("publisher")