Issue 5954 - Disable Transparent Huge Pages

dirsrvtests/tests/suites/healthcheck/health_tunables_test.py

@@ -0,0 +1,150 @@
+# --- BEGIN COPYRIGHT BLOCK ---
+# Copyright (C) 2023 Red Hat, Inc.
+# All rights reserved.
+#
+# License: GPL (version 3 or any later version).
+# See LICENSE for details.
+# --- END COPYRIGHT BLOCK ---
+#
+
+import subprocess
+import pytest
+import re
+import os
+from lib389.utils import *
+from lib389.cli_base import FakeArgs
+from lib389.topologies import topology_st
+from lib389.cli_ctl.health import health_check_run
+from lib389.paths import Paths
+
+CMD_OUTPUT = 'No issues found.'
+JSON_OUTPUT = '[]'
+RET_CODE = 'DSTHPLE0001'
+
+log = logging.getLogger(__name__)
+p = Paths()
+
+
+def run_healthcheck_and_flush_log(topology, instance, searched_code=None, json=False, searched_code2=None,
+                                  list_checks=False, list_errors=False, check=None, searched_list=None):
+    args = FakeArgs()
+    args.instance = instance.serverid
+    args.verbose = instance.verbose
+    args.list_errors = list_errors
+    args.list_checks = list_checks
+    args.check = check
+    args.dry_run = False
+    args.json = json
+
+    log.info('Use healthcheck with --json == {} option'.format(json))
+    health_check_run(instance, topology.logcap.log, args)
+
+    if searched_list is not None:
+        for item in searched_list:
+            assert topology.logcap.contains(item)
+            log.info('Healthcheck returned searched item: %s' % item)
+    else:
+        assert topology.logcap.contains(searched_code)
+        log.info('Healthcheck returned searched code: %s' % searched_code)
+
+    if searched_code2 is not None:
+        assert topology.logcap.contains(searched_code2)
+        log.info('Healthcheck returned searched code: %s' % searched_code2)
+
+    log.info('Clear the log')
+    topology.logcap.flush()
+
+
+def _set_thp_system_mode(mode):
+    thp_path = '/sys/kernel/mm/transparent_hugepage/enabled'
+    with open(thp_path, 'w') as f:
+        log.info(f"Setting THP mode to {mode}")
+        f.write(mode)
+
+
+def _set_thp_instance_mode(inst, disable: bool):
+    systemctl_env = {'SYSTEMD_EDITOR': 'tee'}
+    service_config = f"[Service]\nEnvironment=THP_DISABLE={int(disable)}"
+    subprocess.run(['systemctl', 'edit', '--drop-in=thp.conf', f'dirsrv@{inst.serverid}'],
+                    input=service_config.encode(), check=True, env=systemctl_env)
+    # `systemctl` tries to be helpful and passes +4 as a parameter for the EDITOR.
+    # But when `tee` is used as an EDITOR, it creates another file with our overrides contents.
+    # So let's remove it
+    if os.path.exists("+4"):
+        os.remove("+4")
+
+    inst.restart()
+
+
+def _get_thp_system_mode():
+    thp_path = '/sys/kernel/mm/transparent_hugepage/enabled'
+    enabled_value_pattern = r'\[([^\]]+)\]'
+    with open(thp_path, 'r') as f:
+        text = f.read().strip()
+        mode = re.search(enabled_value_pattern, text)[1]
+        log.info(f"Current THP mode is {mode}")
+        return mode
+
+
+@pytest.fixture(scope="function")
+def thp_reset(request):
+    mode = _get_thp_system_mode()
+
+    def fin():
+        _set_thp_system_mode(mode)
+
+    request.addfinalizer(fin)
+
+
+@pytest.mark.skipif(get_user_is_root() is False,
+                    reason="This test requires root permissions to change kernel tunables.")
+@pytest.mark.skipif(p.with_systemd is False, reason='Needs systemd to run')
+@pytest.mark.skipif(ds_is_older("2.5.0"), reason="Not implemented")
+@pytest.mark.parametrize("system_thp_mode,instance_thp_mode,expected_output",
+                       [("always", False, (RET_CODE, RET_CODE)),
+                        ("always", True, (CMD_OUTPUT, JSON_OUTPUT)),
+                        ("never", False, (CMD_OUTPUT, JSON_OUTPUT)),
+                        ("never", True, (CMD_OUTPUT, JSON_OUTPUT))],
+                        ids=["System and Instance THP ON",
+                             "System THP ON, Instance THP OFF",
+                             "System THP OFF, Instance THP ON",
+                             "System THP OFF, Instance THP OFF"])
+def test_healthcheck_transparent_huge_pages(topology_st, system_thp_mode, instance_thp_mode, expected_output, thp_reset):
+    """Check if HealthCheck returns DSTHPLE0001 code
+    :id: 1f195e10-6403-4c92-8ac9-724b669e8cf2
+    :setup: Standalone instance
+    :parametrized: yes
+    :steps:
+        1. Enable THP system wide and for the instance
+        2. Use HealthCheck without --json option
+        3. Use HealthCheck with --json option
+        4. Enable THP system wide, disable THP for the instance
+        5. Use HealthCheck without --json option
+        6. Use HealthCheck with --json option
+        7. Disable THP system wide, enable THP for the instance
+        8. Use HealthCheck without --json option
+        9. Use HealthCheck with --json option
+        10. Disable THP system wide, disable THP for the instance
+        11. Use HealthCheck without --json option
+        12. Use HealthCheck with --json option
+    :expectedresults:
+        1. Success
+        2. HealthCheck should return code DSHTPLE0001
+        3. HealthCheck should return code DSTHPLE0001
+        4. Success
+        5. HealthCheck reports no issue found
+        6. HealthCheck reports no issue found
+        7. Success
+        8. HealthCheck reports no issue found
+        9. HealthCheck reports no issue found
+        10. Success
+        11. HealthCheck reports no issue found
+        12. HealthCheck reports no issue found
+    """
+    standalone = topology_st.standalone
+    standalone.config.set("nsslapd-accesslog-logbuffering", "on")
+
+    _set_thp_system_mode(system_thp_mode)
+    _set_thp_instance_mode(standalone, instance_thp_mode)
+    run_healthcheck_and_flush_log(topology_st, standalone, expected_output[0], json=False)
+    run_healthcheck_and_flush_log(topology_st, standalone, expected_output[1], json=True)

dirsrvtests/tests/suites/healthcheck/healthcheck_test.py

@@ -153,7 +153,9 @@ def test_healthcheck_list_checks(topology_st):
                    'replication:conflicts',
                    'dseldif:nsstate',
                    'tls:certificate_expiration',
-                   'logs:notes']
+                   'logs:notes',
+                   'tunables:thp',
+                   ]
 
     standalone = topology_st.standalone
 
@@ -205,6 +207,7 @@ def test_healthcheck_list_errors(topology_st):
                    'DSSKEWLE0001 :: Medium time skew',
                    'DSSKEWLE0002 :: Major time skew',
                    'DSSKEWLE0003 :: Extensive time skew',
+                   'DSTHPLE0001 :: Transparent Huge Pages',
                    'DSVIRTLE0001 :: Virtual attribute indexed']
 
     standalone = topology_st.standalone

ldap/servers/slapd/main.c

@@ -67,6 +67,10 @@ union semun
 #include <malloc.h>
 #endif
 
+#ifdef LINUX
+#include <sys/prctl.h>
+#endif
+
 /* Forward Declarations */
 
 struct main_config
@@ -523,6 +527,14 @@ main(int argc, char **argv)
 {
     int return_value = 0;
     struct main_config mcfg = {0};
+#ifdef LINUX
+#if defined(PR_SET_THP_DISABLE)
+    char *thp_disable =  getenv("THP_DISABLE");
+    if (thp_disable != NULL && strcmp(thp_disable, "1") == 0) {
+        prctl(PR_SET_THP_DISABLE, 1, 0, 0, 0);
+    }
+#endif
+#endif
 
     /* Set a number of defaults */
     mcfg.slapd_exemode = SLAPD_EXEMODE_UNKNOWN;

src/lib389/lib389/cli_ctl/health.py

@@ -19,6 +19,7 @@
 from lib389.nss_ssl import NssSsl
 from lib389.dseldif import FSChecks, DSEldif
 from lib389.dirsrv_log import DirsrvAccessLog
+from lib389.tunables import Tunables
 from lib389 import lint
 from lib389 import plugins
 from lib389._constants import DSRC_HOME
@@ -39,6 +40,7 @@
     DSEldif,
     NssSsl,
     DirsrvAccessLog,
+    Tunables,
 ]
 
 

src/lib389/lib389/lint.py

@@ -511,3 +511,29 @@
     'fix': """Stop using this these unknown attributes in the filter, or add the schema
 to the server and make sure it's properly indexed."""
 }
+
+# Transparent Huge Pages
+DSTHPLE0001 = {
+    'dsle': 'DSTHPLE0001',
+    'severity': 'Medium',
+    'description': 'Transparent Huge Pages',
+    'items': ['Possible Performance Impact'],
+    'detail': """Transparent Huge Pages are enabled. This can lead to an unexpected memory
+consumption, especially when using large caches.\n""",
+    'fix': """Disable Transparent Huge Pages.
+System-wide at boot:
+Add "transparent_hugepage=never" to the list of kernel boot parameters.
+
+System-wide at runtime:
+# echo "never" > /sys/kernel/mm/transparent_hugepage/enabled
+# echo "never" > /sys/kernel/mm/transparent_hugepage/defrag
+
+Per instance (for the versions of 389 Directory Server that support it):
+Edit dirsrv unit file:
+# systemctl edit dirsrv@instance_name
+
+And uncomment the following lines:
+[Service]
+Environment=THP_DISABLE=1
+"""
+}

src/lib389/lib389/tunables.py

@@ -0,0 +1,65 @@
+# --- BEGIN COPYRIGHT BLOCK ---
+# Copyright (C) 2023 Red Hat, Inc.
+# All rights reserved.
+#
+# License: GPL (version 3 or any later version).
+# See LICENSE for details.
+# --- END COPYRIGHT BLOCK ---
+#
+
+import os
+import re
+import copy
+from lib389._mapped_object_lint import DSLint
+from lib389 import pid_from_file
+from lib389.lint import DSTHPLE0001
+
+class Tunables(DSLint):
+    """A class for working with system tunables
+    :param instance: An instance
+    :type instance: lib389.DirSrv
+    """
+
+    def __init__(self, instance):
+        self._instance = instance
+        self.pid = str(pid_from_file(instance.ds_paths.pid_file))
+
+
+    @classmethod
+    def lint_uid(cls):
+        return 'tunables'
+
+
+    def _lint_thp(self):
+        """Check if THP is enabled"""
+        def systemwide_thp_enabled() -> bool:
+            thp_path = '/sys/kernel/mm/transparent_hugepage'
+            thp_enabled_path = os.path.join(thp_path, "enabled")
+            thp_status_pattern = r"(.*\[always\].*)|(.*\[madvise\].*)"
+            if os.path.exists(thp_enabled_path):
+                with open(thp_enabled_path, 'r') as f:
+                    thp_status = f.read().strip()
+                    match = re.match(thp_status_pattern, thp_status)
+                    return match is not None
+
+
+        def instance_thp_enabled() -> bool:
+            pid_status_path = f"/proc/{self.pid}/status"
+
+            with open(pid_status_path, 'r') as pid_status:
+                pid_status_content = pid_status.read()
+                thp_line = None
+                for line in pid_status_content.split('\n'):
+                    if 'THP_enabled' in line:
+                        thp_line = line
+                        break
+                if thp_line is not None:
+                    thp_value = int(thp_line.split()[1])
+                    return bool(thp_value)
+
+
+        if instance_thp_enabled() and systemwide_thp_enabled():
+            report = copy.deepcopy(DSTHPLE0001)
+            report['check'] = 'tunables:transparent_huge_pages'
+            yield report
+

wrappers/systemd.template.service.custom.conf.in

@@ -46,6 +46,9 @@ TimeoutStopSec=600
 # Preload jemalloc
 Environment=LD_PRELOAD=@libdir@/@package_name@/lib/libjemalloc.so.2
 
+# Disable Transparent Huge Pages
+Environment=THP_DISABLE=1
+
 ##################################################
 #         Heap profiling with jemalloc           #
 ##################################################