Add guc: gp_resource_group_cgroup_parent (#16738)

Add guc: gp_resource_group_cgroup_parent (only for cgroup v2).

Current gpdb doesn't support change root cgroup path of resource group.
For some situations, it's better if gpdb can change the root cgroup path
of resource group.

For example, on the OS with systemd, user maybe want to create a
delegated cgroup to gpdb via systemd, but the delegated cgroup must end
with .service which typically is /sys/fs/cgroup/gpdb.service. And in
other OS without systemd, user maybe want to use /sys/fs/cgroup/gpdb or
other locations directly. So add the gp_resource_group_cgroup_parent can
make the resource group more flexible.

Author: RMTT

gpMgmt/bin/gpconfig

@@ -15,6 +15,7 @@
 import os
 import sys
 import re
+import psutil
 
 try:
     from gppylib.gpparseopts import OptParser, OptChecker
@@ -170,6 +171,20 @@ class Guc:
                     return msg
             elif newval != "'queue'":
                 return "the value of gp_resource_manager must be 'group' or 'group-v2' or 'queue'"
+        elif self.name == "gp_resource_group_cgroup_parent":
+            base = newval.strip("'")
+            if not re.match("^[0-9a-zA-Z][-._0-9a-zA-Z]*$", base):
+                return "resource group cgroup parent can only contains alphabet, number, and non-leading . _ -"
+
+            path = None
+            for partition in psutil.disk_partitions(all=True):
+                if partition.fstype == "cgroup2":
+                    path = partition.mountpoint + "/" + base
+                    break
+            if path is None:
+                return "cannot find cgroup v2 mountpoint"
+            if not os.path.isdir(path):
+                return "'%s' doesn't exists or is not a directory" % path
 
         elif self.name == 'unix_socket_permissions':
             if newval[0] != '0':

src/backend/cdb/cdbvars.c

@@ -17,9 +17,11 @@
  *
  *-------------------------------------------------------------------------
  */
+#include "catalog/pg_collation_d.h"
 #include "postgres.h"
 
 #include "miscadmin.h"
+#include "regex/regex.h"
 #include "utils/guc.h"
 #include "cdb/cdbvars.h"
 #include "libpq-fe.h"
@@ -567,6 +569,44 @@ gpvars_show_gp_resource_manager_policy(void)
 	}
 }
 
+bool gpvars_check_gp_resource_group_cgroup_parent(char **newval, void **extra, GucSource source)
+{
+	int		  regres;
+	char	  err[128];
+	regex_t  re;
+	char	 *pattern = "^[0-9a-zA-Z][-._0-9a-zA-Z]*$";
+	pg_wchar *wpattern = palloc((strlen(pattern) + 1) * sizeof(pg_wchar));
+	int		  wlen = pg_mb2wchar_with_len(pattern, wpattern, strlen(pattern));
+	pg_wchar *data = palloc((strlen(*newval) + 1) * sizeof(pg_wchar));
+	int		  data_len = pg_mb2wchar_with_len(*newval, data, sizeof(*newval));
+	bool	  match = true;
+
+	regres = pg_regcomp(&re, wpattern, wlen, REG_ADVANCED, DEFAULT_COLLATION_OID);
+	if (regres != REG_OKAY)
+	{
+		pg_regerror(regres, &re, err, sizeof(err));
+		GUC_check_errmsg("compile regex failed: %s", err);
+
+		pfree(wpattern);
+		pfree(data);
+
+		return false;
+	}
+
+	regres = pg_regexec(&re, data, data_len, 0, NULL, 0, NULL, 0);
+	if (regres != REG_OKAY)
+	{
+		match = false;
+		GUC_check_errmsg("gp_resource_group_cgroup_parent can only contains alphabet, number and non-leading . _ -");
+	}
+
+	pg_regfree(&re);
+	pfree(wpattern);
+	pfree(data);
+
+	return match;
+}
+
 /*
  * gpvars_assign_statement_mem
  */

src/backend/utils/misc/guc_gp.c

@@ -239,6 +239,7 @@ double		gp_resource_group_cpu_limit;
 bool		gp_resource_group_bypass;
 bool		gp_resource_group_bypass_catalog_query;
 bool		gp_resource_group_bypass_direct_dispatch;
+char	   *gp_resource_group_cgroup_parent;
 
 /* Metrics collector debug GUC */
 bool		vmem_process_interrupt = false;
@@ -4900,6 +4901,17 @@ struct config_string ConfigureNamesString_gp[] =
 		gpvars_show_gp_resource_manager_policy,
 	},
 
+	{
+		{"gp_resource_group_cgroup_parent", PGC_POSTMASTER, RESOURCES,
+			gettext_noop("The root of gpdb cgroup hierarchy."),
+			NULL,
+			GUC_SUPERUSER_ONLY
+		},
+		&gp_resource_group_cgroup_parent,
+		"gpdb.service",
+		gpvars_check_gp_resource_group_cgroup_parent, NULL, NULL
+	},
+
 	/* for pljava */
 	{
 		{"pljava_vmoptions", PGC_SUSET, CUSTOM_OPTIONS,

src/backend/utils/resgroup/cgroup.c

@@ -174,7 +174,7 @@ buildPathSafe(Oid group,
 		 * for cgroup v2, we just have the top level and child level,
 		 * don't need to care about the component.
 		 */
-		base_dir = base == BASEDIR_GPDB ? "gpdb" : "";
+		base_dir = base == BASEDIR_GPDB ? gp_resource_group_cgroup_parent : "";
 		len = snprintf(path, path_size, "%s/%s%s/%s",
 					   cgroupSystemInfo->cgroup_dir, base_dir, group_dir, filename);
 	}

src/include/utils/guc.h

@@ -860,6 +860,7 @@ extern void gpvars_assign_gp_resource_manager_policy(const char *newval, void *e
 extern const char *gpvars_show_gp_resource_manager_policy(void);
 extern const char *gpvars_assign_gp_resqueue_memory_policy(const char *newval, bool doit, GucSource source);
 extern const char *gpvars_show_gp_resqueue_memory_policy(void);
+extern bool gpvars_check_gp_resource_group_cgroup_parent(char **newval, void **extra, GucSource source);
 extern bool gpvars_check_statement_mem(int *newval, void **extra, GucSource source);
 extern bool gpvars_check_rg_query_fixed_mem(int *newval, void **extra, GucSource source);
 extern int guc_name_compare(const char *namea, const char *nameb);

src/include/utils/resgroup.h

@@ -109,6 +109,7 @@ extern int gp_resource_group_queuing_timeout;
 extern bool gp_resource_group_bypass_catalog_query;
 extern int gp_resource_group_move_timeout;
 extern bool gp_resource_group_bypass_direct_dispatch;
+extern char *gp_resource_group_cgroup_parent;
 
 /*
  * Non-GUC global variables.

src/include/utils/unsync_guc_name.h

@@ -260,6 +260,7 @@
 		"gp_resource_group_bypass_direct_dispatch",
 		"gp_resource_group_queuing_timeout",
 		"gp_resource_group_move_timeout",
+		"gp_resource_group_cgroup_parent",
 		"gp_resource_manager",
 		"gp_resqueue_memory_policy",
 		"gp_resqueue_priority",

src/test/isolation2/expected/resgroup/resgroup_auxiliary_tools_v2.out

@@ -23,6 +23,8 @@ ERROR:  language "plpython3u" already exists
 -- start_ignore
 ! gpconfig -c gp_resource_manager -v group-v2;
 
+! gpconfig -c gp_resource_group_cgroup_parent -v "gpdb"
+
 ! gpconfig -c max_connections -v 250 -m 25;
 
 ! gpconfig -c runaway_detector_activation_percent -v 100;

src/test/isolation2/sql/resgroup/resgroup_auxiliary_tools_v2.sql

@@ -21,6 +21,7 @@ CREATE LANGUAGE plpython3u;
 -- enable resource group and restart cluster.
 -- start_ignore
 ! gpconfig -c gp_resource_manager -v group-v2;
+! gpconfig -c gp_resource_group_cgroup_parent -v "gpdb"
 ! gpconfig -c max_connections -v 250 -m 25;
 ! gpconfig -c runaway_detector_activation_percent -v 100;
 ! gpstop -rai;