Add path column to pg_backend_memory_contexts view

"path" provides a reliable method of determining the parent/child
relationships between memory contexts.  Previously this could be done in
a non-reliable way by writing a recursive query and joining the "parent"
and "name" columns.  This wasn't reliable as the names were not unique,
which could result in joining to the wrong parent.

To make this reliable, "path" stores an array of numerical identifiers
starting with the identifier for TopLevelMemoryContext.  It contains an
element for each intermediate parent between that and the current context.

Incompatibility: Here we also adjust the "level" column to make it
1-based rather than 0-based.  A 1-based level provides a convenient way
to access elements in the "path" array. e.g. path[level] gives the
identifier for the current context.

Identifiers are not stable across multiple evaluations of the view.  In
an attempt to make these more stable for ad-hoc queries, the identifiers
are assigned breadth-first.  Contexts closer to TopLevelMemoryContext
are less likely to change between queries and during queries.

Author: Melih Mutlu <m.melihmutlu@gmail.com>
Discussion: https://postgr.es/m/CAGPVpCThLyOsj3e_gYEvLoHkr5w=tadDiN_=z2OwsK3VJppeBA@mail.gmail.com
Reviewed-by: Andres Freund, Stephen Frost, Atsushi Torikoshi,
Reviewed-by: Michael Paquier, Robert Haas, David Rowley

Author: david-rowley

doc/src/sgml/system-views.sgml

@@ -504,7 +504,22 @@
        <structfield>level</structfield> <type>int4</type>
       </para>
       <para>
-       Distance from TopMemoryContext in context tree
+       The 1-based level of the context in the memory context hierarchy. The
+       level of a context also shows the position of that context in the
+       <structfield>path</structfield> column.
+      </para></entry>
+     </row>
+
+     <row>
+      <entry role="catalog_table_entry"><para role="column_definition">
+       <structfield>path</structfield> <type>int4[]</type>
+      </para>
+      <para>
+       Array of transient numerical identifiers to describe the memory
+       context hierarchy. The first element is for
+       <literal>TopMemoryContext</literal>, subsequent elements contain
+       intermediate parents and the final element contains the identifier for
+       the current context.
       </para></entry>
      </row>
 
@@ -561,6 +576,29 @@
    read only by superusers or roles with the privileges of the
    <literal>pg_read_all_stats</literal> role.
   </para>
+
+  <para>
+   Since memory contexts are created and destroyed during the running of a
+   query, the identifiers stored in the <structfield>path</structfield> column
+   can be unstable between multiple invocations of the view in the same query.
+   The example below demonstrates an effective usage of this column and
+   calculates the total number of bytes used by
+   <literal>CacheMemoryContext</literal> and all of its children:
+
+<programlisting>
+WITH memory_contexts AS (
+    SELECT * FROM pg_backend_memory_contexts
+)
+SELECT sum(c1.total_bytes)
+FROM memory_contexts c1, memory_contexts c2
+WHERE c2.name = 'CacheMemoryContext'
+AND c1.path[c2.level] = c2.path[c2.level];
+</programlisting>
+
+   The <link linkend="queries-with">Common Table Expression</link> is used
+   to ensure the context IDs in the <structfield>path</structfield> column
+   match between both evaluations of the view.
+  </para>
  </sect1>
 
  <sect1 id="view-pg-config">

src/backend/utils/adt/mcxtfuncs.c

@@ -19,55 +19,117 @@
 #include "mb/pg_wchar.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "utils/array.h"
 #include "utils/builtins.h"
+#include "utils/hsearch.h"
 
 /* ----------
  * The max bytes for showing identifiers of MemoryContext.
  * ----------
  */
 #define MEMORY_CONTEXT_IDENT_DISPLAY_SIZE	1024
 
+/*
+ * MemoryContextId
+ *		Used for storage of transient identifiers for
+ *		pg_get_backend_memory_contexts.
+ */
+typedef struct MemoryContextId
+{
+	MemoryContext context;
+	int			context_id;
+}			MemoryContextId;
+
+/*
+ * get_memory_context_name_and_ident
+ *		Populate *name and *ident from the name and ident from 'context'.
+ */
+static void
+get_memory_context_name_and_ident(MemoryContext context, const char **const name,
+								  const char **const ident)
+{
+	*name = context->name;
+	*ident = context->ident;
+
+	/*
+	 * To be consistent with logging output, we label dynahash contexts with
+	 * just the hash table name as with MemoryContextStatsPrint().
+	 */
+	if (ident && strcmp(*name, "dynahash") == 0)
+	{
+		*name = *ident;
+		*ident = NULL;
+	}
+}
+
+/*
+ * int_list_to_array
+ *		Convert an IntList to an array of INT4OIDs.
+ */
+static Datum
+int_list_to_array(const List *list)
+{
+	Datum	   *datum_array;
+	int			length;
+	ArrayType  *result_array;
+
+	length = list_length(list);
+	datum_array = (Datum *) palloc(length * sizeof(Datum));
+
+	foreach_int(i, list)
+		datum_array[foreach_current_index(i)] = Int32GetDatum(i);
+
+	result_array = construct_array_builtin(datum_array, length, INT4OID);
+
+	return PointerGetDatum(result_array);
+}
+
 /*
  * PutMemoryContextsStatsTupleStore
- *		One recursion level for pg_get_backend_memory_contexts.
+ *		Add details for the given MemoryContext to 'tupstore'.
  */
 static void
 PutMemoryContextsStatsTupleStore(Tuplestorestate *tupstore,
 								 TupleDesc tupdesc, MemoryContext context,
-								 const char *parent, int level)
+								 HTAB *context_id_lookup)
 {
-#define PG_GET_BACKEND_MEMORY_CONTEXTS_COLS	10
+#define PG_GET_BACKEND_MEMORY_CONTEXTS_COLS	11
 
 	Datum		values[PG_GET_BACKEND_MEMORY_CONTEXTS_COLS];
 	bool		nulls[PG_GET_BACKEND_MEMORY_CONTEXTS_COLS];
 	MemoryContextCounters stat;
-	MemoryContext child;
+	List	   *path = NIL;
 	const char *name;
 	const char *ident;
 	const char *type;
 
 	Assert(MemoryContextIsValid(context));
 
-	name = context->name;
-	ident = context->ident;
-
 	/*
-	 * To be consistent with logging output, we label dynahash contexts with
-	 * just the hash table name as with MemoryContextStatsPrint().
+	 * Figure out the transient context_id of this context and each of its
+	 * ancestors.
 	 */
-	if (ident && strcmp(name, "dynahash") == 0)
+	for (MemoryContext cur = context; cur != NULL; cur = cur->parent)
 	{
-		name = ident;
-		ident = NULL;
+		MemoryContextId *entry;
+		bool		found;
+
+		entry = hash_search(context_id_lookup, &cur, HASH_FIND, &found);
+
+		if (!found)
+			elog(ERROR, "hash table corrupted");
+		path = lcons_int(entry->context_id, path);
 	}
 
 	/* Examine the context itself */
 	memset(&stat, 0, sizeof(stat));
-	(*context->methods->stats) (context, NULL, (void *) &level, &stat, true);
+	(*context->methods->stats) (context, NULL, NULL, &stat, true);
 
 	memset(values, 0, sizeof(values));
 	memset(nulls, 0, sizeof(nulls));
 
+	get_memory_context_name_and_ident(context, &name, &ident);
+
 	if (name)
 		values[0] = CStringGetTextDatum(name);
 	else
@@ -92,8 +154,15 @@ PutMemoryContextsStatsTupleStore(Tuplestorestate *tupstore,
 	else
 		nulls[1] = true;
 
-	if (parent)
-		values[2] = CStringGetTextDatum(parent);
+	if (context->parent)
+	{
+		const char *parent_name,
+				   *parent_ident;
+
+		get_memory_context_name_and_ident(context->parent, &parent_name,
+										  &parent_ident);
+		values[2] = CStringGetTextDatum(parent_name);
+	}
 	else
 		nulls[2] = true;
 
@@ -117,19 +186,16 @@ PutMemoryContextsStatsTupleStore(Tuplestorestate *tupstore,
 	}
 
 	values[3] = CStringGetTextDatum(type);
-	values[4] = Int32GetDatum(level);
-	values[5] = Int64GetDatum(stat.totalspace);
-	values[6] = Int64GetDatum(stat.nblocks);
-	values[7] = Int64GetDatum(stat.freespace);
-	values[8] = Int64GetDatum(stat.freechunks);
-	values[9] = Int64GetDatum(stat.totalspace - stat.freespace);
-	tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	values[4] = Int32GetDatum(list_length(path));	/* level */
+	values[5] = int_list_to_array(path);
+	values[6] = Int64GetDatum(stat.totalspace);
+	values[7] = Int64GetDatum(stat.nblocks);
+	values[8] = Int64GetDatum(stat.freespace);
+	values[9] = Int64GetDatum(stat.freechunks);
+	values[10] = Int64GetDatum(stat.totalspace - stat.freespace);
 
-	for (child = context->firstchild; child != NULL; child = child->nextchild)
-	{
-		PutMemoryContextsStatsTupleStore(tupstore, tupdesc,
-										 child, name, level + 1);
-	}
+	tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	list_free(path);
 }
 
 /*
@@ -140,10 +206,66 @@ Datum
 pg_get_backend_memory_contexts(PG_FUNCTION_ARGS)
 {
 	ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+	int			context_id;
+	List	   *contexts;
+	HASHCTL		ctl;
+	HTAB	   *context_id_lookup;
+
+	ctl.keysize = sizeof(MemoryContext);
+	ctl.entrysize = sizeof(MemoryContextId);
+	ctl.hcxt = CurrentMemoryContext;
+
+	context_id_lookup = hash_create("pg_get_backend_memory_contexts",
+									256,
+									&ctl,
+									HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
 
 	InitMaterializedSRF(fcinfo, 0);
-	PutMemoryContextsStatsTupleStore(rsinfo->setResult, rsinfo->setDesc,
-									 TopMemoryContext, NULL, 0);
+
+	/*
+	 * Here we use a non-recursive algorithm to visit all MemoryContexts
+	 * starting with TopMemoryContext.  The reason we avoid using a recursive
+	 * algorithm is because we want to assign the context_id breadth-first.
+	 * I.e. all contexts at level 1 are assigned IDs before contexts at level
+	 * 2.  Because contexts closer to TopMemoryContext are less likely to
+	 * change, this makes the assigned context_id more stable.  Otherwise, if
+	 * the first child of TopMemoryContext obtained an additional grandchild,
+	 * the context_id for the second child of TopMemoryContext would change.
+	 */
+	contexts = list_make1(TopMemoryContext);
+
+	/* TopMemoryContext will always have a context_id of 1 */
+	context_id = 1;
+
+	foreach_ptr(MemoryContextData, cur, contexts)
+	{
+		MemoryContextId *entry;
+		bool		found;
+
+		/*
+		 * Record the context_id that we've assigned to each MemoryContext.
+		 * PutMemoryContextsStatsTupleStore needs this to populate the "path"
+		 * column with the parent context_ids.
+		 */
+		entry = (MemoryContextId *) hash_search(context_id_lookup, &cur,
+												HASH_ENTER, &found);
+		entry->context_id = context_id++;
+		Assert(!found);
+
+		PutMemoryContextsStatsTupleStore(rsinfo->setResult,
+										 rsinfo->setDesc,
+										 cur,
+										 context_id_lookup);
+
+		/*
+		 * Append all children onto the contexts list so they're processed by
+		 * subsequent iterations.
+		 */
+		for (MemoryContext c = cur->firstchild; c != NULL; c = c->nextchild)
+			contexts = lappend(contexts, c);
+	}
+
+	hash_destroy(context_id_lookup);
 
 	return (Datum) 0;
 }

src/include/catalog/catversion.h

@@ -57,6 +57,6 @@
  */
 
 /*							yyyymmddN */
-#define CATALOG_VERSION_NO	202407111
+#define CATALOG_VERSION_NO	202407251
 
 #endif

src/include/catalog/pg_proc.dat

@@ -8290,9 +8290,9 @@
   proname => 'pg_get_backend_memory_contexts', prorows => '100',
   proretset => 't', provolatile => 'v', proparallel => 'r',
   prorettype => 'record', proargtypes => '',
-  proallargtypes => '{text,text,text,text,int4,int8,int8,int8,int8,int8}',
-  proargmodes => '{o,o,o,o,o,o,o,o,o,o}',
-  proargnames => '{name, ident, parent, type, level, total_bytes, total_nblocks, free_bytes, free_chunks, used_bytes}',
+  proallargtypes => '{text,text,text,text,int4,_int4,int8,int8,int8,int8,int8}',
+  proargmodes => '{o,o,o,o,o,o,o,o,o,o,o}',
+  proargnames => '{name, ident, parent, type, level, path, total_bytes, total_nblocks, free_bytes, free_chunks, used_bytes}',
   prosrc => 'pg_get_backend_memory_contexts' },
 
 # logging memory contexts of the specified backend

src/include/nodes/memnodes.h

@@ -128,8 +128,8 @@ typedef struct MemoryContextData
 	MemoryContext firstchild;	/* head of linked list of children */
 	MemoryContext prevchild;	/* previous child of same parent */
 	MemoryContext nextchild;	/* next child of same parent */
-	const char *name;			/* context name (just for debugging) */
-	const char *ident;			/* context ID if any (just for debugging) */
+	const char *name;			/* context name */
+	const char *ident;			/* context ID if any */
 	MemoryContextCallback *reset_cbs;	/* list of reset/delete callbacks */
 } MemoryContextData;
 

src/test/regress/expected/rules.out

@@ -1308,12 +1308,13 @@ pg_backend_memory_contexts| SELECT name,
     parent,
     type,
     level,
+    path,
     total_bytes,
     total_nblocks,
     free_bytes,
     free_chunks,
     used_bytes
-   FROM pg_get_backend_memory_contexts() pg_get_backend_memory_contexts(name, ident, parent, type, level, total_bytes, total_nblocks, free_bytes, free_chunks, used_bytes);
+   FROM pg_get_backend_memory_contexts() pg_get_backend_memory_contexts(name, ident, parent, type, level, path, total_bytes, total_nblocks, free_bytes, free_chunks, used_bytes);
 pg_config| SELECT name,
     setting
    FROM pg_config() pg_config(name, setting);

src/test/regress/expected/sysviews.out

@@ -22,10 +22,10 @@ select count(*) >= 0 as ok from pg_available_extensions;
 -- The entire output of pg_backend_memory_contexts is not stable,
 -- we test only the existence and basic condition of TopMemoryContext.
 select type, name, ident, parent, level, total_bytes >= free_bytes
-  from pg_backend_memory_contexts where level = 0;
+  from pg_backend_memory_contexts where level = 1;
    type   |       name       | ident | parent | level | ?column? 
 ----------+------------------+-------+--------+-------+----------
- AllocSet | TopMemoryContext |       |        |     0 | t
+ AllocSet | TopMemoryContext |       |        |     1 | t
 (1 row)
 
 -- We can exercise some MemoryContext type stats functions.  Most of the
@@ -51,6 +51,20 @@ from pg_backend_memory_contexts where name = 'Caller tuples';
 (1 row)
 
 rollback;
+-- Further sanity checks on pg_backend_memory_contexts.  We expect
+-- CacheMemoryContext to have multiple children.  Ensure that's the case.
+with contexts as (
+  select * from pg_backend_memory_contexts
+)
+select count(*) > 1
+from contexts c1, contexts c2
+where c2.name = 'CacheMemoryContext'
+and c1.path[c2.level] = c2.path[c2.level];
+ ?column? 
+----------
+ t
+(1 row)
+
 -- At introduction, pg_config had 23 entries; it may grow
 select count(*) > 20 as ok from pg_config;
  ok