Expose tablet replica attributes as JSONB

keisku · keisku · commit ba894edba62f · 2026-02-13T16:44:52.000+09:00
diff --git a/docs/content/stable/explore/observability/yb-tablet-metadata.md b/docs/content/stable/explore/observability/yb-tablet-metadata.md
@@ -37,8 +37,7 @@ The following table describes the columns of the `yb_tablet_metadata` view.
 | end_hash_code | int | Ending hash code (exclusive) for the tablet. (NULL for range-sharded tables.) |
 | leader | text | IP address, port of the leader node for the tablet. |
 | replicas | text[] | A list of replica IP addresses and port (includes leader) associated with the tablet. |
-| active_ssts_size | bigint[] | Per-replica SST files size in bytes, in the same order as replicas. |
-| wals_size | bigint[] | Per-replica WAL files size in bytes, in the same order as replicas. |
+| attributes | jsonb | Per-replica metrics as a JSONB object. Contains a `"replicas"` key mapping each replica address to its `active_sst_size` and `wal_size` (in bytes). |
 
 ## Examples
 
diff --git a/src/postgres/src/backend/catalog/yb_system_views.sql b/src/postgres/src/backend/catalog/yb_system_views.sql
@@ -68,8 +68,7 @@ CREATE VIEW yb_tablet_metadata AS
         t.end_hash_code,
         t.leader,
         t.replicas,
-        t.active_sst_sizes,
-        t.wal_sizes
+        t.attributes
     FROM
         yb_get_tablet_metadata() t
     LEFT JOIN
diff --git a/src/postgres/src/backend/utils/misc/pg_yb_utils.c b/src/postgres/src/backend/utils/misc/pg_yb_utils.c
@@ -8419,13 +8419,13 @@ replica_index_cmp(const void *a, const void *b, void *arg)
  * - end_hash_code: int32
  * - leader: text
  * - replicas: text[]
- * - active_sst_sizes: int8[]
- * - wal_sizes: int8[]
+ * - attributes: jsonb  (nested structure with per-replica data)
  *
  * The start_hash_code and end_hash_code are the hash codes of the start and end
  * keys of the tablet for hash sharded tables. Leader is provided as a separate
- * column for simpler querying and self-explanatory access. The size arrays are
- * parallel to replicas (same index refers to the same replica).
+ * column for simpler querying and self-explanatory access. The attributes column
+ * contains a JSONB object with a "replicas" key mapping each replica address to
+ * its metrics (active_sst_size, wal_size).
  */
 Datum
 yb_get_tablet_metadata(PG_FUNCTION_ARGS)
@@ -8435,7 +8435,7 @@ yb_get_tablet_metadata(PG_FUNCTION_ARGS)
 	Tuplestorestate *tupstore;
 	MemoryContext per_query_ctx;
 	MemoryContext oldcontext;
-	static int	ncols = 11;
+	static int	ncols = 10;
 
 	/* check to see if caller supports us returning a tuplestore */
 	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
@@ -8504,23 +8504,24 @@ yb_get_tablet_metadata(PG_FUNCTION_ARGS)
 			nulls[6] = true;
 		}
 
-		/* Convert replicas array to PostgreSQL text array and size arrays */
+		/* Convert replicas array to PostgreSQL text array and JSONB attributes */
 		if (tablet->replicas_count > 0)
 		{
 			size_t		nreplicas = tablet->replicas_count;
 			int		   *sort_idxs;
 			Datum	   *text_elems;
-			Datum	   *sst_elems;
-			Datum	   *wal_elems;
+			JsonbParseState *jb_state = NULL;
+			JsonbValue	jb_result;
+			JsonbValue	jb_key;
+			JsonbValue	jb_val;
 
 			Assert(tablet->replicas != NULL);
 
 			/* The last replica is the leader. */
 			values[7] = CStringGetTextDatum(tablet->replicas[nreplicas - 1]);
 
 			/*
-			 * Build a sort index so replicas and their sizes stay parallel
-			 * after lexicographic sorting.
+			 * Build a sort index so replicas stay in lexicographic order.
 			 */
 			sort_idxs = palloc(nreplicas * sizeof(int));
 			for (size_t idx = 0; idx < nreplicas; idx++)
@@ -8530,43 +8531,76 @@ yb_get_tablet_metadata(PG_FUNCTION_ARGS)
 					  replica_index_cmp, tablet->replicas);
 
 			text_elems = palloc(nreplicas * sizeof(Datum));
-			sst_elems = palloc(nreplicas * sizeof(Datum));
-			wal_elems = palloc(nreplicas * sizeof(Datum));
+
+			/* Build JSONB: {"replicas": {"addr": {"active_sst_size": N, "wal_size": N}, ...}} */
+			pushJsonbValue(&jb_state, WJB_BEGIN_OBJECT, NULL);
+
+			jb_key.type = jbvString;
+			jb_key.val.string.val = "replicas";
+			jb_key.val.string.len = 8;
+			pushJsonbValue(&jb_state, WJB_KEY, &jb_key);
+
+			pushJsonbValue(&jb_state, WJB_BEGIN_OBJECT, NULL);
 
 			for (size_t idx = 0; idx < nreplicas; idx++)
 			{
 				int			si = sort_idxs[idx];
+				int64		sst_size;
+				int64		wal_size;
 
 				text_elems[idx] = CStringGetTextDatum(tablet->replicas[si]);
-				sst_elems[idx] = UInt64GetDatum(
-					tablet->replica_sst_sizes
-						? tablet->replica_sst_sizes[si] : 0);
-				wal_elems[idx] = UInt64GetDatum(
-					tablet->replica_wal_sizes
-						? tablet->replica_wal_sizes[si] : 0);
+
+				sst_size = tablet->replica_sst_sizes
+					? tablet->replica_sst_sizes[si] : 0;
+				wal_size = tablet->replica_wal_sizes
+					? tablet->replica_wal_sizes[si] : 0;
+
+				/* Replica address as key */
+				jb_key.type = jbvString;
+				jb_key.val.string.val = (char *) tablet->replicas[si];
+				jb_key.val.string.len = strlen(tablet->replicas[si]);
+				pushJsonbValue(&jb_state, WJB_KEY, &jb_key);
+
+				/* Per-replica object */
+				pushJsonbValue(&jb_state, WJB_BEGIN_OBJECT, NULL);
+
+				jb_key.type = jbvString;
+				jb_key.val.string.val = "active_sst_size";
+				jb_key.val.string.len = 15;
+				pushJsonbValue(&jb_state, WJB_KEY, &jb_key);
+				jb_val.type = jbvNumeric;
+				jb_val.val.numeric = DatumGetNumeric(
+					DirectFunctionCall1(int8_numeric, Int64GetDatum(sst_size)));
+				pushJsonbValue(&jb_state, WJB_VALUE, &jb_val);
+
+				jb_key.type = jbvString;
+				jb_key.val.string.val = "wal_size";
+				jb_key.val.string.len = 8;
+				pushJsonbValue(&jb_state, WJB_KEY, &jb_key);
+				jb_val.type = jbvNumeric;
+				jb_val.val.numeric = DatumGetNumeric(
+					DirectFunctionCall1(int8_numeric, Int64GetDatum(wal_size)));
+				pushJsonbValue(&jb_state, WJB_VALUE, &jb_val);
+
+				pushJsonbValue(&jb_state, WJB_END_OBJECT, NULL);
 			}
 
+			pushJsonbValue(&jb_state, WJB_END_OBJECT, NULL);	/* end replicas */
+			jb_result = *pushJsonbValue(&jb_state, WJB_END_OBJECT, NULL);	/* end outer */
+
 			values[8] = PointerGetDatum(
 				construct_array(text_elems, (int) nreplicas,
 								TEXTOID, -1, false, TYPALIGN_INT));
-			values[9] = PointerGetDatum(
-				construct_array(sst_elems, (int) nreplicas,
-								INT8OID, 8, FLOAT8PASSBYVAL, TYPALIGN_DOUBLE));
-			values[10] = PointerGetDatum(
-				construct_array(wal_elems, (int) nreplicas,
-								INT8OID, 8, FLOAT8PASSBYVAL, TYPALIGN_DOUBLE));
+			values[9] = JsonbPGetDatum(JsonbValueToJsonb(&jb_result));
 
 			pfree(sort_idxs);
 			pfree(text_elems);
-			pfree(sst_elems);
-			pfree(wal_elems);
 		}
 		else
 		{
 			nulls[7] = true;
 			nulls[8] = true;
 			nulls[9] = true;
-			nulls[10] = true;
 		}
 
 		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
diff --git a/src/postgres/src/include/catalog/pg_proc.dat b/src/postgres/src/include/catalog/pg_proc.dat
@@ -12234,9 +12234,9 @@
   proname => 'yb_get_tablet_metadata', prorows => '100',
   proretset => 't', provolatile => 'v', proparallel => 'r',
   prorettype => 'record', proargtypes => '',
-  proallargtypes => '{text,text,text,text,text,int4,int4,text,_text,_int8,_int8}',
-  proargnames => '{tablet_id,object_uuid,namespace,object_name,type,start_hash_code,end_hash_code,leader,replicas,active_sst_sizes,wal_sizes}',
-  proargmodes => '{o,o,o,o,o,o,o,o,o,o,o}',
+  proallargtypes => '{text,text,text,text,text,int4,int4,text,_text,jsonb}',
+  proargnames => '{tablet_id,object_uuid,namespace,object_name,type,start_hash_code,end_hash_code,leader,replicas,attributes}',
+  proargmodes => '{o,o,o,o,o,o,o,o,o,o}',
   prosrc => 'yb_get_tablet_metadata'},
 
 { oid => '8101', descr => 'Get the UUID of the local tserver',
diff --git a/src/postgres/src/test/regress/expected/yb.orig.tablet_metadata.out b/src/postgres/src/test/regress/expected/yb.orig.tablet_metadata.out
@@ -15,35 +15,44 @@ ORDER BY start_hash_code NULLS FIRST;
  test_table_1 | yugabyte |           32768 |         65536
 (3 rows)
 
--- Test that active_sst_sizes and wal_sizes columns are present and aligned with replicas
+-- Test that attributes column is present with correct type and structure
 SELECT
     relname,
-    pg_typeof(active_sst_sizes) AS sst_type,
-    pg_typeof(wal_sizes) AS wal_type,
-    array_length(active_sst_sizes, 1) = array_length(replicas, 1) AS sst_match,
-    array_length(wal_sizes, 1) = array_length(replicas, 1) AS wal_match
+    pg_typeof(attributes) AS attr_type,
+    jsonb_typeof(attributes->'replicas') AS replicas_type,
+    (SELECT count(*) FROM jsonb_object_keys(attributes->'replicas'))::int
+        = array_length(replicas, 1) AS key_count_match
 FROM yb_tablet_metadata
 WHERE relname = 'test_table_1'
 ORDER BY start_hash_code NULLS FIRST
 LIMIT 1;
-   relname    | sst_type | wal_type | sst_match | wal_match 
---------------+----------+----------+-----------+-----------
- test_table_1 | bigint[] | bigint[] | t         | t
+   relname    | attr_type | replicas_type | key_count_match 
+--------------+-----------+---------------+-----------------
+ test_table_1 | jsonb     | object        | t
 (1 row)
 
 -- Test that all size values are non-negative for each replica
-SELECT r.val AS replica, s.val AS sst_size, w.val AS wal_size
+SELECT replica, (attrs->>'active_sst_size')::bigint AS sst_size,
+                (attrs->>'wal_size')::bigint AS wal_size
 FROM yb_tablet_metadata tm,
-     unnest(tm.replicas)         WITH ORDINALITY AS r(val, ord),
-     unnest(tm.active_sst_sizes) WITH ORDINALITY AS s(val, ord),
-     unnest(tm.wal_sizes)        WITH ORDINALITY AS w(val, ord)
+     unnest(tm.replicas) AS replica,
+     LATERAL (SELECT tm.attributes->'replicas'->replica AS attrs) a
 WHERE tm.relname = 'test_table_1'
-  AND r.ord = s.ord AND r.ord = w.ord
-  AND (s.val < 0 OR w.val < 0);
+  AND ((attrs->>'active_sst_size')::bigint < 0
+    OR (attrs->>'wal_size')::bigint < 0);
  replica | sst_size | wal_size 
 ---------+----------+----------
 (0 rows)
 
+-- Test that every replica in the replicas array has an entry in attributes
+SELECT replica
+FROM yb_tablet_metadata tm, unnest(tm.replicas) AS replica
+WHERE tm.relname = 'test_table_1'
+  AND NOT (tm.attributes->'replicas' ? replica);
+ replica 
+---------
+(0 rows)
+
 -- Test that every replica address corresponds to a live tserver
 SELECT unnest(replicas) AS orphan_replica
 FROM yb_tablet_metadata
diff --git a/src/postgres/src/test/regress/sql/yb.orig.tablet_metadata.sql b/src/postgres/src/test/regress/sql/yb.orig.tablet_metadata.sql
@@ -10,27 +10,33 @@ SELECT
 FROM yb_tablet_metadata WHERE relname IN ('test_table_1', 'test_table_2')
 ORDER BY start_hash_code NULLS FIRST;
 
--- Test that active_sst_sizes and wal_sizes columns are present and aligned with replicas
+-- Test that attributes column is present with correct type and structure
 SELECT
     relname,
-    pg_typeof(active_sst_sizes) AS sst_type,
-    pg_typeof(wal_sizes) AS wal_type,
-    array_length(active_sst_sizes, 1) = array_length(replicas, 1) AS sst_match,
-    array_length(wal_sizes, 1) = array_length(replicas, 1) AS wal_match
+    pg_typeof(attributes) AS attr_type,
+    jsonb_typeof(attributes->'replicas') AS replicas_type,
+    (SELECT count(*) FROM jsonb_object_keys(attributes->'replicas'))::int
+        = array_length(replicas, 1) AS key_count_match
 FROM yb_tablet_metadata
 WHERE relname = 'test_table_1'
 ORDER BY start_hash_code NULLS FIRST
 LIMIT 1;
 
 -- Test that all size values are non-negative for each replica
-SELECT r.val AS replica, s.val AS sst_size, w.val AS wal_size
+SELECT replica, (attrs->>'active_sst_size')::bigint AS sst_size,
+                (attrs->>'wal_size')::bigint AS wal_size
 FROM yb_tablet_metadata tm,
-     unnest(tm.replicas)         WITH ORDINALITY AS r(val, ord),
-     unnest(tm.active_sst_sizes) WITH ORDINALITY AS s(val, ord),
-     unnest(tm.wal_sizes)        WITH ORDINALITY AS w(val, ord)
+     unnest(tm.replicas) AS replica,
+     LATERAL (SELECT tm.attributes->'replicas'->replica AS attrs) a
 WHERE tm.relname = 'test_table_1'
-  AND r.ord = s.ord AND r.ord = w.ord
-  AND (s.val < 0 OR w.val < 0);
+  AND ((attrs->>'active_sst_size')::bigint < 0
+    OR (attrs->>'wal_size')::bigint < 0);
+
+-- Test that every replica in the replicas array has an entry in attributes
+SELECT replica
+FROM yb_tablet_metadata tm, unnest(tm.replicas) AS replica
+WHERE tm.relname = 'test_table_1'
+  AND NOT (tm.attributes->'replicas' ? replica);
 
 -- Test that every replica address corresponds to a live tserver
 SELECT unnest(replicas) AS orphan_replica
diff --git a/src/yb/yql/pgwrapper/ysql_migrations/V97__29665__yb_tablet_metadata_sizes.sql b/src/yb/yql/pgwrapper/ysql_migrations/V97__29665__yb_tablet_metadata_sizes.sql
@@ -1,15 +1,15 @@
 BEGIN;
   SET LOCAL yb_non_ddl_txn_for_sys_tables_allowed TO true;
 
-  -- Add active_sst_sizes and wal_sizes columns to yb_get_tablet_metadata().
+  -- Replace active_sst_sizes and wal_sizes with single attributes jsonb column.
   UPDATE pg_catalog.pg_proc SET
-    proallargtypes = '{25,25,25,25,25,23,23,25,1009,1016,1016}',
-    proargmodes = '{o,o,o,o,o,o,o,o,o,o,o}',
-    proargnames = '{tablet_id,object_uuid,namespace,object_name,type,start_hash_code,end_hash_code,leader,replicas,active_sst_sizes,wal_sizes}'
+    proallargtypes = '{25,25,25,25,25,23,23,25,1009,3802}',
+    proargmodes = '{o,o,o,o,o,o,o,o,o,o}',
+    proargnames = '{tablet_id,object_uuid,namespace,object_name,type,start_hash_code,end_hash_code,leader,replicas,attributes}'
   WHERE proname = 'yb_get_tablet_metadata' AND pronamespace = 11;
 COMMIT;
 
--- Recreate the view to include the new columns.
+-- Recreate the view to include the new column.
 CREATE OR REPLACE VIEW pg_catalog.yb_tablet_metadata
 WITH (use_initdb_acl = true) AS
     SELECT
@@ -28,8 +28,7 @@ WITH (use_initdb_acl = true) AS
         t.end_hash_code,
         t.leader,
         t.replicas,
-        t.active_sst_sizes,
-        t.wal_sizes
+        t.attributes
     FROM
         yb_get_tablet_metadata() t
     LEFT JOIN
