YSQL: Add attributes (jsonb) to yb_tablet_metadata

docs/content/stable/explore/observability/yb-tablet-metadata.md

@@ -37,6 +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. |
+| attributes | jsonb | Per-replica metrics as a JSONB object. Contains a `"replicas"` key mapping each replica address to its `active_sst_sizes` and `wal_sizes` (in bytes). |
 
 ## Examples
 
@@ -53,16 +54,16 @@ SELECT * FROM yb_tablet_metadata WHERE db_name = 'yugabyte' AND relname = 'test_
 ```
 
 ```output
-+----------------------------------+-------+----------+-------------+-----------------+---------------+------------------+--------------------------------------------------------+
-| tablet_id                        | oid   | db_name  | relname     | start_hash_code | end_hash_code | leader           | replicas                                               |
-|----------------------------------+-------+----------+-------------+-----------------+---------------+------------------+--------------------------------------------------------|
-| 3987b6a16bf94fbd92262744197350d7 | 16384 | yugabyte | test_table  | 0               | 10922         | 127.0.0.2:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] |
-| dd50b59c7dcb493680093ffa5b195634 | 16384 | yugabyte | test_table  | 10922           | 21845         | 127.0.0.1:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] |
-| bed5b3c3eee747e99622a4e21acf437a | 16384 | yugabyte | test_table  | 21845           | 32768         | 127.0.0.3:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] |
-| da4bad5faa9f448f890cce57c775cd94 | 16384 | yugabyte | test_table  | 32768           | 43690         | 127.0.0.2:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] |
-| 52176c704c614846bbd80f481678519e | 16384 | yugabyte | test_table  | 43690           | 54613         | 127.0.0.1:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] |
-| ea252119fe774ba9bdc585504fae9398 | 16384 | yugabyte | test_table  | 54613           | 65536         | 127.0.0.3:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] |
-+----------------------------------+-------+----------+-------------+-----------------+---------------+------------------+--------------------------------------------------------+
++----------------------------------+-------+----------+-------------+-----------------+---------------+------------------+--------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
+| tablet_id                        | oid   | db_name  | relname     | start_hash_code | end_hash_code | leader           | replicas                                               | attributes                                                                                                                                                                                      |
+|----------------------------------+-------+----------+-------------+-----------------+---------------+------------------+--------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| 3987b6a16bf94fbd92262744197350d7 | 16384 | yugabyte | test_table  | 0               | 10922         | 127.0.0.2:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] | {"replicas": {"127.0.0.1:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.2:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.3:5433": {"active_sst_sizes": 0, "wal_sizes": 0}}} |
+| dd50b59c7dcb493680093ffa5b195634 | 16384 | yugabyte | test_table  | 10922           | 21845         | 127.0.0.1:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] | {"replicas": {"127.0.0.1:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.2:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.3:5433": {"active_sst_sizes": 0, "wal_sizes": 0}}} |
+| bed5b3c3eee747e99622a4e21acf437a | 16384 | yugabyte | test_table  | 21845           | 32768         | 127.0.0.3:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] | {"replicas": {"127.0.0.1:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.2:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.3:5433": {"active_sst_sizes": 0, "wal_sizes": 0}}} |
+| da4bad5faa9f448f890cce57c775cd94 | 16384 | yugabyte | test_table  | 32768           | 43690         | 127.0.0.2:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] | {"replicas": {"127.0.0.1:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.2:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.3:5433": {"active_sst_sizes": 0, "wal_sizes": 0}}} |
+| 52176c704c614846bbd80f481678519e | 16384 | yugabyte | test_table  | 43690           | 54613         | 127.0.0.1:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] | {"replicas": {"127.0.0.1:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.2:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.3:5433": {"active_sst_sizes": 0, "wal_sizes": 0}}} |
+| ea252119fe774ba9bdc585504fae9398 | 16384 | yugabyte | test_table  | 54613           | 65536         | 127.0.0.3:5433   | ['127.0.0.1:5433', '127.0.0.2:5433', '127.0.0.3:5433'] | {"replicas": {"127.0.0.1:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.2:5433": {"active_sst_sizes": 0, "wal_sizes": 0}, "127.0.0.3:5433": {"active_sst_sizes": 0, "wal_sizes": 0}}} |
++----------------------------------+-------+----------+-------------+-----------------+---------------+------------------+--------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 ```
 
 ### Join with yb_servers
@@ -268,4 +269,3 @@ SELECT
 | 99ed7472ccde4af787cb0bcfd4fd90bd | yugabyte | test_table       | 0               | 32768         | RocksDB_NewIterator                | TServer              | DiskIO          | 1     |
 +----------------------------------+----------+------------------+-----------------+---------------+------------------------------------+----------------------+-----------------+-------+
 ```
-

src/postgres/src/backend/catalog/yb_system_views.sql

@@ -67,7 +67,8 @@ CREATE VIEW yb_tablet_metadata AS
         t.start_hash_code,
         t.end_hash_code,
         t.leader,
-        t.replicas
+        t.replicas,
+        t.attributes
     FROM
         yb_get_tablet_metadata() t
     LEFT JOIN

src/postgres/src/backend/utils/misc/pg_yb_utils.c

@@ -118,6 +118,7 @@
 #endif
 #include "storage/procarray.h"
 #include "tcop/utility.h"
+#include "utils/array.h"
 #include "utils/builtins.h"
 #include "utils/datum.h"
 #include "utils/fmgroids.h"
@@ -8482,10 +8483,13 @@ string_list_compare(const ListCell *a, const ListCell *b)
  * - end_hash_code: int32
  * - leader: text
  * - replicas: text[]
+ * - 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.
+ * 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_sizes, wal_sizes).
  */
 Datum
 yb_get_tablet_metadata(PG_FUNCTION_ARGS)
@@ -8495,7 +8499,7 @@ yb_get_tablet_metadata(PG_FUNCTION_ARGS)
 	Tuplestorestate *tupstore;
 	MemoryContext per_query_ctx;
 	MemoryContext oldcontext;
-	static int	ncols = 9;
+	static int	ncols = 10;
 
 	/* check to see if caller supports us returning a tuplestore */
 	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
@@ -8564,31 +8568,100 @@ yb_get_tablet_metadata(PG_FUNCTION_ARGS)
 			nulls[6] = true;
 		}
 
-		/* Convert replicas array to PostgreSQL text array */
+		/* Convert replicas to PostgreSQL text array and build JSONB attributes */
 		if (tablet->replicas_count > 0)
 		{
+			size_t		nreplicas = tablet->replicas_count;
+
 			Assert(tablet->replicas != NULL);
 
 			/* The last replica is the leader. */
-			values[7] = CStringGetTextDatum(tablet->replicas[tablet->replicas_count - 1]);
+			values[7] = CStringGetTextDatum(tablet->replicas[nreplicas - 1]);
 
 			/* Convert char ** to List * */
-			List	   *replicas_list = NIL;
+			{
+				List	   *replicas_list = NIL;
 
-			for (size_t idx = 0; idx < tablet->replicas_count; idx++)
-				replicas_list = lappend(replicas_list, (char *) tablet->replicas[idx]);
+				for (size_t idx = 0; idx < nreplicas; idx++)
+					replicas_list = lappend(replicas_list,
+											(char *) tablet->replicas[idx]);
 
-			/*
-			 * Sort the list lexicographically for consistency, so that all rows
-			 * with same replicas have same entries.
-			 */
-			list_sort(replicas_list, string_list_compare);
-			values[8] = PointerGetDatum(strlist_to_textarray(replicas_list));
+				/*
+				 * Sort the list lexicographically for consistency, so that
+				 * all rows with same replicas have same entries.
+				 */
+				list_sort(replicas_list, string_list_compare);
+				values[8] = PointerGetDatum(strlist_to_textarray(replicas_list));
+			}
+
+			/* Build JSONB attributes with per-replica size metrics. */
+			{
+				JsonbParseState *jb_state = NULL;
+				JsonbValue	jb_result;
+				JsonbValue	jb_key;
+				JsonbValue	jb_val;
+
+				/* {"replicas": {"addr": {"active_sst_sizes": N, "wal_sizes": 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++)
+				{
+					int64		sst_size;
+					int64		wal_size;
+
+					sst_size = tablet->replica_sst_sizes
+						? tablet->replica_sst_sizes[idx] : 0;
+					wal_size = tablet->replica_wal_sizes
+						? tablet->replica_wal_sizes[idx] : 0;
+
+					/* Replica address as key */
+					jb_key.type = jbvString;
+					jb_key.val.string.val = (char *) tablet->replicas[idx];
+					jb_key.val.string.len = strlen(tablet->replicas[idx]);
+					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_sizes";
+					jb_key.val.string.len = 16;
+					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_sizes";
+					jb_key.val.string.len = 9;
+					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[9] = JsonbPGetDatum(JsonbValueToJsonb(&jb_result));
+			}
 		}
 		else
 		{
 			nulls[7] = true;
 			nulls[8] = true;
+			nulls[9] = true;
 		}
 
 		tuplestore_putvalues(tupstore, tupdesc, values, nulls);

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}',
-  proargnames => '{tablet_id,object_uuid,namespace,object_name,type,start_hash_code,end_hash_code,leader,replicas}',
-  proargmodes => '{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',

src/postgres/src/include/catalog/pg_yb_migration.dat

@@ -12,7 +12,7 @@
 [
 
 # For better version control conflict detection, list latest migration filename
-# here: V100__30133__ysql_conn_mgr_coarse_grained_logical_client_version.sql
-{ major => '100', minor => '0', name => '<baseline>', time_applied => '_null_' }
+# here: V101__29665__yb_tablet_metadata_sizes.sql
+{ major => '101', minor => '0', name => '<baseline>', time_applied => '_null_' }
 
 ]

src/postgres/src/test/regress/expected/yb.orig.tablet_metadata.out

@@ -15,6 +15,54 @@ ORDER BY start_hash_code NULLS FIRST;
  test_table_1 | yugabyte |           32768 |         65536
 (3 rows)
 
+-- Test that attributes column is present with correct type and structure
+SELECT
+    relname,
+    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    | 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 replica, (attrs->>'active_sst_sizes')::bigint AS active_sst_sizes,
+                (attrs->>'wal_sizes')::bigint AS wal_sizes
+FROM yb_tablet_metadata tm,
+     unnest(tm.replicas) AS replica,
+     LATERAL (SELECT tm.attributes->'replicas'->replica AS attrs) a
+WHERE tm.relname = 'test_table_1'
+  AND ((attrs->>'active_sst_sizes')::bigint < 0
+    OR (attrs->>'wal_sizes')::bigint < 0);
+ replica | active_sst_sizes | wal_sizes 
+---------+------------------+-----------
+(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
+WHERE relname = 'test_table_1'
+EXCEPT
+SELECT host || ':' || port FROM yb_servers();
+ orphan_replica 
+----------------
+(0 rows)
+
 -- Test that we are able to join with yb_servers()
 SELECT
     ytm.relname,

src/postgres/src/test/regress/sql/yb.orig.tablet_metadata.sql

@@ -10,6 +10,41 @@ SELECT
 FROM yb_tablet_metadata WHERE relname IN ('test_table_1', 'test_table_2')
 ORDER BY start_hash_code NULLS FIRST;
 
+-- Test that attributes column is present with correct type and structure
+SELECT
+    relname,
+    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 replica, (attrs->>'active_sst_sizes')::bigint AS active_sst_sizes,
+                (attrs->>'wal_sizes')::bigint AS wal_sizes
+FROM yb_tablet_metadata tm,
+     unnest(tm.replicas) AS replica,
+     LATERAL (SELECT tm.attributes->'replicas'->replica AS attrs) a
+WHERE tm.relname = 'test_table_1'
+  AND ((attrs->>'active_sst_sizes')::bigint < 0
+    OR (attrs->>'wal_sizes')::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
+FROM yb_tablet_metadata
+WHERE relname = 'test_table_1'
+EXCEPT
+SELECT host || ':' || port FROM yb_servers();
+
 -- Test that we are able to join with yb_servers()
 SELECT
     ytm.relname,

src/yb/master/catalog_manager.cc

@@ -14367,6 +14367,8 @@ void PopulateTabletMetadata(
   }
 
   std::string leader_address;
+  uint64_t leader_sst_files_size = 0;
+  uint64_t leader_wal_files_size = 0;
   auto replica_locations = tablet->GetReplicaLocations();
   for (const auto& [ts_uuid, replica] : *replica_locations) {
     auto ts_desc_result = ts_manager->LookupTSByUUID(ts_uuid);
@@ -14392,15 +14394,21 @@ void PopulateTabletMetadata(
 
       if (replica.role == PeerRole::LEADER) {
         leader_address = server_address;
+        leader_sst_files_size = replica.drive_info.sst_files_size;
+        leader_wal_files_size = replica.drive_info.wal_files_size;
       } else {
         tablet_metadata->add_replicas(server_address);
+        tablet_metadata->add_replica_sst_sizes(replica.drive_info.sst_files_size);
+        tablet_metadata->add_replica_wal_sizes(replica.drive_info.wal_files_size);
       }
     }
   }
 
   // Add leader as the last replica
   if (!leader_address.empty()) {
     tablet_metadata->add_replicas(leader_address);
+    tablet_metadata->add_replica_sst_sizes(leader_sst_files_size);
+    tablet_metadata->add_replica_wal_sizes(leader_wal_files_size);
   }
 
   auto tablet_lock = tablet->LockForRead();

src/yb/tablet/tablet.proto

@@ -76,6 +76,9 @@ message TabletStatusPB {
   // Leader is the last replica in the list.
   repeated string replicas = 24;
   optional bool is_hash_partitioned = 25;
+  // Per-replica SST and WAL sizes in bytes, parallel to the replicas array.
+  repeated uint64 replica_sst_sizes = 26;
+  repeated uint64 replica_wal_sizes = 27;
 }
 
 // Used to present the maintenance manager's internal state.

src/yb/yql/pggate/ybc_pg_typedefs.h

@@ -875,6 +875,8 @@ typedef struct {
   const char** replicas;
   size_t replicas_count;
   bool is_hash_partitioned;
+  const uint64_t* replica_sst_sizes;
+  const uint64_t* replica_wal_sizes;
 } YbcPgGlobalTabletsDescriptor;
 
 typedef struct {