spec(2.4) Sprint A Step 5: SRF/view 23 col + catversion 220 + 2 SQLSTATE + regress

D7 + D11 + D12 + D20.  Atomic catalog change commit (per L46:
catversion bump must cover SRF/view extension + new GUCs +
new SQLSTATE).

Catversion (L46 lower-bound semantics):
  catversion.h    202605210 -> 202605220
  + spec body amend explaining the bump motivation
  + downstream TAP / regress regex are all lower-bound, so 075
    L6/L7, 068 L12, 069 L12 continue to pass without edits.

pg_proc.dat:
  cluster_get_ic_peers SRF (oid 8914):
    proallargtypes: 19 -> 23 (+ int8 + int4 + int8 + int8)
    proargmodes:    19 -> 23 (all 'o' output)
    proargnames:    19 -> 23 (+ stale_epoch_drop_count +
                              chunk_reassembly_active +
                              chunk_reassembly_timeout_count +
                              lamport_observe_advance_count)
    descr appends "+ spec-2.4"

system_views.sql:
  pg_cluster_ic_peers view: 19 -> 23 column SELECT.  Trailing 4
  columns appended after last_error.

cluster_ic_tier1.c::cluster_get_ic_peers SRF body:
  values[19] -> values[23];nulls[19] -> nulls[23]
  Assert(col == 23)
  4 NEW columns filled from per-peer atomic counters:
    stale_epoch_drop_count          (int8;atomic_read_u64)
    chunk_reassembly_active         (int4;atomic_read_u32)
    chunk_reassembly_timeout_count  (int8;atomic_read_u64)
    lamport_observe_advance_count   (int8;atomic_read_u64)

errcodes.txt -- 2 NEW SQLSTATE:
  53R20 ERRCODE_CLUSTER_IC_STALE_EPOCH_DROP
  53R21 ERRCODE_CLUSTER_IC_CHUNK_REASSEMBLY_TIMEOUT
  Both appear in 53R block contiguous with existing cluster
  errcodes (53R01 .. 53R12);PG codegen rebuilds errcodes.h /
  errcodes-table.h on next build.

cluster_ic_envelope.c step 7 epoch reject:
  errcode upgraded ERRCODE_INTERNAL_ERROR -> ERRCODE_CLUSTER_IC_STALE_EPOCH_DROP

cluster_ic_chunk.c reassembly timeout:
  errcode upgraded ERRCODE_INTERNAL_ERROR -> ERRCODE_CLUSTER_IC_CHUNK_REASSEMBLY_TIMEOUT
  + cluster_ic_tier1_close_peer call after timeout (force reconnect)

cluster_regress ic_envelope_smoke.sql + expected:
  spec-2.3 4 SQL blocks unchanged
  + block 5: spec-2.4 D9 5 NEW PGC_POSTMASTER GUCs visible
  + block 6: pg_cluster_ic_peers view = 23 columns

src/test/regress/expected/rules.out:
  pg_cluster_ic_peers view dump updated to 23 columns
  (PG view rule capture).

cluster_tap 075 (L6/L7) + 068 (L12) + 069 (L12):
  All catversion regex are lower-bound (>= 202605200/210/181)
  so 220 satisfies all three -- NO test edits needed.

cluster_tap 082_envelope_epoch_enforce.pl + 083_envelope_chunked_round_trip.pl
  spec-2.4 D17/D18 NEW negative tests deferred:
  - 082 epoch fault inject requires raw socket negative test
    pattern not yet implemented;cluster_unit U10 fully covers
    the verify reject branch.
  - 083 chunked round-trip requires tier1 recv variable-length
    payload extension (Step 4 wired send-side + reassembly state
    machine but not the recv-side variable-length frame read).
    Will land in spec-2.4 hardening or spec-2.13 GES caller use.
  Production protection still applies via cluster_unit (envelope
  step 7 epoch enforce + chunk_reset_peer cleanup).

Verification surfaces (local):
  - cluster_unit:    34/34 binaries
  - cluster_regress: 5/5 (ic_envelope_smoke output extended)
  - PG regression:   219/219 (rules.out updated)
  - clang-format-18: 0 violations
  - cppcheck:        0 new findings

Spec authority: pgrac:specs/spec-2.4-* §1.2 D7 catversion + D11
SRF/view + D12 SQLSTATE + D20 regress (frozen v0.2 2026-05-08).

Spec: spec-2.4-framing-epoch-enforce-lamport-piggyback.md

Author: sqlrush

src/backend/catalog/system_views.sql

@@ -1488,7 +1488,11 @@ CREATE VIEW pg_cluster_ic_peers AS
            connect_error_count,
            last_errno,
            last_error_code,
-           last_error
+           last_error,
+           stale_epoch_drop_count,
+           chunk_reassembly_active,
+           chunk_reassembly_timeout_count,
+           lamport_observe_advance_count
       FROM cluster_get_ic_peers();
 
 REVOKE ALL ON pg_cluster_ic_peers FROM PUBLIC;

src/backend/cluster/cluster_ic_chunk.c

@@ -334,13 +334,17 @@ cluster_ic_chunk_scan_reassembly_timeouts(void)
 		if (elapsed_ms > cluster_interconnect_chunk_reassembly_timeout_ms) {
 			cluster_ic_tier1_bump_chunk_reassembly_timeout((int32)peer);
 			ereport(WARNING,
-					(errcode(ERRCODE_INTERNAL_ERROR),
+					(errcode(ERRCODE_CLUSTER_IC_CHUNK_REASSEMBLY_TIMEOUT),
 					 errmsg("cluster_ic chunk reassembly timeout for peer %d "
 							"(elapsed %ld ms > %d ms threshold)",
 							peer, elapsed_ms, cluster_interconnect_chunk_reassembly_timeout_ms)));
 			cluster_ic_chunk_reset_peer((int32)peer);
-			/* spec-2.4 D8 close-peer hook lands here in Step 4 wiring;
-			 * this Step 3 commit only handles the reassembly cleanup. */
+			/*
+			 * spec-2.4 D8 (already shipped in Step 4):
+			 * cluster_ic_tier1_close_peer's hook calls chunk_reset_peer;
+			 * here we close the peer to force reconnection on timeout.
+			 */
+			cluster_ic_tier1_close_peer((int32)peer, "chunk reassembly timeout");
 		}
 	}
 }

src/backend/cluster/cluster_ic_envelope.c

@@ -228,7 +228,7 @@ cluster_ic_envelope_verify(const ClusterICEnvelope *env, const void *payload, ui
 		memcpy(&env_epoch, &env->epoch, sizeof(uint64)); /* L34 unaligned */
 		if (env_epoch != my_epoch) {
 			cluster_ic_tier1_bump_stale_epoch_drop((int32)env->source_node_id);
-			ereport(LOG, (errcode(ERRCODE_INTERNAL_ERROR),
+			ereport(LOG, (errcode(ERRCODE_CLUSTER_IC_STALE_EPOCH_DROP),
 						  errmsg("cluster_ic dropped envelope from node %u: "
 								 "stale epoch " UINT64_FORMAT " != current " UINT64_FORMAT,
 								 env->source_node_id, env_epoch, my_epoch),

src/backend/cluster/cluster_ic_tier1.c

@@ -1574,8 +1574,8 @@ cluster_get_ic_peers(PG_FUNCTION_ARGS)
 
 	for (i = 0; i < CLUSTER_MAX_NODES; i++) {
 		ClusterICPeerStateShmem *p = &Tier1Shmem->peers[i];
-		Datum values[19];
-		bool nulls[19];
+		Datum values[23];
+		bool nulls[23];
 		int col = 0;
 
 		if (cluster_conf_lookup_node(i) == NULL)
@@ -1611,8 +1611,14 @@ cluster_get_ic_peers(PG_FUNCTION_ARGS)
 		values[col++] = Int32GetDatum(p->last_errno);
 		values[col++] = CStringGetTextDatum(p->last_error_code[0] ? p->last_error_code : "");
 		values[col++] = CStringGetTextDatum(p->last_error[0] ? p->last_error : "");
-
-		Assert(col == 19);
+		/* spec-2.4 D11: 4 NEW columns (19 -> 23). */
+		values[col++] = Int64GetDatum((int64)pg_atomic_read_u64(&p->stale_epoch_drop_count));
+		values[col++] = Int32GetDatum((int32)pg_atomic_read_u32(&p->chunk_reassembly_active));
+		values[col++]
+			= Int64GetDatum((int64)pg_atomic_read_u64(&p->chunk_reassembly_timeout_count));
+		values[col++] = Int64GetDatum((int64)pg_atomic_read_u64(&p->lamport_observe_advance_count));
+
+		Assert(col == 23);
 		tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
 	}
 

src/backend/utils/errcodes.txt

@@ -460,6 +460,8 @@ Section: Class 53 - Insufficient Resources (pgrac extension)
 53R10    E    ERRCODE_CLUSTER_STATS_SPAWN_FAILED                             cluster_stats_spawn_failed
 53R11    E    ERRCODE_CLUSTER_STATS_NOT_READY                                cluster_stats_not_ready
 53R12    E    ERRCODE_CLUSTER_SCN_WRAPAROUND_PANIC                           cluster_scn_wraparound_panic
+53R20    E    ERRCODE_CLUSTER_IC_STALE_EPOCH_DROP                            cluster_ic_stale_epoch_drop
+53R21    E    ERRCODE_CLUSTER_IC_CHUNK_REASSEMBLY_TIMEOUT                    cluster_ic_chunk_reassembly_timeout
 
 Section: Class 54 - Program Limit Exceeded
 

src/include/catalog/catversion.h

@@ -194,6 +194,18 @@
  * remain lower-bound (>= 202605210) not equal-bound; see grep
  * audit in spec-2.3 §7 DoD.
  */
-#define CATALOG_VERSION_NO	202605210
+/*
+ * spec-2.4 D7 (2026-05-08) -- bump 202605210 -> 202605220 for
+ * pg_cluster_ic_peers SRF/view extension to 23 columns
+ * (+ stale_epoch_drop_count + chunk_reassembly_active +
+ * chunk_reassembly_timeout_count + lamport_observe_advance_count) +
+ * 5 NEW PGC_POSTMASTER GUCs (interconnect_payload_max_bytes +
+ * chunk_reassembly_timeout_ms + tcp_keepidle_sec / keepintvl_sec /
+ * keepcnt) + 2 NEW SQLSTATE (53R20 + 53R21).  Per L46:downstream
+ * TAP / regress hardcoded catversion regex MUST remain lower-bound
+ * (>= 202605220) not equal-bound;see grep audit in spec-2.4 §7
+ * DoD.
+ */
+#define CATALOG_VERSION_NO 202605220
 
 #endif

src/include/catalog/pg_proc.dat

@@ -12248,13 +12248,13 @@
 # TRANSPORT-LEVEL liveness only (connecting / connected / down /
 # rejected); does NOT map to membership / quorum / fence (those land
 # in spec-2.5+ / 2.6+ / 2.28+).
-{ oid => '8914', descr => 'list cluster IC peer connection state + telemetry (spec-2.2)',
+{ oid => '8914', descr => 'list cluster IC peer connection state + telemetry (spec-2.2 + spec-2.4)',
   proname => 'cluster_get_ic_peers', prorows => '8',
   proretset => 't', provolatile => 'v', proparallel => 'r',
   prorettype => 'record', proargtypes => '',
-  proallargtypes => '{int4,text,text,timestamptz,timestamptz,timestamptz,timestamptz,timestamptz,int8,int8,int8,int8,int8,int8,int4,int4,int4,text,text}',
-  proargmodes => '{o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
-  proargnames => '{node_id,state,interconnect_addr,last_connect_at,last_send_at,last_recv_at,last_heartbeat_sent_at,last_heartbeat_recv_at,heartbeat_send_count,heartbeat_recv_count,msg_send_count,msg_recv_count,bytes_send,bytes_recv,reconnect_count,connect_error_count,last_errno,last_error_code,last_error}',
+  proallargtypes => '{int4,text,text,timestamptz,timestamptz,timestamptz,timestamptz,timestamptz,int8,int8,int8,int8,int8,int8,int4,int4,int4,text,text,int8,int4,int8,int8}',
+  proargmodes => '{o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
+  proargnames => '{node_id,state,interconnect_addr,last_connect_at,last_send_at,last_recv_at,last_heartbeat_sent_at,last_heartbeat_recv_at,heartbeat_send_count,heartbeat_recv_count,msg_send_count,msg_recv_count,bytes_send,bytes_recv,reconnect_count,connect_error_count,last_errno,last_error_code,last_error,stale_epoch_drop_count,chunk_reassembly_active,chunk_reassembly_timeout_count,lamport_observe_advance_count}',
   prosrc => 'cluster_get_ic_peers' },
 
 # spec-2.3 D8 cluster_get_ic_msg_types -- list registered IC message

src/test/cluster_regress/expected/ic_envelope_smoke.out

@@ -72,3 +72,37 @@ SELECT msg_type, name, handler_present, broadcast_ok
         1 | heartbeat | t               | f
 (1 row)
 
+-- ----------
+-- 5. spec-2.4 D9 NEW 5 PGC_POSTMASTER GUCs (chunked + KeepAlive).
+-- ----------
+SELECT name, vartype, context
+  FROM pg_settings
+ WHERE name IN ('cluster.interconnect_payload_max_bytes',
+                'cluster.interconnect_chunk_reassembly_timeout_ms',
+                'cluster.interconnect_tcp_keepidle_sec',
+                'cluster.interconnect_tcp_keepintvl_sec',
+                'cluster.interconnect_tcp_keepcnt')
+ ORDER BY name;
+                       name                       | vartype |  context   
+--------------------------------------------------+---------+------------
+ cluster.interconnect_chunk_reassembly_timeout_ms | integer | postmaster
+ cluster.interconnect_payload_max_bytes           | integer | postmaster
+ cluster.interconnect_tcp_keepcnt                 | integer | postmaster
+ cluster.interconnect_tcp_keepidle_sec            | integer | postmaster
+ cluster.interconnect_tcp_keepintvl_sec           | integer | postmaster
+(5 rows)
+
+-- ----------
+-- 6. spec-2.4 D11 pg_cluster_ic_peers view extended to 23 columns
+--	(was 19 in spec-2.2 D9).  4 NEW columns: stale_epoch_drop_count /
+--	chunk_reassembly_active / chunk_reassembly_timeout_count /
+--	lamport_observe_advance_count.
+-- ----------
+SELECT count(*) AS column_count
+  FROM information_schema.columns
+ WHERE table_name = 'pg_cluster_ic_peers';
+ column_count 
+--------------
+           23
+(1 row)
+

src/test/cluster_regress/expected/ic_tcp_smoke.out

@@ -53,6 +53,6 @@ SELECT count(*) AS column_count
  WHERE table_name = 'pg_cluster_ic_peers';
  column_count 
 --------------
-           19
+           23
 (1 row)
 

src/test/cluster_regress/sql/ic_envelope_smoke.sql

@@ -57,3 +57,27 @@ SELECT count(*) AS column_count
 SELECT msg_type, name, handler_present, broadcast_ok
   FROM pg_cluster_ic_msg_types
  WHERE msg_type = 1;
+
+
+-- ----------
+-- 5. spec-2.4 D9 NEW 5 PGC_POSTMASTER GUCs (chunked + KeepAlive).
+-- ----------
+SELECT name, vartype, context
+  FROM pg_settings
+ WHERE name IN ('cluster.interconnect_payload_max_bytes',
+                'cluster.interconnect_chunk_reassembly_timeout_ms',
+                'cluster.interconnect_tcp_keepidle_sec',
+                'cluster.interconnect_tcp_keepintvl_sec',
+                'cluster.interconnect_tcp_keepcnt')
+ ORDER BY name;
+
+
+-- ----------
+-- 6. spec-2.4 D11 pg_cluster_ic_peers view extended to 23 columns
+--	(was 19 in spec-2.2 D9).  4 NEW columns: stale_epoch_drop_count /
+--	chunk_reassembly_active / chunk_reassembly_timeout_count /
+--	lamport_observe_advance_count.
+-- ----------
+SELECT count(*) AS column_count
+  FROM information_schema.columns
+ WHERE table_name = 'pg_cluster_ic_peers';