feat(cluster): S3 forensics step 1a -- lost-write / GES-timeout / TT-retention attribution

src/backend/cluster/cluster_debug.c

@@ -1743,6 +1743,21 @@ dump_ges(ReturnSetInfo *rsinfo)
 			 fmt_int64((int64)cluster_txw_get_wakeup_count()));
 	emit_row(rsinfo, "ges", "tx_enqueue_timeout_count",
 			 fmt_int64((int64)cluster_txw_get_timeout_count()));
+	/* S3 forensics step 1 — breakdown of the "cluster lock acquire timeout"
+	 * fold: how many surfaced timeouts genuinely ran out the clock vs failed
+	 * immediately on capacity / send / probe (never waited at all). */
+	emit_row(rsinfo, "ges", "ges_timeout_true_wait_count",
+			 fmt_int64((int64)cluster_ges_timeout_true_wait_count()));
+	emit_row(rsinfo, "ges", "ges_timeout_capacity_count",
+			 fmt_int64((int64)cluster_ges_timeout_capacity_count()));
+	emit_row(rsinfo, "ges", "ges_timeout_send_fail_count",
+			 fmt_int64((int64)cluster_ges_timeout_send_fail_count()));
+	emit_row(rsinfo, "ges", "ges_timeout_retransmit_exhausted_count",
+			 fmt_int64((int64)cluster_ges_timeout_retransmit_exhausted_count()));
+	emit_row(rsinfo, "ges", "ges_timeout_native_probe_count",
+			 fmt_int64((int64)cluster_ges_timeout_native_probe_count()));
+	emit_row(rsinfo, "ges", "ges_timeout_master_reject_count",
+			 fmt_int64((int64)cluster_ges_timeout_master_reject_count()));
 }
 
 
@@ -1958,6 +1973,10 @@ dump_pcm(ReturnSetInfo *rsinfo)
 			 fmt_int64((int64)cluster_pcm_get_restore_aba_detected_count()));
 	emit_row(rsinfo, "pcm", "invalidate_parked_grant_pending_count",
 			 fmt_int64((int64)cluster_pcm_get_invalidate_parked_grant_pending_count()));
+	/* S3 forensics step 1b — watermark-advance provenance table insert drops
+	 * (table full): non-zero means prov_query absence is inconclusive. */
+	emit_row(rsinfo, "pcm", "wm_prov_insert_fail_count",
+			 fmt_int64((int64)cluster_pcm_get_wm_prov_insert_fail_count()));
 }
 
 
@@ -2794,6 +2813,13 @@ dump_undo(ReturnSetInfo *rsinfo)
 			 fmt_int64((int64)cluster_tt_slot_retention_off_recycle_count()));
 	emit_row(rsinfo, "undo", "tt_retention_rollover_count",
 			 fmt_int64((int64)cluster_undo_tt_retention_rollover_count()));
+	/* S3 forensics step 1a — TT-rollover failure split (writer 53R9E family). */
+	emit_row(rsinfo, "undo", "tt_rollover_fail_hard_cap_count",
+			 fmt_int64((int64)cluster_undo_tt_rollover_fail_hard_cap_count()));
+	emit_row(rsinfo, "undo", "tt_rollover_fail_extend_count",
+			 fmt_int64((int64)cluster_undo_tt_rollover_fail_extend_count()));
+	emit_row(rsinfo, "undo", "tt_rollover_fail_activate_count",
+			 fmt_int64((int64)cluster_undo_tt_rollover_fail_activate_count()));
 
 	/* spec-6.2: Smart Fusion terminal-authority substrate counters. */
 	emit_row(rsinfo, "undo", "terminal_authority_check_count",