From db3b5714fd788cb41693e3f8e215984475a8f131 Mon Sep 17 00:00:00 2001
From: mboyd1 <5233780+mboyd1@users.noreply.github.com>
Date: Mon, 19 Oct 2026 19:36:43 -0500
Subject: [PATCH] fix(cmd): start blob-store age sweeper in the all-in-one
 binary (#267)

The sweeper was wired only into cmd/block-processor, so all-in-one
deployments never age-swept file:// subtree blobs. A production store grew
to 5.5M files in one directory, filled ext4's htree index and failed every
write with ENOSPC at 19% disk usage, stalling SEEN_ON_NETWORK.
---
 cmd/merkle-service/main.go | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/cmd/merkle-service/main.go b/cmd/merkle-service/main.go
index c92ccb5..38d4e4e 100644
--- a/cmd/merkle-service/main.go
+++ b/cmd/merkle-service/main.go
@@ -42,6 +42,14 @@ func main() {
 	}
 	defer func() { _ = registry.Close() }()
 
+	// Blob-store age sweeper: the all-in-one process hosts the block-processor,
+	// so it owns the sweep exactly as cmd/block-processor does. Without it,
+	// subtree blobs whose delete-at-height never fires accumulate forever in
+	// one flat directory until ext4's htree index fills and every write fails
+	// with ENOSPC regardless of free space.
+	stopSweeper := store.StartAgeSweeperFromConfig(registry.Blob, cfg.BlobStore, logger)
+	defer stopSweeper()
+
 	subtreeProducer, err := kafka.NewProducer(cfg.Kafka.Brokers, cfg.Kafka.SubtreeTopic, cfg.Kafka.TopicPartitions(), cfg.Kafka.TopicRetention(), logger)
 	if err != nil {
 		log.Fatal("failed to create subtree producer: ", err)