fix(langfuse): raise clickhouse nofile limit and bound system log retention

docker-compose-langfuse.yml

@@ -152,13 +152,22 @@ services:
     user: "101:101"
     container_name: langfuse-clickhouse
     hostname: langfuse-clickhouse
+    # ClickHouse merges open thousands of sparse subcolumn files at once; the
+    # Docker default nofile (4096) breaks them with errno 24 and the failing
+    # merge is hot-retried forever, pinning CPU cores. Same limit the
+    # observability clickhouse already runs with.
+    ulimits:
+      nofile:
+        soft: 262144
+        hard: 262144
     environment:
       CLICKHOUSE_DB: ${LANGFUSE_CLICKHOUSE_DB:-default}
       CLICKHOUSE_USER: ${LANGFUSE_CLICKHOUSE_USER:-clickhouse}
       CLICKHOUSE_PASSWORD: ${LANGFUSE_CLICKHOUSE_PASSWORD:-clickhouse}
     volumes:
       - langfuse-clickhouse-data:/var/lib/clickhouse
       - langfuse-clickhouse-logs:/var/log/clickhouse-server
+      - ./langfuse/clickhouse/system-logs.xml:/etc/clickhouse-server/config.d/system-logs.xml:ro
     healthcheck:
       test: wget --no-verbose --tries=1 --spider http://localhost:8123/ping || exit 1
       interval: 5s

langfuse/clickhouse/system-logs.xml

@@ -0,0 +1,41 @@
+<!--
+  Bounded retention for ClickHouse system log tables.
+
+  Without TTLs the system tables (trace_log above all) grow unbounded, which
+  keeps scheduling giant merges over tens of thousands of historic parts. With
+  the nofile limit raised these merges complete, but they still should not
+  exist: analytics-grade tracing only needs the last few days locally.
+-->
+<clickhouse>
+    <logger>
+        <level>warning</level>
+    </logger>
+
+    <trace_log>
+        <ttl>event_date + INTERVAL 3 DAY DELETE</ttl>
+    </trace_log>
+
+    <text_log>
+        <ttl>event_date + INTERVAL 3 DAY DELETE</ttl>
+    </text_log>
+
+    <metric_log>
+        <ttl>event_date + INTERVAL 7 DAY DELETE</ttl>
+    </metric_log>
+
+    <query_log>
+        <ttl>event_date + INTERVAL 7 DAY DELETE</ttl>
+    </query_log>
+
+    <part_log>
+        <ttl>event_date + INTERVAL 7 DAY DELETE</ttl>
+    </part_log>
+
+    <processors_profile_log>
+        <ttl>event_date + INTERVAL 7 DAY DELETE</ttl>
+    </processors_profile_log>
+
+    <asynchronous_metric_log>
+        <ttl>event_date + INTERVAL 7 DAY DELETE</ttl>
+    </asynchronous_metric_log>
+</clickhouse>