observability: fix local Loki ingest stalls and Alloy flush latency

packages/observability/alloy/config.alloy

@@ -99,5 +99,12 @@ loki.process "filter_old" {
 loki.write "local" {
 	endpoint {
 		url = "http://loki:3100/loki/api/v1/push"
+		// Default batch_wait is 1s, but indexing jobs frequently emit
+		// only 3 progress lines (started / file-visited / finished)
+		// for incremental updates, which doesn't always trip Alloy's
+		// flush trigger in time. Tighter batch_wait keeps low-volume
+		// flushes responsive — at the cost of slightly more HTTP
+		// traffic to Loki, which is fine for local dev.
+		batch_wait = "500ms"
 	}
 }

packages/observability/loki/config.yaml

@@ -30,6 +30,19 @@ schema_config:
         prefix: index_
         period: 24h
 
+# Loki 3.x reads the chunks-store path from `storage_config.filesystem.directory`.
+# When unset, the chunks flusher falls back to a relative `mkdir <tenant>`
+# against Loki's CWD (`/`), which the `loki` user cannot write — producing
+# "mkdir fake: permission denied" errors (the default tenant id is `fake`,
+# not a permission name). The flush failures cascade: chunks pile in memory
+# → ingester ring marks itself unhealthy → Alloy's writes get back 500
+# "empty ring" and silently stall ingest. Verified via the `/config`
+# endpoint that this is the field the chunks flusher actually reads;
+# `common.storage.filesystem.chunks_directory` does NOT propagate here.
+storage_config:
+  filesystem:
+    directory: /loki/chunks
+
 limits_config:
   # Accept Docker logs of any age — local containers may have months of
   # backlog when they're first picked up by Alloy.