deploy/systemd/README.md

@@ -9,6 +9,7 @@ Systemd **user** units for clawstor.
 | `clawstor-cluster.service` | daemon (gossip + RPC + Prometheus + build cache). Not shipped here — deployed per-node from the fleet playbook. |
 | `clawstor-fuse.service` | Phase 6 read-only FUSE mount at `~/clawstor-mount/`. Depends on `clawstor-cluster.service`. |
 | `clawstor-scrub.service` + `clawstor-scrub.timer` | Weekly (Sun 04:00) BLAKE3 verify every chunk against its manifest. Non-zero exit = integrity failure — surfaced by systemd status. |
+| `clawstor-gc.service` + `clawstor-gc.timer` | Nightly (03:30) orphan-chunk sweep. Override ExecStart via drop-in to add `--evict-to-gb N` for size-cap fleets. Sequenced before the Sunday scrub so scrub reads a fresh layout. |
 
 ## Install `clawstor-fuse.service`
 

deploy/systemd/clawstor-gc.service

@@ -0,0 +1,19 @@
+[Unit]
+Description=Clawstor nightly GC (orphan-chunk sweep + optional size-cap eviction)
+Documentation=https://git.redclaw.dev/clawverse/clawstor
+After=clawstor-cluster.service
+
+[Service]
+Type=oneshot
+# Default: orphan-chunk sweep only (safe on any node).
+# For a size-cap fleet, override with a drop-in:
+#   systemctl --user edit clawstor-gc.service
+#   [Service]
+#   ExecStart=
+#   ExecStart=%h/clawstor-deploy/claw-store --config %h/clawstor-deploy/config.toml cluster-gc --evict-to-gb 200
+ExecStart=%h/clawstor-deploy/claw-store --config %h/clawstor-deploy/config.toml cluster-gc
+StandardOutput=journal
+StandardError=journal
+
+[Install]
+WantedBy=default.target

deploy/systemd/clawstor-gc.timer

@@ -0,0 +1,13 @@
+[Unit]
+Description=Nightly Clawstor GC
+Documentation=https://git.redclaw.dev/clawverse/clawstor
+
+[Timer]
+# 03:30 local, every day. Slots ahead of the Sunday 04:00 scrub so
+# scrub always sees a fresh (post-GC) blob layout.
+OnCalendar=*-*-* 03:30:00
+Persistent=true
+Unit=clawstor-gc.service
+
+[Install]
+WantedBy=timers.target