deploy/scripts/fleet-status.sh

@@ -0,0 +1,37 @@
+#!/usr/bin/env bash
+# fleet-status.sh — one-shot health snapshot of every clawstor
+# node in the fleet. Run from a workstation with ssh access to
+# all of them.
+#
+# Reports per node:
+#   * daemon service state
+#   * FUSE mount state + layer listing
+#   * each timer's next fire + last result
+#   * blob store byte count
+#
+# Configure via NODES env var (space-separated), default = the
+# current LAN fleet.
+
+set -uo pipefail
+
+NODES=${NODES:-"tank architect morpheus"}
+
+hr() { printf '%.0s─' {1..66}; echo; }
+
+for host in $NODES; do
+  hr
+  echo "▎$host"
+  hr
+  ssh -o ConnectTimeout=5 -o BatchMode=yes "$host" '
+    printf "%-24s " "daemon:";       systemctl --user is-active clawstor-cluster.service 2>&1
+    printf "%-24s " "fuse mount:";   systemctl --user is-active clawstor-fuse.service    2>&1
+    printf "%-24s " "fuse layers:";  ls ~/clawstor-mount 2>/dev/null | tr "\n" " " ; echo
+    printf "%-24s " "blob store bytes:"; du -sb ~/clawstor-deploy/data 2>/dev/null | awk "{print \$1}"
+    for t in clawstor-scrub clawstor-gc clawstor-ref-sweep clawstor-snapshot-rotate; do
+      NEXT=$(systemctl --user list-timers "$t.timer" --no-pager 2>/dev/null | awk "NR==2 {print \$1, \$2, \$3, \$4}")
+      LAST=$(systemctl --user show "$t.service" --no-pager -p Result 2>/dev/null | cut -d= -f2)
+      printf "%-24s next=%-25s last=%s\n" "$t:" "${NEXT:--}" "${LAST:--}"
+    done
+  ' 2>&1 || echo "unreachable"
+done
+hr