104 lines
2.6 KiB
Bash
104 lines
2.6 KiB
Bash
|
#!/usr/bin/env bash
|
||
|
|
set -o errexit
|
||
|
|
set -o nounset
|
||
|
|
set -o pipefail
|
||
|
|
|
||
|
|
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||
|
|
# shellcheck source=00_env.sh
|
||
|
|
. "$SCRIPT_DIR/00_env.sh"
|
||
|
|
|
||
|
|
usage() {
|
||
|
|
printf 'Usage: %s --fs <filesystem>\n' "$(basename "$0")"
|
||
|
|
}
|
||
|
|
|
||
|
|
while [[ "$#" -gt 0 ]]; do
|
||
|
|
case "$1" in
|
||
|
|
--fs)
|
||
|
|
FILESYSTEM="${2:-}"
|
||
|
|
shift 2
|
||
|
|
;;
|
||
|
|
-h|--help)
|
||
|
|
usage
|
||
|
|
exit 0
|
||
|
|
;;
|
||
|
|
*)
|
||
|
|
critical "Unknown argument: $1"
|
||
|
|
usage
|
||
|
|
exit 2
|
||
|
|
;;
|
||
|
|
esac
|
||
|
|
done
|
||
|
|
|
||
|
|
if [[ -z "$FILESYSTEM" ]]; then
|
||
|
|
critical "Missing required --fs <filesystem>"
|
||
|
|
usage
|
||
|
|
exit 2
|
||
|
|
fi
|
||
|
|
|
||
|
|
missing=0
|
||
|
|
for cmd in mmgetstate mmlscluster mmlsfs mmlsdisk mmlsmount mmlsmgr df; do
|
||
|
|
require_cmd "$cmd" || missing=1
|
||
|
|
done
|
||
|
|
|
||
|
|
if [[ "$missing" -ne 0 ]]; then
|
||
|
|
exit 2
|
||
|
|
fi
|
||
|
|
|
||
|
|
issues=0
|
||
|
|
|
||
|
|
section "GPFS daemon state"
|
||
|
|
state_output="$(mmgetstate -a 2>&1 || true)"
|
||
|
|
printf '%s\n' "$state_output" | tee -a "$LOG_FILE"
|
||
|
|
if printf '%s\n' "$state_output" | awk 'NR > 1 && $0 !~ / active / { found=1 } END { exit found ? 0 : 1 }'; then
|
||
|
|
warning "Not all GPFS nodes appear active"
|
||
|
|
fi
|
||
|
|
|
||
|
|
section "Target filesystem definition"
|
||
|
|
if mmlsfs "$FILESYSTEM" 2>&1 | tee -a "$LOG_FILE"; then
|
||
|
|
ok "Filesystem exists: $FILESYSTEM"
|
||
|
|
else
|
||
|
|
critical "Filesystem does not exist or cannot be queried: $FILESYSTEM"
|
||
|
|
exit 1
|
||
|
|
fi
|
||
|
|
|
||
|
|
section "Target filesystem mount state"
|
||
|
|
mount_output="$(mmlsmount "$FILESYSTEM" 2>&1 || true)"
|
||
|
|
printf '%s\n' "$mount_output" | tee -a "$LOG_FILE"
|
||
|
|
if printf '%s\n' "$mount_output" | grep -Eiq 'not mounted|no file systems were found|not found'; then
|
||
|
|
warning "Filesystem may not be mounted anywhere: $FILESYSTEM"
|
||
|
|
fi
|
||
|
|
|
||
|
|
section "Existing disks"
|
||
|
|
if ! mmlsdisk "$FILESYSTEM" 2>&1 | tee -a "$LOG_FILE"; then
|
||
|
|
critical "Unable to list disks for filesystem: $FILESYSTEM"
|
||
|
|
issues=1
|
||
|
|
fi
|
||
|
|
|
||
|
|
section "Filesystem capacity"
|
||
|
|
df -h 2>&1 | awk -v fs="$FILESYSTEM" 'NR == 1 || $0 ~ fs || $0 ~ /gpfs|mmfs/' | tee -a "$LOG_FILE"
|
||
|
|
|
||
|
|
section "Cluster health"
|
||
|
|
if command -v mmhealth >/dev/null 2>&1; then
|
||
|
|
health_output="$(mmhealth cluster show 2>&1 || true)"
|
||
|
|
printf '%s\n' "$health_output" | tee -a "$LOG_FILE"
|
||
|
|
if printf '%s\n' "$health_output" | grep -Eiq 'degraded|failed|down|error|unhealthy'; then
|
||
|
|
warning "Cluster health output indicates a degraded condition"
|
||
|
|
fi
|
||
|
|
else
|
||
|
|
warning "mmhealth command not available, skipping health check"
|
||
|
|
fi
|
||
|
|
|
||
|
|
section "Managers and quorum"
|
||
|
|
mmlsmgr 2>&1 | tee -a "$LOG_FILE" || {
|
||
|
|
critical "Unable to query GPFS manager/quorum information"
|
||
|
|
issues=1
|
||
|
|
}
|
||
|
|
|
||
|
|
if [[ "$issues" -eq 0 ]]; then
|
||
|
|
ok "Precheck completed for filesystem: $FILESYSTEM"
|
||
|
|
exit 0
|
||
|
|
fi
|
||
|
|
|
||
|
|
critical "Precheck found operational validation failures"
|
||
|
|
exit 1
|