Update check_xfs.py

This commit is contained in:
gru
2025-12-02 10:54:53 +01:00
parent d12d05378d
commit d747dda3f5

View File

@@ -10,6 +10,10 @@ ESC = {"reset": "\033[0m", "red": "\033[31m", "yellow": "\033[33m", "green": "\0
def color_text(text, color): def color_text(text, color):
return f"{ESC[color]}{text}{ESC['reset']}" return f"{ESC[color]}{text}{ESC['reset']}"
def is_interactive():
"""Detect if running interactively (not Nagios)"""
return sys.stdout.isatty()
def run_cmd(cmd): def run_cmd(cmd):
try: try:
return subprocess.check_output(cmd, shell=True, text=True, stderr=subprocess.STDOUT).strip() return subprocess.check_output(cmd, shell=True, text=True, stderr=subprocess.STDOUT).strip()
@@ -31,11 +35,13 @@ def is_xfs_mount(mount):
return "xfs" in mounts_output.lower() return "xfs" in mounts_output.lower()
def check_disk_usage(mount): def check_disk_usage(mount):
output = run_cmd(f"df -h {mount}") """Use df -P for consistent POSIX output"""
output = run_cmd(f"df -P -h {mount}")
for line in output.splitlines(): for line in output.splitlines():
if mount in line:
parts = line.split() parts = line.split()
usage_str = parts[4] # e.g. '45%' if len(parts) >= 6 and mount in parts[5]: # Target is LAST column in -P
usage_str = parts[4]
if '%' in usage_str:
usage_pct = int(usage_str.strip('%')) usage_pct = int(usage_str.strip('%'))
size = parts[1] size = parts[1]
avail = parts[3] avail = parts[3]
@@ -97,9 +103,9 @@ Examples:
else: else:
target_mounts = find_xfs_mounts() target_mounts = find_xfs_mounts()
# Nagios: No mounts = UNKNOWN (3) # Nagios: No mounts = UNKNOWN (3) - SINGLE LINE
if not target_mounts: if not target_mounts:
print(color_text("UNKNOWN No XFS mount points found or specified|", "yellow")) print("UNKNOWN No XFS mount points found or specified|")
sys.exit(3) sys.exit(3)
warn_threshold = args.warn_threshold warn_threshold = args.warn_threshold
@@ -107,68 +113,76 @@ Examples:
global_status = 0 global_status = 0
all_perfdata = [] all_perfdata = []
# Human-readable output (colorful) # ONLY show verbose colorful output in INTERACTIVE mode
if is_interactive():
print(color_text("Checking XFS filesystems:", "bold")) print(color_text("Checking XFS filesystems:", "bold"))
print(f"Thresholds: WARN>{warn_threshold}%% CRIT>{crit_threshold}%%") print(f"Thresholds: WARN>{warn_threshold}% CRIT>{crit_threshold}%")
print("=" * 60) print("=" * 60)
for mount in target_mounts: for mount in target_mounts:
if is_interactive():
print(f"\n{color_text(mount, 'cyan')}") print(f"\n{color_text(mount, 'cyan')}")
# Validate it's actually XFS # Validate it's actually XFS
if not is_xfs_mount(mount): if not is_xfs_mount(mount):
if is_interactive():
print(color_text(" UNKNOWN: Mount point not found or not XFS", "red")) print(color_text(" UNKNOWN: Mount point not found or not XFS", "red"))
global_status = max(global_status, 3) global_status = max(global_status, 3)
all_perfdata.append(f"{mount}_status=3") all_perfdata.append(f"{mount.replace('/', '_')}_status=3")
continue continue
usage_pct, size, avail = check_disk_usage(mount) usage_pct, size, avail = check_disk_usage(mount)
if usage_pct is None: if usage_pct is None:
if is_interactive():
print(color_text(" UNKNOWN: Unable to read disk usage", "red")) print(color_text(" UNKNOWN: Unable to read disk usage", "red"))
global_status = max(global_status, 3) global_status = max(global_status, 3)
all_perfdata.append(f"{mount}_status=3") all_perfdata.append(f"{mount.replace('/', '_')}_status=3")
continue continue
# Usage status
usage_status = 0 usage_status = 0
if usage_pct > crit_threshold: if usage_pct > crit_threshold:
usage_status = 2 usage_status = 2
elif usage_pct > warn_threshold: elif usage_pct > warn_threshold:
usage_status = 1 usage_status = 1
if is_interactive():
usage_color = "green" if usage_status == 0 else "yellow" if usage_status == 1 else "red" usage_color = "green" if usage_status == 0 else "yellow" if usage_status == 1 else "red"
print(f" Usage: {color_text(f'{usage_pct}%', usage_color)} (Size: {size}, Available: {avail})") print(f" Usage: {color_text(f'{usage_pct}%', usage_color)} (Size: {size}, Available: {avail})")
# XFS repair check
repair_ok, repair_out = check_xfs_repair(mount) repair_ok, repair_out = check_xfs_repair(mount)
xfs_status = 2 if not repair_ok else 0
if is_interactive():
if not repair_ok: if not repair_ok:
print(color_text(" CRITICAL: XFS filesystem issues detected", "red")) print(color_text(" CRITICAL: XFS filesystem issues detected", "red"))
print(f" {repair_out[:200]}...")
global_status = max(global_status, 2)
xfs_status = 2
else: else:
print(color_text(" XFS filesystem OK (xfs_repair dry-run)", "green")) print(color_text(" XFS filesystem OK (xfs_repair dry-run)", "green"))
xfs_status = 0
# Final status for this mount # Final status for this mount
mount_status = max(usage_status, xfs_status) mount_status = max(usage_status, xfs_status)
global_status = max(global_status, mount_status) global_status = max(global_status, mount_status)
# Nagios perfdata: metric=value;warn;crit;min;max # Nagios perfdata format
perf = f"used_pct={usage_pct};{warn_threshold};{crit_threshold};0;100 size={size} avail={avail} xfs_repair={xfs_status}" mount_name = mount.replace('/', '_').replace(' ', '_')
perf = f"{mount_name}_used_pct={usage_pct};{warn_threshold};{crit_threshold};0;100 size={size} avail={avail} xfs_status={xfs_status}"
all_perfdata.append(perf) all_perfdata.append(perf)
perfdata = " ".join(all_perfdata) perfdata = "| " + " ".join(all_perfdata)
status_text = {0: "OK", 1: "WARNING", 2: "CRITICAL", 3: "UNKNOWN"} status_text = {0: "OK", 1: "WARNING", 2: "CRITICAL", 3: "UNKNOWN"}
# Colorful human output # ALWAYS: Clean Nagios output FIRST LINE (parseable)
print(f"{status_text[global_status]} XFS check: {len(target_mounts)} mount(s){perfdata}")
# ONLY interactive: Additional colorful summary
if is_interactive():
status_color = "green" if global_status == 0 else "yellow" if global_status == 1 else "red" status_color = "green" if global_status == 0 else "yellow" if global_status == 1 else "red"
print("\n" + "=" * 60) print("\n" + "=" * 60)
print(color_text(f"FINAL STATUS: {status_text[global_status]}", status_color)) print(color_text(f"FINAL STATUS: {status_text[global_status]}", status_color))
print(f"| {perfdata}") print(f"{perfdata}")
# NAGIOS: Clean single-line output (no colors, parseable) # Nagios standard exit codes: 0=OK, 1=WARNING, 2=CRITICAL, 3=UNKNOWN
print(f"\n{status_text[global_status]} XFS check complete| {perfdata}")
# Nagios standard exit codes
sys.exit(global_status) sys.exit(global_status)
if __name__ == "__main__": if __name__ == "__main__":