prometheus_client: alle Filesysteme anzeigen (ZFS, LVM, NVMe)

Root-Partition + Extra-Datastores (ZFS-Pools, /var/lib/vz etc.)
werden jetzt korrekt in Übersicht und Detail angezeigt.

Made-with: Cursor
This commit is contained in:
root 2026-03-09 14:04:09 +07:00
parent 2981e98530
commit 586bedc0eb

View file

@ -86,7 +86,7 @@ def get_disk(host: str = None) -> list[dict]:
def get_disk_bytes(host: str = None) -> list[dict]:
"""Returns available and total bytes per host."""
"""Returns available and total bytes for root partition per host."""
filt = f', host="{host}"' if host else ""
avail = _by_host(instant_query(
f'node_filesystem_avail_bytes{{mountpoint="/", host!=""{filt}}}'), "avail_bytes")
@ -104,6 +104,47 @@ def get_disk_bytes(host: str = None) -> list[dict]:
return result
def get_all_filesystems(host: str = None) -> list[dict]:
"""All non-trivial filesystems (skips tmpfs, dev, run, boot)."""
filt = f', host="{host}"' if host else ""
skip = "tmpfs|devtmpfs|efivarfs"
q_avail = f'node_filesystem_avail_bytes{{fstype!~"{skip}", host!=""{filt}}}'
q_total = f'node_filesystem_size_bytes{{fstype!~"{skip}", host!=""{filt}}}'
data_a = instant_query(q_avail)
data_t = instant_query(q_total)
if data_a.get("status") != "success":
return []
total_map = {}
for r in data_t.get("data", {}).get("result", []):
m = r.get("metric", {})
key = (m.get("host", m.get("instance", "?")), m.get("mountpoint", ""))
total_map[key] = float(r.get("value", [0, 0])[1])
results = []
for r in data_a.get("data", {}).get("result", []):
m = r.get("metric", {})
h = m.get("host", m.get("instance", "?"))
mp = m.get("mountpoint", "")
if mp in ("/boot/efi", "/boot"):
continue
avail = float(r.get("value", [0, 0])[1])
total = total_map.get((h, mp), 0)
if total < 500 * 1024 * 1024:
continue
used_pct = ((total - avail) / total * 100) if total > 0 else 0
results.append({
"host": h,
"mountpoint": mp,
"total_gb": total / (1024**3),
"avail_gb": avail / (1024**3),
"used_pct": used_pct,
"device": m.get("device", ""),
})
return sorted(results, key=lambda x: (x["host"], x["mountpoint"]))
def get_uptime(host: str = None) -> list[dict]:
filt = f', host="{host}"' if host else ""
q = f'node_time_seconds{{host!=""{filt}}} - node_boot_time_seconds{{host!=""{filt}}}'
@ -150,6 +191,11 @@ def format_overview() -> str:
disk_gb = {r["host"]: r for r in get_disk_bytes()}
uptime = {r["host"]: r["value"] for r in get_uptime()}
load = {r["host"]: r["value"] for r in get_load()}
all_fs = get_all_filesystems()
extra_fs = {}
for fs in all_fs:
if fs["mountpoint"] != "/":
extra_fs.setdefault(fs["host"], []).append(fs)
hosts = sorted(set(list(cpu.keys()) + list(mem.keys()) + list(disk.keys())))
if not hosts:
@ -177,10 +223,17 @@ def format_overview() -> str:
mem_str = f"{m:.0f}%" if m >= 0 else "n/a"
load_str = f"{l5:.1f}" if l5 >= 0 else "n/a"
extra_line = ""
if h in extra_fs:
parts = []
for efs in extra_fs[h]:
parts.append(f"{efs['mountpoint']}: {efs['avail_gb']:.0f}/{efs['total_gb']:.0f} GB frei")
extra_line = "\n Storage: " + ", ".join(parts)
lines.append(
f"{emoji} {h}{warn}\n"
f" CPU: {cpu_str} RAM: {mem_str} Disk: {disk_str}\n"
f" Load5: {load_str} Uptime: {_fmt_uptime(u)}"
f" Load5: {load_str} Uptime: {_fmt_uptime(u)}{extra_line}"
)
warnings = get_warnings()
@ -200,8 +253,7 @@ def format_host_detail(host: str) -> str:
cpu = get_cpu(host)
mem = get_memory(host)
disk = get_disk(host)
disk_gb = get_disk_bytes(host)
filesystems = get_all_filesystems(host)
uptime = get_uptime(host)
load = get_load(host)
@ -213,11 +265,14 @@ def format_host_detail(host: str) -> str:
lines.append(f"CPU: {cpu[0]['value']:.1f}%")
if mem:
lines.append(f"RAM: {mem[0]['value']:.1f}%")
if disk:
d_str = f"Disk: {disk[0]['value']:.1f}%"
if disk_gb:
d_str += f" ({disk_gb[0]['avail_gb']:.1f} / {disk_gb[0]['total_gb']:.1f} GB frei)"
lines.append(d_str)
if filesystems:
lines.append("Speicher:")
for fs in filesystems:
warn = " ⚠️" if fs["used_pct"] >= WARN_DISK else ""
lines.append(
f" {fs['mountpoint']}: {fs['used_pct']:.0f}% belegt "
f"({fs['avail_gb']:.0f} / {fs['total_gb']:.0f} GB frei){warn}"
)
if load:
lines.append(f"Load (5m): {load[0]['value']:.2f}")
if uptime: