272 lines
12 KiB
Markdown
272 lines
12 KiB
Markdown
# Вариант 1 — Пересоздание XFS на том же логическом томе (`vg0/data`)
|
||
|
||
Этот вариант подходит, если ты готов удалить все виртуальные машины и контейнеры (VM/LXC)
|
||
и настроить их заново. Он полностью очищает файловую систему `/var/lib/vz`
|
||
и создаёт новую XFS с оптимальными параметрами под RAID0 (6×NVMe, chunk 128 KB).
|
||
|
||
---
|
||
|
||
## 🔹 Шаги
|
||
|
||
### 1. Останови все контейнеры и ВМ
|
||
```bash
|
||
qm list
|
||
pct list
|
||
systemctl stop pvedaemon pve-cluster pve-storage pveproxy
|
||
```
|
||
|
||
---
|
||
|
||
### 2. Размонтируй `/var/lib/vz`
|
||
```bash
|
||
umount /var/lib/vz
|
||
```
|
||
|
||
Если появится ошибка `busy`, проверь, какие процессы удерживают каталог:
|
||
```bash
|
||
lsof +D /var/lib/vz
|
||
```
|
||
и заверши эти процессы.
|
||
|
||
---
|
||
|
||
### 3. Пересоздай файловую систему XFS
|
||
Оптимальные параметры для RAID0 с chunk = 1 Mb и 6 дисков для блокчейнов:
|
||
```bash
|
||
mkfs.xfs -f -s size=4096 -l size=1g -n size=4096 -d su=1m,sw=3 /dev/md3
|
||
```
|
||
Проверка, что XFS «увидел» выравнивание:
|
||
```bash
|
||
xfs_info /dev/md3 | egrep 'sunit|swidth|sectsz'
|
||
# Ожидаем: sunit=256 (256 * 4K = 1MiB), swidth=768 (768 * 4K = 3MiB), sectsz=4096
|
||
```
|
||
|
||
```bash
|
||
mkdir -p /var/lib/vz
|
||
UUID=$(blkid -s UUID -o value /dev/md3)
|
||
echo "UUID=$UUID /var/lib/vz xfs defaults,noatime,inode64,logbufs=8,logbsize=256k 0 2" >> /etc/fstab
|
||
mount -a
|
||
```
|
||
|
||
|
||
### 6. Перезапусти службы Proxmox
|
||
```bash
|
||
systemctl start pve-cluster pvedaemon pve-storage pveproxy
|
||
```
|
||
---
|
||
|
||
## ✅ Результат
|
||
|
||
После выполнения этих шагов:
|
||
|
||
- `/var/lib/vz` будет чистой, новой XFS-файловой системой,
|
||
оптимизированной под RAID0 (3×NVMe, chunk 1 МБ).
|
||
- Proxmox автоматически подхватит хранилище после старта сервисов.
|
||
- Система готова к развертыванию новых VM и LXC-контейнеров.
|
||
---
|
||
|
||
### 📋 Проверка
|
||
|
||
После монтирования убедись в правильности параметров:
|
||
```bash
|
||
xfs_info /var/lib/vz | grep -E "sunit|swidth"
|
||
df -h /var/lib/vz
|
||
```
|
||
Ожидаемые значения:
|
||
```
|
||
= sunit=256 swidth=1536 blks
|
||
= sectsz=4096 sunit=1 blks, lazy-count=1
|
||
Filesystem Size Used Avail Use% Mounted on
|
||
/dev/md3 11T 205G 11T 2% /var/lib/vz
|
||
```
|
||
|
||
-----
|
||
Да — ✅ эти настройки стоит применить, особенно если у тебя RAID 0 из нескольких NVMe и ты используешь ноду (например, Reth / Nethermind / Erigon) с большими sequential-чтениями и trace/debug.
|
||
Разберём по пунктам, зачем каждая настройка:
|
||
|
||
🧠 1. read_ahead_kb
|
||
|
||
Позволяет ядру читать данные «с запасом» при последовательном чтении.
|
||
|
||
Устройство Значение Эффект
|
||
/sys/block/md1/queue/read_ahead_kb 4096 RAID-массив (XFS): читать по 4 МБ заранее
|
||
/sys/block/nvme*/queue/read_ahead_kb
|
||
|
||
👉 На Ethereum-архивах (особенно при trace и debug) это даёт +10-20 % скорости при холодных чтениях и снижает IOPS-нагрузку.
|
||
|
||
⚙️ 2. irqbalance
|
||
|
||
Этот демон равномерно распределяет аппаратные прерывания NVMe-дисков между CPU-ядрами.
|
||
Без него все IRQ могут обрабатываться одним ядром, и даже самый быстрый RAID будет «тормозить» при многопоточном доступе.
|
||
|
||
Установка и включение:
|
||
```bash
|
||
apt update
|
||
apt install irqbalance -y
|
||
systemctl enable --now irqbalance
|
||
systemctl status irqbalance
|
||
```
|
||
|
||
```bash
|
||
echo 4096 | sudo tee /sys/block/md3/queue/read_ahead_kb
|
||
for d in /sys/block/nvme*n1/queue/read_ahead_kb; do echo 2048 | sudo tee "$d"; done
|
||
```
|
||
|
||
💾 3. Применение постоянных значений
|
||
|
||
Чтобы после перезагрузки не сбрасывались, добавь в /etc/rc.local или systemd-юнит, например:
|
||
```bash
|
||
cat >/usr/local/sbin/raid-tuning.sh <<'EOF'
|
||
#!/usr/bin/env bash
|
||
set -euo pipefail
|
||
|
||
# RAID массив(ы): 4 MiB readahead
|
||
[ -e /sys/block/md3/queue/read_ahead_kb ] && echo 4096 > /sys/block/md3/queue/read_ahead_kb
|
||
|
||
# NVMe диски: 2 MiB readahead + scheduler=none
|
||
for d in /sys/block/nvme*n1; do
|
||
[ -e "$d/queue/read_ahead_kb" ] && echo 2048 > "$d/queue/read_ahead_kb"
|
||
[ -e "$d/queue/scheduler" ] && echo none > "$d/queue/scheduler"
|
||
done
|
||
EOF
|
||
|
||
chmod +x /usr/local/sbin/raid-tuning.sh
|
||
|
||
```
|
||
|
||
|
||
```bash
|
||
cat >/etc/systemd/system/raid-tuning.service <<'EOF'
|
||
[Unit]
|
||
Description=RAID/NVMe tuning
|
||
After=mdadm-raid.service local-fs.target
|
||
Wants=mdadm-raid.service
|
||
|
||
[Service]
|
||
Type=oneshot
|
||
ExecStart=/usr/local/sbin/raid-tuning.sh
|
||
RemainAfterExit=yes
|
||
|
||
[Install]
|
||
WantedBy=multi-user.target
|
||
EOF
|
||
|
||
```
|
||
|
||
```bash
|
||
systemctl daemon-reload
|
||
systemctl enable --now raid-tuning.service
|
||
systemctl enable --now fstrim.timer
|
||
systemctl status raid-tuning.service --no-pager
|
||
```
|
||
```bash
|
||
for f in /sys/block/md*/queue/read_ahead_kb; do printf "%s: %s\n" "$f" "$(cat "$f")"; done
|
||
for f in /sys/block/nvme*n1/queue/read_ahead_kb; do printf "%s: %s\n" "$f" "$(cat "$f")"; done
|
||
```
|
||
|
||
|
||
🧠 Что делает irqbalance
|
||
irqbalance автоматически распределяет аппаратные прерывания (IRQ) от NVMe, сетевых и других устройств по разным CPU-ядрам.
|
||
Для серверов с:
|
||
несколькими NVMe,
|
||
RAID0,
|
||
большим количеством потоков (64 ядра, как у тебя на EPYC 7502P),
|
||
это снижает latency и делает нагрузку по NVMe-дискам ровнее.
|
||
🔍 Проверка распределения IRQ
|
||
После запуска можно убедиться, что прерывания действительно разнесены по ядрам:
|
||
```bash
|
||
grep -i nvme /proc/interrupts
|
||
```
|
||
```bash
|
||
161: 1000000 0 0 0 IR-PCI-MSI nvme0q1
|
||
162: 0 900000 0 0 IR-PCI-MSI nvme1q1
|
||
163: 0 0 800000 0 IR-PCI-MSI nvme2q1
|
||
```
|
||
|
||
Быстрый sanity-тест диска:
|
||
```bash
|
||
fio --name=seqread --filename=/var/lib/vz/fio.test --size=20G --rw=read --bs=1M --iodepth=32 --direct=1 --runtime=30 --time_based --group_reporting
|
||
fio --name=seqwrite --filename=/var/lib/vz/fio.test --size=20G --rw=write --bs=1M --iodepth=32 --direct=1 --runtime=30 --time_based --group_reporting
|
||
rm -f /var/lib/vz/fio.test
|
||
```
|
||
RAID0x6
|
||
```bash
|
||
seqread: (g=0): rw=read, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=psync, iodepth=32
|
||
fio-3.39
|
||
Starting 1 process
|
||
seqread: Laying out IO file (1 file / 20480MiB)
|
||
note: both iodepth >= 1 and synchronous I/O engine are selected, queue depth will be capped at 1
|
||
Jobs: 1 (f=1): [R(1)][100.0%][r=2943MiB/s][r=2942 IOPS][eta 00m:00s]
|
||
seqread: (groupid=0, jobs=1): err= 0: pid=211291: Thu Oct 23 08:39:48 2025
|
||
read: IOPS=2923, BW=2924MiB/s (3066MB/s)(85.7GiB/30001msec)
|
||
clat (usec): min=182, max=2295, avg=341.71, stdev=229.11
|
||
lat (usec): min=183, max=2295, avg=341.74, stdev=229.11
|
||
clat percentiles (usec):
|
||
| 1.00th=[ 186], 5.00th=[ 186], 10.00th=[ 186], 20.00th=[ 188],
|
||
| 30.00th=[ 192], 40.00th=[ 192], 50.00th=[ 192], 60.00th=[ 194],
|
||
| 70.00th=[ 619], 80.00th=[ 619], 90.00th=[ 619], 95.00th=[ 619],
|
||
| 99.00th=[ 635], 99.50th=[ 1057], 99.90th=[ 2180], 99.95th=[ 2278],
|
||
| 99.99th=[ 2278]
|
||
bw ( MiB/s): min= 2864, max= 2984, per=100.00%, avg=2925.56, stdev=28.52, samples=59
|
||
iops : min= 2864, max= 2984, avg=2925.56, stdev=28.52, samples=59
|
||
lat (usec) : 250=66.13%, 500=0.04%, 750=33.32%, 1000=0.01%
|
||
lat (msec) : 2=0.31%, 4=0.19%
|
||
cpu : usr=0.18%, sys=20.96%, ctx=87762, majf=0, minf=264
|
||
IO depths : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
|
||
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
|
||
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
|
||
issued rwts: total=87717,0,0,0 short=0,0,0,0 dropped=0,0,0,0
|
||
latency : target=0, window=0, percentile=100.00%, depth=32
|
||
|
||
Run status group 0 (all jobs):
|
||
READ: bw=2924MiB/s (3066MB/s), 2924MiB/s-2924MiB/s (3066MB/s-3066MB/s), io=85.7GiB (92.0GB), run=30001-30001msec
|
||
|
||
Disk stats (read/write):
|
||
dm-2: ios=87372/3, sectors=178921472/40, merge=0/0, ticks=27671/0, in_queue=27671, util=92.36%, aggrios=701744/661, aggsectors=179644416/101672, aggrmerge=0/0, aggrticks=76026/107, aggrin_queue=76133, aggrutil=92.29%
|
||
md1: ios=701744/661, sectors=179644416/101672, merge=0/0, ticks=76026/107, in_queue=76133, util=92.29%, aggrios=116956/89, aggsectors=29940736/16945, aggrmerge=1/20, aggrticks=20080/38, aggrin_queue=20118, aggrutil=58.99%
|
||
nvme1n1: ios=116964/81, sectors=29942784/16624, merge=4/10, ticks=13644/33, in_queue=13676, util=17.87%
|
||
nvme5n1: ios=116944/125, sectors=29937664/17208, merge=0/39, ticks=51833/67, in_queue=51900, util=58.99%
|
||
nvme2n1: ios=116944/87, sectors=29937664/16960, merge=0/48, ticks=16230/34, in_queue=16263, util=7.48%
|
||
nvme3n1: ios=116960/75, sectors=29941760/17488, merge=4/1, ticks=14628/31, in_queue=14660, util=12.67%
|
||
nvme0n1: ios=116964/84, sectors=29942784/16792, merge=0/22, ticks=10502/34, in_queue=10535, util=7.43%
|
||
nvme4n1: ios=116960/85, sectors=29941760/16600, merge=0/4, ticks=13645/34, in_queue=13678, util=14.23%
|
||
seqwrite: (g=0): rw=write, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=psync, iodepth=32
|
||
fio-3.39
|
||
Starting 1 process
|
||
note: both iodepth >= 1 and synchronous I/O engine are selected, queue depth will be capped at 1
|
||
Jobs: 1 (f=1): [W(1)][100.0%][w=2927MiB/s][w=2927 IOPS][eta 00m:00s]
|
||
seqwrite: (groupid=0, jobs=1): err= 0: pid=211615: Thu Oct 23 08:40:18 2025
|
||
write: IOPS=2940, BW=2940MiB/s (3083MB/s)(86.1GiB/30001msec); 0 zone resets
|
||
clat (usec): min=166, max=2283, avg=336.58, stdev=234.58
|
||
lat (usec): min=169, max=2286, avg=339.79, stdev=234.58
|
||
clat percentiles (usec):
|
||
| 1.00th=[ 169], 5.00th=[ 169], 10.00th=[ 172], 20.00th=[ 174],
|
||
| 30.00th=[ 178], 40.00th=[ 188], 50.00th=[ 188], 60.00th=[ 190],
|
||
| 70.00th=[ 619], 80.00th=[ 619], 90.00th=[ 627], 95.00th=[ 627],
|
||
| 99.00th=[ 644], 99.50th=[ 660], 99.90th=[ 2114], 99.95th=[ 2245],
|
||
| 99.99th=[ 2278]
|
||
bw ( MiB/s): min= 2858, max= 2992, per=100.00%, avg=2943.25, stdev=27.86, samples=59
|
||
iops : min= 2858, max= 2992, avg=2943.25, stdev=27.86, samples=59
|
||
lat (usec) : 250=66.17%, 500=0.01%, 750=33.33%, 1000=0.01%
|
||
lat (msec) : 2=0.32%, 4=0.16%
|
||
cpu : usr=2.57%, sys=13.38%, ctx=88256, majf=0, minf=6
|
||
IO depths : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
|
||
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
|
||
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
|
||
issued rwts: total=0,88208,0,0 short=0,0,0,0 dropped=0,0,0,0
|
||
latency : target=0, window=0, percentile=100.00%, depth=32
|
||
|
||
Run status group 0 (all jobs):
|
||
WRITE: bw=2940MiB/s (3083MB/s), 2940MiB/s-2940MiB/s (3083MB/s-3083MB/s), io=86.1GiB (92.5GB), run=30001-30001msec
|
||
|
||
Disk stats (read/write):
|
||
dm-2: ios=0/87854, sectors=0/179902504, merge=0/0, ticks=0/19604, in_queue=19604, util=65.42%, aggrios=0/705804, aggsectors=0/180651424, aggrmerge=0/0, aggrticks=0/57168, aggrin_queue=57168, aggrutil=64.74%
|
||
md1: ios=0/705804, sectors=0/180651424, merge=0/0, ticks=0/57168, in_queue=57168, util=64.74%, aggrios=0/117619, aggsectors=0/30108570, aggrmerge=0/14, aggrticks=0/18083, aggrin_queue=18083, aggrutil=43.24%
|
||
nvme1n1: ios=0/117617, sectors=0/30109704, merge=0/4, ticks=0/12824, in_queue=12824, util=7.39%
|
||
nvme5n1: ios=0/117610, sectors=0/30105712, merge=0/1, ticks=0/44537, in_queue=44537, util=43.24%
|
||
nvme2n1: ios=0/117610, sectors=0/30106168, merge=0/61, ticks=0/14725, in_queue=14725, util=7.29%
|
||
nvme3n1: ios=0/117625, sectors=0/30109768, merge=0/4, ticks=0/13479, in_queue=13478, util=7.64%
|
||
nvme0n1: ios=0/117626, sectors=0/30110240, merge=0/17, ticks=0/10437, in_queue=10438, util=6.08%
|
||
nvme4n1: ios=0/117627, sectors=0/30109832, merge=0/2, ticks=0/12499, in_queue=12500, util=9.68%
|
||
```
|