From 015a8fa5d4c9e1a144b71e12a932227d545fbb17 Mon Sep 17 00:00:00 2001
From: Benoit S <git@benpro.fr>
Date: Sat, 6 Feb 2021 17:03:07 +0900
Subject: [PATCH] Add zfs-scrub

---
 files/zfs-scrub.service   | 11 +++++++++
 files/zfs-scrub.timer     | 10 +++++++++
 group_data/all.py         |  4 ++++
 setup-zfs-and-lxd.py      | 47 ++++++++++++++++++++++++++++++++++++++-
 templates/zfs-scrub.sh.j2 | 18 +++++++++++++++
 5 files changed, 89 insertions(+), 1 deletion(-)
 create mode 100644 files/zfs-scrub.service
 create mode 100644 files/zfs-scrub.timer
 create mode 100644 templates/zfs-scrub.sh.j2

diff --git a/files/zfs-scrub.service b/files/zfs-scrub.service
new file mode 100644
index 0000000..858574d
--- /dev/null
+++ b/files/zfs-scrub.service
@@ -0,0 +1,11 @@
+[Unit]
+Description=zpool scrub
+
+[Service]
+Nice=19
+IOSchedulingClass=idle
+KillSignal=SIGINT
+ExecStart=/usr/local/bin/zfs-scrub.sh
+
+[Install]
+WantedBy=multi-user.target
diff --git a/files/zfs-scrub.timer b/files/zfs-scrub.timer
new file mode 100644
index 0000000..a21a252
--- /dev/null
+++ b/files/zfs-scrub.timer
@@ -0,0 +1,10 @@
+[Unit]
+Description=Monthly zpool scrub
+
+[Timer]
+OnCalendar=monthly
+RandomizedDelaySec=86400
+Persistent=true
+
+[Install]
+WantedBy=multi-user.target
diff --git a/group_data/all.py b/group_data/all.py
index 656a154..14f96a0 100644
--- a/group_data/all.py
+++ b/group_data/all.py
@@ -11,3 +11,7 @@ borg_passphrase = b_borg_passphrase.decode('utf-8')
 
 b_borg_healthchecks = get_secret(b'1$2$AbHQ2WFelnJfIi5SjEtgAugd3xHSJDfikh0RBV3QY80=$Z0FBQUFBQmdGOWdKZnp1d3B0YUJUck9kOVNBVWxTRjJZZnpubUZmNmV0cmlCUm9kLUhYNndqWkhMOG5WQU44ZUVMQ2lEdW1jLUhSOTJub0VMZWFWdHBBb0JKdVptanNCQ2xyQXQ2cEdfMjBaUy00Y2dlcnJzc1ViaE5sQ0RUUS1idkViQ0hOOWdRWUlqZElQSlAzYmpjSUl5bjNuTDhDLTNBPT0=')
 borg_healthchecks = b_borg_healthchecks.decode('utf-8')
+
+b_zfs_healthchecks = get_secret(b'1$2$dsh3T6POjU9RvOqr_Itg1paAckP4K2cIRN8L22OsGVo=$Z0FBQUFBQmdIa21UYUtGeVdmLWNvbGhYUDNfNzdTd0gwN21ISHRTV3B6SDBrTFA4aTRCaF84bWY5UlhYV3RSOUxsekRyTVd4REhqczZTTW00WHhzSjJSV2JuNjZ5NXdGWmtrQjFkWjNNdmZYTm4xMXUxOHlYNlpJb0pGQmoyb0JTajRjanQwMHlEUzhUSTNjRFlXenBwdDRTMzA5c2NzeFJBPT0=')
+zfs_healthchecks = b_zfs_healthchecks.decode('utf-8')
+
diff --git a/setup-zfs-and-lxd.py b/setup-zfs-and-lxd.py
index 057083b..86e14cd 100644
--- a/setup-zfs-and-lxd.py
+++ b/setup-zfs-and-lxd.py
@@ -1,5 +1,5 @@
 from pyinfra import host
-from pyinfra.operations import server, files, systemd
+from pyinfra.operations import server, files, systemd, apt
 
 SUDO = True
 
@@ -19,6 +19,12 @@ SUDO = True
 # sdc       8:32   0 223.6G  0 disk  
 # └─sdc1    8:33   0    24G  0 part
 
+apt.packages(
+    name='Install packages',
+    packages=['zfsutils-linux'],
+    update=True,
+)
+
 if not host.fact.command('lsblk | grep sda4 || true'):
     server.shell(
         name='Create sda4 for zpool',
@@ -92,3 +98,42 @@ if not host.fact.command('lxc storage volume list default | grep images || true'
         commands=['lxc storage volume create default images', 'lxc config set storage.images_volume default/images']
     )
 
+files.template(
+    name='Push zfs-scrub script',
+    src='templates/zfs-scrub.sh.j2',
+    dest='/usr/local/bin/zfs-scrub.sh',
+    mode='700',
+    user='root',
+    group='root',
+)
+
+files.put(
+    name='Push zfs-scrub timer',
+    src='files/zfs-scrub.timer',
+    dest='/etc/systemd/system/zfs-scrub.timer',
+    user='root',
+    group='root',
+    mode='644',
+)
+
+files.put(
+    name='Push zfs-scrub service',
+    src='files/zfs-scrub.service',
+    dest='/etc/systemd/system/zfs-scrub.service',
+    user='root',
+    group='root',
+    mode='644',
+)
+
+systemd.daemon_reload(
+    name='Reload systemd',
+    user_mode=False,
+)
+
+systemd.service(
+    name='Enable zfs-scrub timer',
+    service='zfs-scrub.timer',
+    running=True,
+    enabled=True,
+)
+
diff --git a/templates/zfs-scrub.sh.j2 b/templates/zfs-scrub.sh.j2
new file mode 100644
index 0000000..b91fb65
--- /dev/null
+++ b/templates/zfs-scrub.sh.j2
@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+# https://serverfault.com/questions/538978/how-to-run-a-command-once-a-zfs-scrub-completes
+set -euo pipefail
+
+curl -m 10 --retry 5 {{ host.data.zfs_healthchecks }}/start
+zpool scrub local
+# wait until scrub is finished
+while zpool status local | grep -q 'scan:  *scrub in progress'; do
+  zpool status local
+  sleep 600
+done
+
+zpool status local
+
+# Get stdout from journalctl
+LOG=$(journalctl -o cat -r -u zfs-scrub.service -n 100)
+curl -fsS -m 10 --retry 5 --data-raw "$LOG" {{ host.data.zfs_healthchecks }}
+exit 0