🚜 The Mow'n'Plow MK2

Overview

An efficient Chia Plot mower and mover. It's purpose is minimise the amount of time that drives are unavailable for harvesting while replotting (or even normal plotting).

Uses inotify to watch for new Chia plots and then fires off rsync to move them to their final destination.

Based on '🚜 The Plow' (https://github.com/lmacken/plow) it also:

MK1

Prioritises moving of plots to fill the disks in order of definition
(Optionally) removes plots older than a given timestamp
Collects all of the available mounted plot drives so you don't have specify them manually (though you can if you want to)

MK2

This is a fairly major restructure to the codebase which adds:

a separate configuration file
the option to remove all matching plots on the destination drive at the beginning of plowing
the ability to remove a destination drive from harvesting until it is 'full'

Rather than spilling too much ink here (who reads doco anyway?) there are comments in the config.yaml which should point you in the right direction.

Usage Notes

This tool is an 'opinionated' method for mowing and plowing and only works with a single harvester at a time. It has been developed and tested on Linux so YMMV might vary on Windows.

Directory Structure

This script reflects the way that I organise the plots on my harvester. Each plot drive is mounted under \data\chia\plots like so:

├── c0b0
│   ├── plot-k32-2023-01-10-05-52-f0037e35e748d775aa5d7743c1280343e9c4779bd6eca195e2281bff9c13111c.plot
│   ├── plot-k32-2023-01-10-06-05-11dd9e96363612db2a643fe32592f9003f1908cb3c9d91fa6b327c9d4e2773a0.plot
│   ...
├── c0b1
│   ├── plot-k32-2023-01-10-05-52-f0037e35e748d775aa5d7743c1280343e9c4779bd6eca195e2281bff9c13111c.plot
│   ├── plot-k32-2023-01-10-06-05-11dd9e96363612db2a643fe32592f9003f1908cb3c9d91fa6b327c9d4e2773a0.plot
│   ...
├── c1b0
│   ├── plot-k32-2023-01-10-05-52-f0037e35e748d775aa5d7743c1280343e9c4779bd6eca195e2281bff9c13111c.plot
│   ├── plot-k32-2023-01-10-06-05-11dd9e96363612db2a643fe32592f9003f1908cb3c9d91fa6b327c9d4e2773a0.plot
│   ...
│...

Rsync

rsync is run as a service on the harvester. This by-passes the encryption of the file during transfer (there is very little value to be obtained by sniffing the traffic and I mostly trust the servers on my network).

Configuration files

# /etc/rsyncd.conf

# Configuration file for rsync daemon mode
# See rsyncd.conf man page for more options.

# configuration example:
uid = chia
gid = chia
port = 12000

# plots is the rsync 'module' which becomes the site root
[plots]
path = /data/chia/plots
comment = Chia plots
read only = false

# /etc/systemd/system/rsync.service

[Unit]
Description=fast remote file copy program daemon
ConditionPathExists=/etc/rsyncd.conf
After=network.target
Documentation=man:rsync(1) man:rsyncd.conf(5)

[Service]
ExecStart=/usr/bin/nocache /usr/bin/ionice -c 3 /usr/bin/rsync --daemon --no-detach
RestartSec=1

# Citing README.md:
#
#   [...] Using ssh is recommended for its security features.
#
#   Alternatively, rsync can run in `daemon' mode, listening on a socket.
#   This is generally used for public file distribution, [...]
#
# So let's assume some extra security is more than welcome here. We do full
# system protection (which makes /usr, /boot, & /etc read-only) and hide
# devices. To override these defaults, it's best to do so in the drop-in
# directory, often done via `systemctl edit rsync.service`. The file needs
# just the bare minimum of the right [heading] and override values.
# See systemd.unit(5) and search for "drop-in" for full details.

ProtectSystem=full
#ProtectHome=on|off|read-only
PrivateDevices=on
NoNewPrivileges=on

[Install]
WantedBy=multi-user.target

Note: File permissions can be a source of frustration. Make sure that your plot directories are owned by the user specified in the rsyncd.conf file

Disk Scheduler

Plotting

Writing plot files while harvesting can have a large impact on the latency on harvester lookups (which can increase the number of stale plots being submitted to the pool). ionice (specified in the rsync ExecStart command above) only works if the disk scheduler in use is either CFQ (obsolete) or BFQ. Unfortunately, Debian Bullseye and Ubuntu 22.04 default to using the mq-deadline scheduler so ionice is ineffective.

To check the current schedules in use run:

grep "" /sys/block/*/queue/scheduler

If BFQ is not in use for your hard disks then to get ionice to work the following configuration files will need to be added/changed:

# /etc/modules-load.d/bfq.conf
# Make sure the bfq module is loaded at boot

bfq

# /etc/udev/rules.d/60-ioschedulers.rules 

# set scheduler for rotating disks
ACTION=="add|change", KERNEL=="sd[a-z]*", ATTR{queue/rotational}=="1", ATTR{queue/scheduler}="bfq"

After changing these files a reboot will be required.

Normal Plotting

The mq-scheduler has lower latency during normal operation of the harvester so after plotting is complete reverse the changes above and reboot again.

Nocache

The rsync service also uses the nocache utility which ensures that rsync writes bypass the filesystem cache. To install:

apt-get install nocache

SSH

The script assumes that there is an entry in the authorized_keys file for the DEST_USER specified. If necessary, see Understanding SSH authorized_keys file with Examples for instructions.

Footnote

I built this tool for my own use (and the lol's :innocent:) but if you find it useful and feel the urge to buy me a drink use:

xch1vlnelz9ef43z3xa4x6a3zzfm7cezwvmq332p97xlflmxxcgzdrpsqamyee

graemes / mownplow