Experimental Features


The features described here are unfinished and in an alpha or beta stage.

Query Optimization

You can provide the --fast-query option of rtcontrol to set a level of optimization to use when querying rTorrent for items. The default for that option is set via the fast_query config parameter, and is 0 if not changed. That means optimization is normally off, and can be activated via -Q1. It is recommended to keep it that way for now, and use -Q1 explicitly in scripts and other background processing to reduce the load they generate. Only activating it in scripts usually means the filters used don’t change that much, i.e. you can be pretty sure the optimization does what you expect it to do.

Level 1 is less aggressive and safe by definition (i.e. produces correct results in all cases, unless there’s a bug), while -Q2 is highly experimental and in some circumstances likely produces results that are too small or empty.

Optimization works by giving a pre-filter condition to rTorrent, to reduce the overhead involved in sending items over XMLRPC and processing them, only to be then discarded in the rtcontrol filter machinery. That pre-filter evaluation needs features of rTorrent-PS 1.1 or later, and will produce errors when used with anything else.

This goal of reducing the number of items sent to rtcontrol is best achieved if you put a highly selective condition first in a series of conditions combined by AND. For cron-type jobs, this can often be achieved by looking at recent items only – older items should already be processed by previous runs. Even a very lenient window like “last week” drastically reduces items that need to be processed.

Consider this example:

$ rtcontrol loaded=-6w is_ignored=0 -o- -v -Q0
DEBUG    Matcher is: loaded=-6w is_ignored=no
DEBUG    Got 131 items with 20 attributes …
INFO     Filtered 13 out of 131 torrents.
DEBUG    XMLRPC stats: 25 req, out 5.6 KiB [1.4 KiB max], in 104.9 KiB [101.5 KiB max], …
INFO     Total time: 0.056 seconds.

$ rtcontrol loaded=-6w is_ignored=0 -o- -v -Q1
INFO     !!! pre-filter: greater=value=$d.custom=tm_loaded,value=1488920876
DEBUG    Got 17 items with 20 attributes …
INFO     Filtered 13 out of 131 torrents.
DEBUG    XMLRPC stats: 25 req, out 5.7 KiB [1.5 KiB max], in 16.6 KiB [13.2 KiB max], …
INFO     Total time: 0.028 seconds.

You can see that the 2nd command executes faster (the effect is larger with more overall items), and only looks at 17 items to select the final 13 ones, while with -Q0 all 131 items need to be looked at, and thus transferred via XMLRPC. That means 105 KiB instead of only 16.6 KiB need to be serialized, read, and parsed again.

Putting the right condition first is quite important, as you can see when the conditions are swapped and the less selective one is used for the pre-filter:

$ rtcontrol is_ignored=0 loaded=-6w -o- -v -Q1
INFO     !!! pre-filter: equal=d.ignore_commands=,value=0
DEBUG    Got 117 items with 20 attributes …

Be careful when mixing --anneal and --fast-query, since most of the post-processing steps also look at deselected items, and produce unexpected results if they are missing due to pre-filtering. Put another way, always include -Q0 when you use --anneal, to be on the safe side.

Connecting via SSH

Starting with version 0.4.1, you can use URLs of the form


to connect securely to a remote rTorrent instance. For this to work, the following preconditions have to be met:

  • the provided account has to have full permissions (rwx) on the given socket.
  • you have to use either public key authentication via authorized_keys, or a SSH agent that holds your password.
  • the remote host needs to have the socat executable available (on Debian/Ubuntu, install the socat package).

You also need to extend the rtorrent.rc of the remote instance with this snippet:

# COMMAND: Return startup time (can be used to calculate uptime)
method.insert = startup_time,value|const,$system.time=

For example, the following queries the remote instance ID using rtxmlrpc:

rtxmlrpc -v -Dscgi_url=scgi+ssh://user@example.com/var/torrent/.scgi_local session.name

This typically takes several seconds due to the necessary authentication.

Using the Monitoring Web Service



This feature is not finished and should not be considered stable at this time (i.e. it might change drastically).

The monitoring subsystem is an optional part of pyrotorque and includes a web service that creates the monitoring pages which can be viewed in your browser. There is a live view that continuously updates current performance indicators of rTorrent and the host it runs on, something similar to this:


Screenshot of the Monitoring View

What can you see here?

  • rTorrent and host uptimes.
  • rTorrent upload and download activity.
  • number of rTorrent items in total (♯), active (⚡), having a message (↯), complete (✔), incomplete (◑), seeding (▲), downloading (▼). started (☀), stopped (■).
  • and key host performance indicators.

The web interface follows responsive web design (RWD) principles, which means it’ll adapt to different devices and their display size.

(This is not yet implemented…) Also, the StatsArchiver job of the pyrotorque demon writes a lot of statistical data to RRD archives (round robin database) in 1 minute intervals. See http://oss.oetiker.ch/rrdtool/doc/rrdtool.en.html for the theory behind RRD, and the standard implementation used in a lot of systems.

Installation & Configuration

As previously mentioned, monitoring is an optional part of pyrotorque, so first see rTorrent Queue Manager on how to set it up in case you didn’t do that already. After pyrotorque is successfully running, follow these additional steps to activate the web server.

A few additional Python libraries and external CSS/Javascript resources need to be installed, which are not part of the core distribution.

  1. Install current code and dependencies:

    ~/.local/pyroscope/bin/pip install -r ~/.local/pyroscope/requirements-torque.txt
  2. Activate the web server option by adding this to your ~/.pyroscope/torque.ini:

    httpd.active = True
  3. Download resources to ~/.pyroscope/htdocs:

    pyroadmin --create-config
  4. Finally, restart the demon:

    pyrotorque --cron --restart
    # use "pyrotorque --fg --restart -v" instead, in case something doesn't work,
    # so you can directly read the log

If you didn’t change the defaults, the web interface is now available using the URL http://localhost:8042/, which will show you something similar to the screen shot further above,

Additional Configuration Options

As with other config files, ~/.pyroscope/torque.ini.default lists all the available options and a short description. The following just lists those that are quite often changed from the defaults.

The address the web server listens on. The default is (i.e. localhost), and can be changed to to listen to any interface. Note that the latter is only safe in your home LAN, behind a firewall or NAT. Add a reverse proxy to your Apache/nginx/… for exposing the web service to the internet, ideally adding password protection and using SSL.
TCP port the web server listens on, default is 8042.
Path used to get disk used/total, this can also be a list of paths to different partitions, separated by :. The default is your home directory ~.


The following values are gathered. Most (all?) of them are also available per tracker (and per media type?).

Item Numbers

d.total, d.started, d.stopped, d.complete, d.incomplete, d.seeding, d.leeching, d.active, d.messages

These are the associated view sizes; could be sampled more often, and the average values taken.

Item Size
d.size_bytes, d.left_bytes, d.size_files
d.up_rate, d.down_rate, d.skip_rate

open_sockets, cputime, pcpu, pmem, sz, rsz, vsz

See man ps for most of these.

Also, the usual machine statistics (CPU load, disk usage and I/O, network traffic) are sampled (by collectd, or using collectd plugins, or some system stats package?).

Later Extensions

These are probably not sampled that often, or we need to define an extra view to allow efficient sampling.

As histogram counters?

event_closed, …

Counters for all event.download.* events.

peers_total, peers_encrypted, peers_incoming, peers_obfuscated, peers_preferred, peers_snubbed, peers_unwanted

With some patches compiled into rTorrent, the additional values network.http.open, and network.open_files are available.

Event Handling

TODO – see the old docs for anything not yet moved.

Queue Manager: Planned Features

These aren’t implemented yet…

ExecCommand (planned)

TODO pyrocore.torrent.jobs:ExecCommand runs an external command in a shell, i.e. it simply runs cron jobs. The reasons for not using cron instead are these: 1. You can have all your rTorrent-related background processing at one place, and the commands see the same environment as pyrotorque. 1. pyrotorque offers more flexible scheduling, including the ability to run jobs at sub-minute intervals.

RemoteWatch (planned)

TODO pyrocore.torrent.watch:RemoteWatch polls a (S)FTP source for new .torrent files, creates a local copy, and loads that into the client.

ItemPoller (planned)

TODO pyrocore.torrent.: maintains an updated copy of all rTorrent items, as a service for the other jobs.

ActionRule (planned)

TODO pyrocore.torrent.filter:ActionRule is rtcontrol in form of a house-keeping job, and using this is way more efficient than an equivalent rtcontrol cron job; due to that, they can be run a lot more frequently.

TorrentMirror (planned)

TODO pyrocore.torrent.filter:TorrentMirror allows you to transfer a torrent’s data from the local client to other remote clients using a specified tracker (at the start, a locally running “bttrack”). In a nutshell, it allows you to transfer any filtered item automatically to a remote location via bittorrent.

CompletionHandler (planned)

TODO pyrocore.torrent.: moves completed data to a target directory, according to flexible rules.

StatsArchiver (planned)

TODO pyrocore.torrent.: keeps a continuous archive of some statistical values (like bandwidth) so they can later be rendered into graphs.

See RtorrentMonitoring for more details.