piwheels¶

Welcome to the developer and administrator documentation for the piwheels service. If you simply want to use piwheels to install Python packages more quickly on your Raspberry Pi, head over to the piwheels homepage for the relevant information.

However, if you want to set up your own instance of piwheels, or hack on the piwheels codebase, you’re in the right place! These documents are far from comprehensive and there’s really no substitute for just playing with the system at first, but hopefully they’ll provide some answers to anyone who gets confused wandering through the code (although much of the documentation is derived from the code) and some starting points for those that want to get involved. For reference, the piwheels code is available from GitHub (naturally) which also hosts the issue tracker. Note there is a separate issue tracker for reporting issues with packages built by piwheels.org, which can be found at piwheels/packages.

Table of Contents¶

Overview¶

The piwheels project is designed to automate building of wheels from packages on PyPI for a set of pre-configured ABIs. As the name suggests, it was originally built for Raspberry Pis but there’s nothing particular in the codebase that should limit it to that platform. The system relies on the following components:

Component	Description
piw-master	Coordinates the various build slaves, using the database to store all relevant information, and keeps the web site up to date.
piw-slave	Builds package on behalf of the piwheels master. Is intended to run on separate machines to the master, partly for performance and partly for security.
piw-monitor	Provides a friendly curses-based UI for interacting with the piwheels master.
piw-sense	Provides a friendly Sense HAT-based UI for interacting with the piwheels master.
piw-initdb	A simple maintenance script for initializing or upgrading the database to the current version.
piw-import	A tool for importing wheels manually into the piwheels database and file-system.
piw-remove	A tool for manually removing builds from the database and file-system.
piw-rebuild	A tool for regenerating certain elements of the piwheels web-site.
piw-logger	A tool for transferring download statistics into the piwheels database.
database server	Currently only PostgreSQL is supported (and frankly that’s all we’re ever likely to support). This provides the master’s data store.
web server	Anything that can serve from a static directory is fine here. We use Apache in production.

Note

At present the master is a monolithic application, but the internal architecture is such that it could, in future, be split into three parts: one that deals exclusively with the database server, one that deals exclusively with the file-system served by the web server, and one that talks to the piwheels slave and monitor processes.

piw-master¶

The piw-master script is intended to be run on the database and file-server machine. It is recommended you do not run piw-slave on the same machine as the piw-master script. The database specified in the configuration must exist and have been configured with the piw-initdb script. It is recommended you run piw-master as an ordinary unprivileged user, although obviously it will need write access to the output directory.

Synopsis¶

piw-master [-h] [--version] [-c FILE] [-q] [-v] [-l FILE] [-d DSN]
                [-o PATH] [--dev-mode] [--pypi-xmlrpc URL]
                [--pypi-simple URL] [--status-queue ADDR]
                [--control-queue ADDR] [--import-queue ADDR]
                [--log-queue ADDR] [--slave-queue ADDR] [--file-queue ADDR]
                [--web-queue ADDR] [--builds-queue ADDR] [--db-queue ADDR]
                [--fs-queue ADDR] [--stats-queue ADDR]

Description¶

-h, --help¶: Show this help message and exit

--version¶: Show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

-q, --quiet¶: Produce less console output

-v, --verbose¶: Produce more console output

-l FILE, --log-file FILE¶: Log messages to the specified file

-d DSN, --dsn DSN¶: The database to use; this database must be configured with piw-initdb and the user must not be a PostgreSQL superuser (default: postgres:///piwheels)

-o PATH, --output-path PATH¶: The path under which the website should be written; must be writable by the current user

--dev-mode¶: Run the master in development mode, which reduces some timeouts and tweaks some defaults

--pypi-xmlrpc URL¶: The URL of the PyPI XML-RPC service (default: https://pypi.python.org/pypi)

--pypi-simple URL¶: The URL of the PyPI simple API (default: https://pypi.python.org/simple)

--status-queue ADDR¶: The address of the queue used to report status to monitors (default: ipc:///tmp/piw-status); this is usually an ipc address

--control-queue ADDR¶: The address of the queue a monitor can use to control the master (default: ipc:///tmp/piw-control); this is usually an ipc address

--import-queue ADDR¶: The address of the queue used by piw-import (default: ipc:///tmp/piw-import); this should always be an ipc address

--log-queue ADDR¶: The address of the queue used by piw-logger (default: ipc:///tmp/piw-logger); this should always be an ipc address

--slave-queue ADDR¶: The address of the queue used to talk to the build slaves (default: tcp://*:5555); this is usually a tcp address

--file-queue ADDR¶: The address of the queue used to transfer files from slaves (default: tcp://*:5556); this is usually a tcp address

--builds-queue ADDR¶: The address of the queue used to store pending builds (default: inproc://builds)

--db-queue ADDR¶: The address of the queue used to talk to the database server (default: inproc://db)

--fs-queue ADDR¶: The address of the queue used to talk to the file- system server (default: inproc://fs)

--stats-queue ADDR¶: The address of the queue used to send statistics to the collator task (default: inproc://stats)

Deployment¶

A typical deployment of the master service on a Raspbian server goes something like this (each step assumes you start as root):

Install the pre-requisite software:

# apt install postgresql-9.6 apache2 python3-psycopg2 python3-geoip
# apt install python3-sqlalchemy python3-urwid python3-zmq python3-voluptuous python3-chameleon
# pip install piwheels[monitor,master,logger]

Set up the (unprivileged) piwheels user and the output directory:

# groupadd piwheels
# useradd -g piwheels -m piwheels
# mkdir /var/www/piwheels
# chown piwheels:piwheels /var/www/piwheels

Set up the database:

# su - postgres
$ createuser piwheels
$ createdb -O postgres piwheels
$ piw-initdb

Set up the web server:
- Point the document root to the output path (/var/www/piwheels above, but it can be anywhere your piwheels user has write access to; naturally you want to make sure your web-server’s user only has read access to the location).
- Set up SSL for the web server (e.g. with Let’s Encrypt; the dehydrated utility is handy for getting and maintaining the SSL certificates).
Start the master running (it’ll take quite a while to populate the list of packages and versions from PyPI on the initial run so get this going before you start bringing up build slaves):
```
# su - piwheels
$ piw-master -v
```
Deploy some build slaves; see piw-slave for deployment instructions.

Automatic start¶

If you wish to ensure that the master starts on every boot-up, you may wish to define a systemd unit for it. Example units can be also be found in the root of the piwheels repository:

# wget https://raw.githubusercontent.com/piwheels/piwheels/master/piwheels-master.service
# cp piwheels-master.service /etc/systemd/system/
# systemctl daemon-reload
# systemctl enable piwheels-master
# systemctl start piwheels-master

Upgrades¶

The master will check that build slaves have the same version number and will reject them if they do not. Furthermore, it will check the version number in the database’s configuration table matches its own and fail if it does not. Re-run the piw-initdb script as the PostgreSQL super-user to upgrade the database between versions (downgrades are not supported, so take a backup first!).

piw-slave¶

The piw-slave script is intended to be run on a standalone machine to build packages on behalf of the piw-master script. It is intended to be run as an unprivileged user with a clean home-directory. Any build dependencies you wish to use must already be installed. The script will run until it is explicitly terminated, either by Ctrl+C, SIGTERM, or by the remote piw-master script.

Synopsis¶

piw-slave [-h] [--version] [-c FILE] [-q] [-v] [-l FILE] [-m HOST]
          [-t DURATION]

Description¶

-h, --help¶: Show this help message and exit

--version¶: Show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

-q, --quiet¶: Produce less console output

-v, --verbose¶: Produce more console output

-l FILE, --log-file FILE¶: Log messages to the specified file

-m HOST, --master HOST¶: The IP address or hostname of the master server (default: localhost)

-t DURATION, --timeout DURATION¶: The time to wait before assuming a build has failed (default: 3h)

Deployment¶

Our typical method of deployment is to spin up a new Pi as a build slave (through Mythic Beasts’ control panel) then execute a script to install the piwheels code, and all the build dependencies that we feel are reasonable to support under various Raspbian versions. The deployment script can be found in the root of the piwheels repository:

# wget https://raw.githubusercontent.com/piwheels/piwheels/master/deploy_slave.sh
# chmod +x deploy_slave.sh
# ./deploy_slave.sh

However, you will very likely wish to customize this script for your own purposes, e.g. to support a different set of dependencies, or to customize the typical build environment.

Once the script is complete, simply switch to the unprivileged user used to run the build slave, and execute piw-slave. For example, assuming the master’s IP address is 10.0.0.1:

# su - piwheels
$ piw-slave -m 10.0.0.1

Automatic start¶

If you wish to ensure that the build slave starts on every boot-up, you may wish to define a systemd unit for it. Example units can be also be found in the root of the piwheels repository:

# wget https://raw.githubusercontent.com/piwheels/piwheels/master/piwheels-slave.service
# cp piwheels-slave.service /etc/systemd/system/
# systemctl daemon-reload
# systemctl enable piwheels-slave
# systemctl start piwheels-slave

Warning

Be aware that this example unit forces a reboot in the case that the build slave fails (as occasionally happens with excessively complex packages).

Because of this you must ensure that the slave executes successfully prior to installing the unit, otherwise you’re liable to leave your build slave in permanent reboot cycle. This isn’t a huge issue for a build slave that’s physically in front of you (from which you can detach and tweak the storage), but it may be an issue if you’re dealing with a cloud builder.

piw-monitor¶

The piw-monitor application is used to monitor (and optionally control) the piw-master script. Upon startup it will request the status of all build slaves currently known to the master, and will then continually update its display as the slaves progress through builds. The controls at the bottom of the display allow the administrator to pause or resume the master script, kill build slaves that are having issues (e.g. excessive resource consumption from a huge build) or terminate the master itself.

Synopsis¶

piw-monitor [-h] [--version] [-c FILE] [--status-queue ADDR]
            [--control-queue ADDR]

Description¶

-h, --help¶: Show this help message and exit

--version¶: Show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

--status-queue ADDR¶: The address of the queue used to report status to monitors (default: ipc:///tmp/piw-status)

--control-queue ADDR¶: The address of the queue a monitor can use to control the master (default: ipc:///tmp/piw-control)

Usage¶

The monitor application should be started on the same machine as the master after the piw-master script has been started. After initialization it will request the current status of all build slaves from the master, displaying this in a list in the middle of the screen.

The Tab key can be used to navigate between the list of build slaves and the controls at the bottom of the screen. Mouse control is also supported, provided the terminal emulator supports it. Finally, hot-keys for all actions are available. The actions are as follows:

Pause¶

Hotkey: p

Pauses operations on the master. This causes Cloud Gazer to stop querying PyPI, Slave Driver to return “SLEEP” in response to any build slave requesting new packages, and so on. This is primarily a debugging tool to permit the developer to peek at the system in a more or less frozen state before resuming things.

Resume¶

Hotkey: r

Resumes operations on the master when paused.

Kill Slave¶

Hotkey: k

The next time the selected build slave requests a new package (with “IDLE”) the master will return “BYE” indicating the slave should terminate. Note that this cannot kill a slave in the middle of a build (that would require a more complex asynchronous protocol in Slave Driver), but is useful for shutting things down in an orderly fashion.

Terminate Master¶

Hotkey: t

Tells the master to shut itself down. In a future version, the master should request all build slaves to terminate as well, but currently this is unimplemented.

Quit¶

Hotkey: q

Terminate the monitor. Note that this won’t affect the master.

piw-sense¶

The piw-sense application is an alternative monitor for the piw-master script that uses the Raspberry Pi Sense HAT as its user interface. Upon startup it will request the status of all build slaves currently known to the master, and will then continually update its display as the slaves progress through builds. The Sense HAT’s joystick can be used to navigate information about current builds, and kill builds slaves that are having issues, or terminate the master itself.

Synopsis¶

piw-sense [-h] [--version] [-c FILE] [--status-queue ADDR]
               [--control-queue ADDR] [-r DEGREES]

Description¶

-h, --help¶: Show this help message and exit

--version¶: Show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

--status-queue ADDR¶: The address of the queue used to report status to monitors (default: ipc:///tmp/piw-status)

--control-queue ADDR¶: The address of the queue a monitor can use to control the master (default: ipc:///tmp/piw-control)

-r DEGREES, --rotate DEGREES¶: The rotation of the HAT in degrees; must be 0 (the default), 90, 180, or 270

Usage¶

The Sense monitor can be started on the same machine as the master after the piw-master script has been started. After initialization it will request the current status of all build slaves from the master.

Layout¶

The top three (normally blue) rows of the display are used for some important statistics:

The top row represents the ping time from the master, or more specifically the time since the last message was received. This will continually increase (changing white), and reset with each message received. If 30 seconds elapse without any messages being received, this row will pulse red until another message is received, resetting the count.
The second row represents available disk space for the output directory on the master. White pixels represent remaining space, and the scale is simply percentage (all blue = 0%, all white = 100%).
The third row represents the number of pending builds on the master. The scale is one white pixel = 8 builds in the queue (with partial shades representing <8 builds).

The remaining rows represent all build slaves. Each pixel represents a single build slave, working vertically then horizontally. Build slaves are sorted first by ABI, then by label (as in piw-monitor).

A gray pixel indicates an idle build slave.
A green pixel indicates an active build.
A blue pixel indicates an active file transfer after a successful build.
A purple pixel indicates a build slave cleaning up after a build.
A yellow pixel indicates an active build that’s been running for more than 15 minutes; not necessarily a problem but longer than average.
A red pixel indicates a build slave that’s either timed out or been terminated; it should disappear from the display within a few seconds.

Navigation¶

The pixel that pulses white indicates your current position, which can be moved with the Sense HAT joystick. Pressing the joystick in when a build-slave is selected (indicated by it pulsing white) will bring up detailed information on that build slave.

Scroll left and right to navigate through the build-slave information (label, ABI, current task, and kill option). Press the joystick in to return to the main display (optionally killing the build slave if the kill screen is selected).

Scroll the cursor off the top of the display to go to detailed statistics information. Scroll left and right to navigate through the available statistics (ping time, disk free, queue size, build rate, total build time, and total build size). Most statistics are displayed as scrolling text, and a background fill representing the information graphically. Scroll down to return to the main screen.

Scroll the cursor off the bottom of the display to go to the quit and terminate options (scroll left and right to navigate between them). Press the joystick in to activate either option, or scroll up to return to the main screen.

piw-initdb¶

The piw-initdb script is used to initialize or upgrade the piwheels master database. The target PostgreSQL database must already exist, and the DSN should connect as a cluster superuser (e.g. the postgres user), in contrast to the piw-master script which should not use the cluster superuser. The script will prompt before making any permanent alterations, and all actions will be executed within a single transaction so that in the event of failure the database will be left unchanged. Nonetheless, it is strongly recommended you take a backup of your database before using this script for upgrades.

Synopsis¶

piw-initdb [-h] [--version] [-c FILE] [-q] [-v] [-l FILE] [-d DSN]
           [-u NAME] [-y]

Description¶

-h, --help¶: show this help message and exit

--version¶: show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

-q, --quiet¶: produce less console output

-v, --verbose¶: produce more console output

-l FILE, --log-file FILE¶: log messages to the specified file

-d DSN, --dsn DSN¶: The database to create or upgrade; this DSN must connect as the cluster superuser (default: postgres:///piwheels)

-u NAME, --user NAME¶: The name of the ordinary piwheels database user (default: piwheels); this must not be a cluster superuser

-y, --yes¶: Proceed without prompting before init/upgrades

Usage¶

This script is intended to be used after installation to initialize the piwheels master database. Note that it does not create the database or the users for the database. It merely creates the tables, views, and other structures within an already existing database. See the Overview chapter for typical usage.

The script can also be used to upgrade an existing piwheels database to the latest version. The update scripts used attempt to preserve all data, and all upgrades are performed in a single transaction so that, theoretically, if anything goes wrong the database should be rolled back to its original state. However, it is still strongly recommended that you back up your master database before proceeding with any upgrade.

piw-import¶

The piw-import script is used to inject the specified file(s) manually into the piwheels database and file-system. This script must be run on the same node as the piw-master script. If multiple files are specified, they are registered as produced by a single build.

Synopsis¶

piw-import [-h] [--version] [-c FILE] [-q] [-v] [-l FILE]
           [--package PACKAGE] [--package-version VERSION] [--abi ABI]
           [--duration DURATION] [--output FILE] [-y] [-d]
           [--import-queue ADDR]
           files [files ...]

Description¶

-h, --help¶: Show this help message and exit

--version¶: Show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

-q, --quiet¶: Produce less console output

-v, --verbose¶: Produce more console output

-l FILE, --log-file FILE¶: Log messages to the specified file

--package PACKAGE¶: The name of the package to import; if omitted this will be derived from the file(s) specified

--package-version VERSION¶: The version of the package to import; if omitted this will be derived from the file(s) specified

--abi ABI¶: The ABI of the package to import; if omitted this will be derived from the file(s) specified

--duration DURATION¶: The time taken to build the package (default: 0s)

--output FILE¶: The filename containing the build output to insert into the database; if this is omitted an appropriate message will be inserted instead

-y, --yes¶: Run non-interactively; never prompt during operation

-d, --delete¶: Remove the specified file(s) after a successful import; if the import fails, no files will be removed

--import-queue ADDR¶: The address of the queue used by piw-import (default: (ipc:///tmp/piw-import); this should always be an ipc address

Usage¶

This utility is used to import wheels manually into the system. This is useful with packages which have no source available on PyPI, or binary-only packages from third parties. If invoked with multiple files, all files will be associated with a single “build” and the build will be for the package and version of the first file specified. No checks are made for equality of package name or version (as several packages on PyPI would violate such a rule!).

The utility can be run in a batch mode with --yes but still requires invoking once per build required (you cannot register multiple builds in a single invocation).

The return code will be 0 if the build was registered and all files were uploaded successfully. Additionally the --delete option can be specified to remove the source files once all uploads are completed successfully. If anything fails, the return code will be non-zero and no files will be deleted.

The utility should only ever be run directly on the master node (opening the import queue to other machines is a potential security risk).

piw-rebuild¶

The piw-rebuild script is used to inject rebuild requests for various web pages into the piwheels system. This script must be run on the same node as the piw-master script.

Synopsis¶

piw-rebuild [-h] [--version] [-c FILE] [-q] [-v] [-l FILE] [-y]
            [--import-queue ADDR]
            part [package]

Description¶

-h, --help¶: Show this help message and exit

--version¶: Show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

-q, --quiet¶: Produce less console output

-v, --verbose¶: Produce more console output

-l FILE, --log-file FILE¶: Log messages to the specified file

-y, --yes¶: Run non-interactively; never prompt during operation

--import-queue ADDR¶: The address of the queue used by piw-rebuild (default: (ipc:///tmp/piw-import); this should always be an ipc address

Usage¶

This utility is used to request rebuilds of parts of the piwheels website. This is primarily useful after manual fixes to the database, manipulation of the file-system, or after large-scale upgrades which require rebuilding many pages.

The mandatory part parameter can be one of the following values, which specify which part of the website to rebuild:

Part	Description
home	Rebuild the home-page (/index.html)
search	Rebuild the JSON search-index (/packages.json)
project	Rebuild the project-page for the specified package (/project/package/index.html)
index	Rebuild the simple-index and the project-page for the specified package (/simple/package/index.html and /project/package/index.html)

If part is “project” or “index” you may optionally specify a package name for which to rebuild the specified part. If the package name is omitted, the utility will request a rebuild of the specified part for all known packages in the system.

Warning

In the case a rebuild of all packages is requested, you will be prompted to make sure you wish to continue (this option can take hours to process on a system with many builds). The --yes option can be used to skip this prompt but should be used carefully!

Note that the utility only requests the rebuild of the specified part. This request will be queued, and acted upon as soon as The Scribe reaches it but there is no guarantee this has occurred by the time the utility exits. The return code will be 0 if the rebuild request was queued successfully. If anything fails the return code will be non-zero and the request may or may not have been queued.

The utility should only ever be run directly on the master node (opening the import queue to other machines is a potential security risk).

piw-remove¶

The piw-remove script is used to manually remove a package (or a version of a package) from the system. All builds for the specified package (or version) will be forgotten, and all files generated by such builds will be deleted.

By default, the package (or version) removed will be deleted entirely (not marked to skip). Optionally, you can provide a skip reason; in this case the version will not be deleted (though its builds and files will), but will be left marked to skip to prevent future rebuilds.

Synopsis¶

piw-remove [-h] [--version] [-c FILE] [-q] [-v] [-l FILE] [-y]
           [-s REASON] [--import-queue ADDR]
           package [version]

Description¶

package¶: The name of the package to remove

version¶: The version of the package to remove. If omitted, removes the entire package

-h, --help¶: Show this help message and exit

--version¶: Show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

-q, --quiet¶: Produce less console output

-v, --verbose¶: Produce more console output

-l FILE, --log-file FILE¶: Log messages to the specified file

-y, --yes¶: Run non-interactively; never prompt during operation

-s REASON, --skip REASON¶: Leave the version in place, but marked with a reason to prevent future build attempts

--import-queue ADDR¶: The address of the queue used by piw-remove (default: (ipc:///tmp/piw-import); this should always be an ipc address

Usage¶

This utility is typically used in response to a request from a package maintainer to remove a specific build from the system. Usually because the presence of a piwheels build is causing issues in and of itself.

Note

Older versions of piwheels didn’t heed PyPI deletion messages. This is no longer the case and this utility is no longer required to manually remove deleted packages.

The utility can be run in a batch mode with --yes but still requires invoking once per deletion required (you cannot remove multiple versions in a single invocation).

The return code will be 0 if the package (or version) was successfully removed. If anything fails, the return code will be non-zero and no files should be deleted (but this cannot be guaranteed in all circumstances).

The utility should only ever be run directly on the master node (opening the import queue to other machines is a potential security risk).

piw-logger¶

The piw-logger script is intended for use as an Apache “piped log script” but can also be used to feed pre-existing Apache logs to the master by feeding logs to the script’s stdin. This script must be run on the same node as the piw-master script.

Synopsis¶

piw-logger [-h] [--version] [-c FILE] [-q] [-v] [-l FILE]
                [--format FORMAT] [--log-queue ADDR] [--drop]
                [files [files ...]]

Description¶

files¶: The log file(s) to load into the master; if omitted or “-” then stdin will be read which is the default for piped log usage

-h, --help¶: Show this help message and exit

--version¶: Show program’s version number and exit

-c FILE, --configuration FILE¶: Specify a configuration file to load

-q, --quiet¶: Produce less console output

-v, --verbose¶: Produce more console output

-l FILE, --log-file FILE¶: Log messages to the specified file

--format FORMAT¶: The Apache log format that log lines will be expected to be in (default: combined); the short-cuts common, combined and common_vhost can be used in addition to Apache LogFormat strings

--log-queue ADDR¶: The address of the queue used by piw-logger (default: (ipc:///tmp/piw-logger); this should always be an ipc address

--drop¶: Drop log records if unable to send them to the master after a short timeout; this should generally be specified when piw-logger is used as a piped log script

Usage¶

This utility is typically used to pipe logs from a web-server, such as Apache into the piwheels database where they can be used for analysis, and to keep the stats on the homepage up to date. Apache provides a capability to pipe all logs to a given script which can be used directly with piw-logger.

A typical configuration under a Debian-like operating system might use the Apache CustomLog directive as follows, within the Apache virtual host reponsible for serving files to pip clients:

ErrorLog ${APACHE_LOG_DIR}/ssl_error.log
CustomLog ${APACHE_LOG_DIR}/ssl_access.log combined
CustomLog "|/usr/local/bin/piw-logger --drop" combined

Development¶

The main GitHub repository for the project can be found at:

https://github.com/piwheels/piwheels

After cloning, we recommend you set up a virtualenv for development and then execute make develop within that virtualenv. This should install all requirements for executing all tools, building the documentation and executing the test suite.

Testing¶

Executing the test suite requires that you have a local PostgreSQL installation configured with an unprivileged user, a privileged super user, and a test database.

The test suite uses environment variables to discover the name of the test database, and the aforementioned users. See the top of tests/conftest.py for more details. A typical execution of the test suite might look as follows:

$ export PIWHEELS_TESTDB=piwtest
$ export PIWHEELS_USER=piwheels
$ export PIWHEELS_PASS=piwheels
$ export PIWHEELS_SUPERUSER=piwsuper
$ export PIWHEELS_SUPERPASS=foobar
$ cd piwheels
$ make test

You may wish to construct a script for exporting the environment variables, or add these values to your ~/.bashrc.

Note

If you are not using your local PostgreSQL installation for anything else you may wish to set fsync=off and synchronous_commit=off in your local postgresql.conf to speed up execution of the test suite. Do NOT do this on any production PostgreSQL server!

Design¶

Although the piwheels master appears to be a monolithic script, it’s actually composed of numerous (often extremely simple) tasks. Each task runs its own thread and all communication between tasks takes place over ZeroMQ sockets. This is also how communication occurs between the master and the piw-slave, and the piw-monitor.

The following diagram roughly illustrates all the tasks in the system (including those of the build slaves and the monitor), along with details of the type of ZeroMQ socket used to communicate between them:

It may be confusing that the file server and database server appear to be separate to the master in the diagram. This is deliberate as the system’s architecture is such that certain tasks can be easily broken off into entirely separate processes (potentially on separate machines), if required in future (either for performance or security reasons).

Tasks¶

The following sections document the tasks shown above (listed from the “front” at PyPI to the “back” at Users):

Cloud Gazer¶

Implemented in: piwheels.master.cloud_gazer.CloudGazer.

This task is the “front” of the system. It follows PyPI’s event log for new package and version registrations, and writes those entries to the database. It does this via The Oracle.

The Oracle¶

Implemented in: piwheels.master.the_oracle.TheOracle.

This task is the main interface to the database. It accepts requests from other tasks (“register this new package”, “log this build”, “what files were built with this package”, etc.) and executes them against the database. Because database requests are extremely variable in their execution time, there are actually several instances of the oracle which sit behind Seraph.

Seraph¶

Implemented in: piwheels.master.seraph.Seraph.

Seraph is a simple load-balancer for the various instances of The Oracle. This is the task that actually accepts database requests. It finds a free oracle and passes the request along, passing back the reply when it’s finished.

The Architect¶

Implemented in: piwheels.master.the_architect.TheArchitect.

This task is the final database related task in the master script. Unlike The Oracle it periodically queries the database for the packages that need building and passes this information along to the Slave Driver.

Slave Driver¶

Implemented in: piwheels.master.slave_driver.SlaveDriver.

This task is the main coordinator of the build slaves’ activities. When a build slave first comes online it introduces itself to this task (with information including the ABI it can build for), and asks for a package to build. If there is a pending package matching the build slave’s ABI, it will be told to build that package.

Periodically, The Architect refreshes this task’s list of packages that require building.

Eventually the build slave will communicate whether or not the build succeeded, along with information about the build (log output, files generated, etc.). This task writes this information to the database via The Oracle. If the build was successful, it informs the File Juggler that it should expect a file transfer from the relevant build slave.

Finally, when all files from the build have been transferred, the Slave Driver informs the The Scribe that the package’s index and project page will need (re)writing. It also periodically informs Big Brother of the size of the build queue.

Mr. Chase¶

Implemented in: piwheels.master.mr_chase.MrChase.

This task talks to piw-import and handles importing builds manually into the system. It is essentially a cut-down version of the Slave Driver with a correspondingly simpler protocol. It is also the end-point for piw-rebuild and piw-remove.

This task writes information to the database via The Oracle. If the imported build was successful, it informs the File Juggler that it should expect a file transfer from the importer.

Finally, when all files from the build have been transferred, it informs the The Scribe that the package’s index and project pages will need (re)writing.

File Juggler¶

Implemented in: piwheels.master.file_juggler.FileJuggler.

This task handles file transfers from the build slaves to the master. Files are transferred in multiple (relatively small) chunks and are verified with the hash reported by the build slave (retrieved from the database via The Oracle).

Big Brother¶

Implemented in: piwheels.master.big_brother.BigBrother.

This task is a bit of a miscellaneous one. It sits around periodically generating statistics about the system as a whole (number of files, number of packages, number of successful builds, number of builds in the last hour, free disk space, etc.) and sends these off to the The Scribe.

The Scribe¶

Implemented in: piwheels.master.the_scribe.TheScribe.

This task generates the web output for piwheels. It generates the home-page with statistics from Big Brother, the overall package index, individual package file lists, and project pages with messages from Slave Driver.

The Secretary¶

Implemented in piwheels.master.the_secretary.TheSecretary.

This task sits in front of The Scribe and attempts to mitigate many of the repeated requests that typically get sent to it. For example, project pages (which are relatively expensive to generate, in database terms), may need regenerating every time a file is registered against a package version.

This often happens in a burst when a new package version is released, resulting in several (redundant) requests to re-write the same page with minimally changed information. The secretary buffers up such requests, eliminating duplicates before finally passing them to The Scribe for processing.

Queues¶

It should be noted that the diagram omits several queues for the sake of brevity. For instance, there is a simple PUSH/PULL control queue between the master’s “main” task and each sub-task which is used to relay control messages like PAUSE, RESUME, and QUIT.

Most of the protocols used by the queues are (currently) undocumented with the exception of those between the build slaves and the Slave Driver and File Juggler tasks (documented in the piw-slave chapter).

However, all protocols share a common basis: messages are lists of Python objects. The first element is always string containing the action. Further elements are parameters specific to the action. Messages are encoded with CBOR.

Protocols¶

The following sections document the protocols used between the build slaves and the three sub-tasks that they talk to in the piw-master. Each protocol operates over a separate queue. All messages in the piwheels system follow a similar structure of being a tuple containing:

A short unicode string indicating what sort of message it is.
Data. The structure of the data is linked to the type of the message, and validated on both transmission and reception (see piwheels.protocols for more information).

For example the message telling a build slave what package and version to build looks like this in Python syntax:

['BUILD', 'numpy', '1.14.0']

If a message is not associated with any data whatsoever, it is transmitted as a simple unicode string (without the list encapsulation). The serialization format for all messages in the system is currently CBOR.

Slave Driver¶

The queue that talks to Slave Driver is a ZeroMQ REQ socket, hence the protocol follows a strict request-reply sequence which is illustrated below:

The new build slave sends “HELLO” with data [build_timeout, master_timeout, py_version_tag, abi_tag, platform_tag, label, os_name, os_version, board_revision, board_serial] where:
- build_timeout is the slave’s configured timeout (the length of time after which it will assume a build has failed and attempt to terminate it) as a timedelta.
- master_timeout is the maximum length of time the slave will wait for communication from the master. After this timeout it will assume the connection has failed, terminate and clean-up any on-going build, then attempt to restart the connection to the master.
- py_version_tag is the python version the slave will build for (e.g. “27”, “35”, etc.)
- abi_tag is the ABI the slave will build for (e.g. “cp35m”)
- platform_tag is the platform of the slave (e.g. “linux_armv7l”)
- label is an identifying label for the slave (e.g. “slave2”); note that this label doesn’t have to be anything specific, it’s purely a convenience for administrators displayed in the monitor. In the current implementation this is the unqualified hostname of the slave
- os_name is a string identifying the operating system, e.g. “Raspbian GNU/Linux”.
- os_version is a string identifying the release of the operating system, e.g. “10 (buster)”.
- board_revision is a code indicating the revision of the board that the slave is running upon, e.g. “c03111” for a Raspberry Pi 4B.
- board_serial is the serial number of the board that the slave is running upon.
The master replies sends “ACK” with data [slave_id, pypi_url] where slave_id is an integer identifier for the slave. Strictly speaking, the build slave doesn’t need this identifier but it can be helpful for admins or developers to see the same identifier in logs on the master and the slave which is the only reason it is communicated.

The pypi_url is the URL the slave should use to fetch packages from PyPI.
The build slave sends “IDLE” to indicate that it is ready to accept a build job. The “IDLE” message is accompanied with the data [now, disk_total, disk_free, mem_total, mem_free, load_avg, cpu_temp] where:
- now is a datetime indicating the current time on the build slave.
- disk_total is the total size (in bytes) of the file-system used to build wheels.
- disk_free is the number of bytes free in the file-system used to build wheels.
- mem_total is the total size (in bytes) of the RAM on the build slave.
- mem_free is the number of bytes of RAM currently available (not necessarily unused, but potentially useable by builds).
- load_avg is the one minute load average.
- cpu_temp is the temperature, in degrees celsius of the CPU.
The master can reply with “SLEEP” which indicates that no jobs are currently available for that slave (e.g. the master is paused, or the build queue is empty, or there are no builds for the slave’s particular ABI at this time). In this case the build slave should pause a while (the current implementation waits 10 seconds) before retrying “IDLE”.
The master can also reply with “DIE” which indicates the build slave should shutdown. In this case, after cleaning up any resources the build slave should send back “BYE” and terminate (generally speaking, whenever the slave terminates it should send “BYE” no matter where in the protocol it occurs; the master will take this as a sign of termination).
The master can also reply “BUILD” with data [package, version] where package is the name of a package to build and version is the version to build. At this point, the build slave should attempt to locate the package on PyPI and build a wheel from it.
While the build is underway, the slave must periodically ping the master with the “BUSY” message, which is accompanied by the exact same stats as in the “IDLE” message.
If the master wishes the build slave to continue with the build it will reply with “CONT”. If the master wants to build slave to terminate the build early it will reply with “DONE” (goto step 13).
Assuming the master doesn’t request termination of the build, eventually it will finish. In response to the next “CONT” message, the slave sends “BUILT” with data [status, duration, output, files]:
- status is True if the build succeeded and False otherwise.
- duration is a timedelta value indicating the length of time it took to build in seconds.
- output is a string containing the complete build log.
- files is a list of file state tuples containing the following fields in the specified order:
  - filename is the filename of the wheel.
  - filesize is the size in bytes of the wheel.
  - filehash is the SHA256 hash of the wheel contents.
  - package_tag is the package tag extracted from the filename.
  - package_version_tag is the version tag extracted from the filename.
  - py_version_tag is the python version tag extracted from the filename.
  - abi_tag is the ABI tag extracted from the filename (sanitized).
  - platform_tag is the platform tag extracted from the filename.
  - dependencies is a set of dependency tuples containing the following fields in the specified order:
    - tool is the name of the tool used to install the dependency
    - package is the name of the package to install with the tool
If the build succeeded, the master will send “SEND” with data filename where filename is one of the names transmitted in the prior “BUILT” message.
At this point the slave should use the File Juggler protocol documented below to transmit the contents of the specified file to the master. When the file transfer is complete, the build slave sends “SENT” to the master.
If the file transfer fails to verify, or if there are more files to send the master will repeat the “SEND” message. Otherwise, if all transfers have completed and have been verified, the master replies with “DONE”.
The build slave is now free to destroy all resources associated with the build, and returns to step 3 (“IDLE”).

If at any point, the master takes longer than master_timeout (default: 5 minutes) to respond to a slave’s request, the slave will assume the master has disappeared. If a build is still active, it will be cleaned up and terminated, the connection to the master will be closed, the slave’s ID will be reset and the slave must restart the protocol from the top (“HELLO”).

This permits the master to be upgraded or replaced without having to shutdown and restart the slaves manually. It is possible that the master is restarted too fast for the slave to notice. In this case the slave’s next message will be mis-interpreted by the master as an invalid initial message, and it will be ignored. However, this is acceptable behaviour as the re-connection protocol described above will then effectively restart the slave after the master_timeout has elapsed.

Mr Chase (importing)¶

The queue that talks to Mr. Chase is a ZeroMQ REQ socket, hence the protocol follows a strict request-reply sequence which is illustrated below (see below for documentation of the “REMOVE” path):

The importer sends “IMPORT” with data [slave_id, package, version, abi_tag, status, duration, output, files]:
- slave_id is the integer id of the build slave that created the wheel. This is usually 0 and is ignored by the master anyway.
- package is the name of the package that the build is for.
- version is the version of the package that the build is for.
- abi_tag is either None, indicating that the master should use the “default” (minimum) build ABI registered in the system, or is a string indicating the ABI that the build was attempted for.
- status is True if the build succeeded and False otherwise.
- duration is a float value indicating the length of time it took to build in seconds.
- output is a string containing the complete build log.
- files is a list of file state tuples containing the following fields in the specified order:
  - filename is the filename of the wheel.
  - filesize is the size in bytes of the wheel.
  - filehash is the SHA256 hash of the wheel contents.
  - package_tag is the package tag extracted from the filename.
  - package_version_tag is the version tag extracted from the filename.
  - py_version_tag is the python version tag extracted from the filename.
  - abi_tag is the ABI tag extracted from the filename (sanitized).
  - platform_tag is the platform tag extracted from the filename.
  - dependencies is a set of dependency tuples containing the following fields in the specified order:
    - tool is the name of the tool used to install the dependency
    - package is the name of the package to install with the tool
If the import information is insufficient or incorrect, the master will send “ERROR” with data message which is the description of the error that occurred.
If the import information is okay, the master will send “SEND” with data filename for each file mentioned in the build.
At this point the importer should use the File Juggler protocol to transmit the contents of the specified file to the master. When the file transfer is complete, the importer sends “SENT” to the master.
If the file transfer fails to verify, or if there are more files to send the master will repeat the “SEND” message. Otherwise, if all transfers have completed and have been verified, the master replies with “DONE”.
The importer is now free to remove all files associated with the build, if requested to.

Mr Chase (removing)¶

The queue that talks to Mr. Chase is a ZeroMQ REQ socket, hence the protocol follows a strict request-reply sequence which is illustrated below (see above for documentation of the IMPORT path):

The utility sends “REMOVE” with data [package, version, skip]:
- package is the name of the package to remove.
- version is the version of the package to remove.
- skip is a string containing the reason the version should never be built again, or is a blank string indicating the version should be rebuilt.
If the removal fails (e.g. if the package or version does not exist), the master will send “ERROR” with data message (a string describing the error that occurred).
If the removal is successful, the master replies with “DONE”.

Mr Chase (rebuilding)¶

The queue that talks to Mr. Chase is a ZeroMQ REQ socket, hence the protocol follows a strict request-reply sequence which is illustrated below (see above for documentation of the IMPORT path):

The utility sends “REBUILD” with data [part, package]:
- part is the part of the website to rebuild. It must be one of “HOME”, “SEARCH”, “PROJECT” or “BOTH”.
- package is the name of the package to rebuild indexes and/or project pages for or None if pages for all packages should be rebuilt. This parameter is omitted if part is “HOME” or “SEARCH”.
If the rebuild request fails (e.g. if the package does not exist), the master will send “ERROR” with data message (a string describing the error that occurred).
If the rebuild request is successful, the master replies with “DONE”.

File Juggler¶

The queue that talks to File Juggler is a ZeroMQ DEALER socket. This is because the protocol is semi-asynchronous (for performance reasons). For the sake of illustration, a synchronous version of the protocol is illustrated below:

The build slave initially sends “HELLO” with data slave_id where slave_id is the integer identifier of the slave. The master knows what file it requested from this slave (with “SEND” to the Slave Driver), and knows the file hash it is expecting from the “BUILT” message.
The master replies with “FETCH” with data [offset, length] where offset is a byte offset into the file, and length is the number of bytes to send.
The build slave replies with “CHUNK” with data where data is a byte-string containing the requested bytes from the file.
The master now either replies with another “FETCH” message or, when it has all chunks successfully received, replies with “DONE” indicating the build slave can now close the file (though it can’t delete it yet; see the “DONE” message on the Slave Driver side for that).

“FETCH” messages may be repeated if the master drops packets (due to an overloaded queue). Furthermore, because the protocol is semi-asynchronous multiple “FETCH” messages will be sent before the master waits for any returning “CHUNK” messages.

Security¶

Care must be taken when running the build slave. Building all packages in PyPI effectively invites the denizens of the Internet to run arbitrary code on your machine. For this reason, the following steps are recommended:

Never run the build slave on the master; ensure they are entirely separate machines.
Run the build slave as an unprivileged user which has access to nothing it doesn’t absolutely require (it shouldn’t have any access to the master’s file-system, the master’s database, etc.)
Install the build slave’s code in a location the build slave’s unprivileged user does not have write access (i.e. not in a virtualenv under the user’s home dir).
Consider whether to make the unprivileged user’s home-directory read-only.

We have experimented with read-only home directories, but a significant portion of (usually scientifically oriented) packages attempt to be “friendly” and either write data to the user’s home directory or modify the user’s profile (~/.bashrc and so forth).

The quandry is whether it is better to fail with such packages (a read-only home-directory will most likely crash such setup scripts, failing the build), or partially support them (leaving the home-directory writeable even though the modifications on the build-slave won’t be recorded in the resulting wheel and thus won’t be replicated on user’s machines). There is probably no universally good answer.

Currently, while the build slave cleans up the temporary directory used by pip during wheel building, it doesn’t attempt to clean its own home directory (which setup scripts are free to write to). This is something that ought to be addressed in future as it’s a potentially exploitable hole.

Module Reference¶

This chapter contains all the documentation auto-generated from the source code. It is probably not terribly useful for reading through, but may be useful as a searchable reference.

piwheels.master ¶

piwheels.master.db ¶

This module defines the low level database API, Database. This is a simple core SQLAlchemy affair which runs trivial queries against the PostgreSQL database. All the serious logic is defined within views in the database itself.

class piwheels.master.db.Database(dsn)[source]¶

PiWheels database connection class

add_new_package(package, skip='', description='')[source]¶: Insert a new package record into the database. Returns True if the row was inserted successfully, or False if a key violation occurred.

add_new_package_version(package, version, released=None, skip='')[source]¶: Insert a new package version record into the database. Returns True if the row was inserted successfully, or False if a key violation occurred.

add_package_name(package, name, seen)[source]¶: Add a new package alias or update the last seen timestamp.

delete_build(package, version)[source]¶: Remove all builds for the specified package and version, along with all files records.

delete_package(package)[source]¶: Remove the specified package, along with all builds and files.

delete_version(package, version)[source]¶: Remove the specified version of the specified package, along with all builds and files.

get_all_package_versions()[source]¶: Returns the set of all known (package, version) tuples.

get_all_packages()[source]¶: Returns the set of all known package names.

get_build_abis()[source]¶: Return the set of ABIs that the master should attempt to build.

get_build_queue(limit=1000)[source]¶: Returns a mapping of ABI tags to an ordered list of up to limit package version tuples which currently need building for that ABI.

get_file_apt_dependencies(filename)[source]¶: Returns a set of the apt dependencies for the specified filename.

get_package_aliases(package)[source]¶: Retrieve all aliases for package (not including the canonical name itself).

get_package_description(package)[source]¶: Retrieve the description for package in the packages table.

get_package_files(package)[source]¶: Returns a mapping of filenames to file hashes; this is all the data required to build the simple index.html for the specified package.

get_project_display_name(package)[source]¶: Retrieve the last seen name for package.

get_project_files(package)[source]¶: Returns all details required to build the files table in the project page of the specified package.

get_project_versions(package)[source]¶: Returns all details required to build the versions table in the project page of the specified package.

get_pypi_serial()[source]¶: Return the serial number of the last PyPI event.

get_search_index()[source]¶: Return a mapping of all packages to their download count for the last month. This is used to construct the searchable package index.

get_statistics()[source]¶: Return various build related statistics from the database.

get_version_files(package, version)[source]¶: Returns the names of all files for version of package.

get_version_skip(package, version)[source]¶: Returns the reason for skipping version of package.

get_versions_deleted(package)[source]¶: Return any versions of package which have been marked for deletion.

load_rewrites_pending()[source]¶: Loads the rewrites-pending queue (the internal state of TheSecretary) from the database.

log_build(build)[source]¶: Log a build attempt in the database, including build output and wheel info if successful.

log_download(download)[source]¶: Log a download in the database, including data derived from JSON in pip’s user-agent.

log_json(json)[source]¶: Log a project’s JSON page hit in the database.

log_page(page)[source]¶: Log a web page hit in the database.

log_project(project)[source]¶: Log a project page hit in the database.

log_search(search)[source]¶: Log a search in the database, including data derived from JSON in pip’s user-agent.

package_marked_deleted(package)[source]¶: Check whether package has been marked for deletion.

save_rewrites_pending(queue)[source]¶: Save the rewrites-pending queue (the internal state of TheSecretary) in the database. The queue parameter is expected to be a list of RewritePendingRow tuples.

set_package_description(package, description)[source]¶: Update the description for package in the packages table.

set_pypi_serial(serial)[source]¶: Update the serial number of the last PyPI event.

skip_package(package, reason)[source]¶: Mark a package with a reason to prevent future builds of all versions (and all future versions).

skip_package_version(package, version, reason)[source]¶: Mark a version of a package with a reason to prevent future build attempts.

test_package(package)[source]¶: Check whether package already exists in the database. Returns a boolean.

test_package_version(package, version)[source]¶: Check whether version of package already exists in the database. Returns a boolean.

unyank_version(package, version)[source]¶: Mark the specified version of the specified package version as “unyanked”.

yank_version(package, version)[source]¶: Mark the specified version of the specified package version as “yanked”.

piwheels.master.cloud_gazer ¶

piwheels.master.the_oracle ¶

Defines TheOracle task and the DbClient RPC class for talking to it.

class piwheels.master.the_oracle.TheOracle(config)[source]¶

This task provides an RPC-like interface to the database; it handles requests such as registering a new package, version, or build, and answering queries about the hashes of files. The primary clients of this class are SlaveDriver, TheScribe, and CloudGazer.

Note that because database requests are notoriously variable in length the client RPC class (DbClient) doesn’t directly talk to TheOracle. Rather, multiple instances of TheOracle are spawned and Seraph sits in front of these acting as a simple load-sharing router for the RPC clients.

close()[source]¶: Close all registered queues. This should be overridden to close any additional queues the task holds which aren’t registered.

do_allpkgs()[source]¶: Handler for “ALLPKGS” message, sent by DbClient to request the set of all packages define known to the database.

do_allvers()[source]¶: Handler for “ALLVERS” message, sent by DbClient to request the set of all (package, version) tuples known to the database.

do_delbuild(package, version)[source]¶: Handler for “DELBUILD” message, sent by DbClient to remove all builds (and files and downloads by cascade) for version of package.

do_delpkg(package)[source]¶: Handler for “DELPKG” message, sent by DbClient to delete a package.

do_delver(package, version)[source]¶: Handler for “DELVER” message, sent by DbClient to delete a specific version of a package.

do_filedeps(filename)[source]¶: Handler for “FILEDEPS” message, sent by DbClient to request apt dependencies for filename, returned as a set of dependencies excluding those which are preinstalled in the distro version with the corresponding ABI tag.

do_getabis()[source]¶: Handler for “GETABIS” message, sent by DbClient to request the list of all ABIs to build for.

do_getdesc(package)[source]¶: Handler for “GETDESC” message, sent by DbClient to retrieve a package’s project description.

do_getpkgnames(package)[source]¶: Handler for “GETPKGNAMES” message, sent by DbClient to retrieve all aliases for package (not including the canonical name itself).

do_getprojname(package)[source]¶: Handler for “GETPROJNAME” message, sent by DbClient to retrieve the last seen name for package.

do_getpypi()[source]¶: Handler for “GETPYPI” message, sent by DbClient to request the record of the last serial number from the PyPI changelog.

do_getsearch()[source]¶: Handler for “GETSEARCH” message, sent by DbClient to request the recent download statistics, returned as a mapping of package to (downloads_recent, downloads_all) tuples.

do_getskip(package, version)[source]¶: Handler for “GETSKIP” message, send by DbClient to request the reason for skipping builds of version of package.

do_getstats()[source]¶: Handler for “GETSTATS” message, sent by DbClient to request the latest database statistics, returned as a list of (field, value) tuples.

do_loadrwp()[source]¶: Handler for “LOADRWP” message, sent by DbClient to request the content of the rewrites_pending table.

do_logbuild(build)[source]¶: Handler for “LOGBUILD” message, sent by DbClient to register a new build result.

do_logdownload(download)[source]¶: Handler for “LOGDOWNLOAD” message, sent by DbClient to register a new download.

do_logjson(json)[source]¶: Handler for “LOGJSON” message, sent by DbClient to register a new project JSON download.

do_logpage(page)[source]¶: Handler for “LOGPAGE” message, sent by DbClient to register a new web page hit.

do_logproject(project)[source]¶: Handler for “LOGPROJECT” message, sent by DbClient to register a new project page hit.

do_logsearch(search)[source]¶: Handler for “LOGSEARCH” message, sent by DbClient to register a new search.

do_newpkg(package, skip, description)[source]¶: Handler for “NEWPKG” message, sent by DbClient to register a new package.

do_newpkgname(package, name, seen)[source]¶: Handler for “NEWPKGNAME” message, sent by DbClient to register a new package alias or update the last seen timestamp.

do_newver(package, version, released, skip)[source]¶: Handler for “NEWVER” message, sent by DbClient to register a new (package, version) tuple.

do_pkgdeleted(package)[source]¶: Handler for “PKGDELETED” message, sent by DbClient to request whether or not the specified package has been marked for deletion.

do_pkgexists(package)[source]¶: Handler for “PKGEXISTS” message, sent by DbClient to request whether or not the specified package exists.

do_pkgfiles(package)[source]¶: Handler for “PKGFILES” message, sent by DbClient to request details of all wheels assocated with package.

do_projfiles(package)[source]¶: Handler for “PROJFILES” message, sent by DbClient to request file details of all versions of package.

do_projvers(package)[source]¶: Handler for “PROJVERS” message, sent by DbClient to request build and skip details of all versions of package.

do_saverwp(queue)[source]¶: Handler for “SAVERWP” message, sent by DbClient to request that queue is saved to the rewrites_pending table.

do_setdesc(package, description)[source]¶: Handler for “SETDESC” message, sent by DbClient to update a package’s project description.

do_setpypi(serial)[source]¶: Handler for “SETPYPI” message, sent by DbClient to update the last seen serial number from the PyPI changelog.

do_skippkg(package, reason)[source]¶: Handler for “SKIPPKG” message, sent by DbClient to skip building all versions of a package.

do_skipver(package, version, reason)[source]¶: Handler for “SKIPVER” message, sent by DbClient to skip building a specific version of a package.

do_unyankver(package, version)[source]¶: Handler for “UNYANKVER” message, sent by DbClient to mark a specific version of a package as not “yanked”.

do_verexists(package, version)[source]¶: Handler for “VEREXISTS” message, sent by DbClient to request whether or not the specified version of package exists.

do_verfiles(package, version)[source]¶: Handler for “VERFILES” message, sent by DbClient to request the filenames of all wheels associated with version of package.

do_versdeleted(package)[source]¶: Handler for “VERSDELETED” message, sent by DbClient to request any versions for package which have been marked for deletion.

do_yankver(package, version)[source]¶: Handler for “YANKVER” message, sent by DbClient to mark a specific version of a package as “yanked”.

handle_db_request(queue)[source]¶: Handle incoming requests from DbClient instances.

class piwheels.master.the_oracle.DbClient(config, logger=None)[source]¶

RPC client class for talking to TheOracle.

add_new_package(package, skip='', description='')[source]¶: See db.Database.add_new_package().

add_new_package_version(package, version, released=None, skip='')[source]¶: See db.Database.add_new_package_version().

add_package_name(package, name, seen=datetime.datetime(1970, 1, 1, 0, 0, tzinfo=datetime.timezone.utc))[source]¶: See db.Database.add_package_name().

delete_build(package, version)[source]¶: See db.Database.delete_build().

delete_package(package)[source]¶: See db.Database.delete_package().

delete_version(package, version)[source]¶: See db.Database.delete_version().

get_all_package_versions()[source]¶: See db.Database.get_all_package_versions().

get_all_packages()[source]¶: See db.Database.get_all_packages().

get_build_abis()[source]¶: See db.Database.get_build_abis().

get_file_apt_dependencies(filename)[source]¶: See db.Database.get_file_apt_dependencies().

get_package_aliases(package)[source]¶: See db.Database.get_package_aliases().

get_package_description(package)[source]¶: See db.Database.get_project_description().

get_package_files(package)[source]¶: See db.Database.get_package_files().

get_project_display_name(package)[source]¶: See db.Database.get_project_display_name().

get_project_files(package)[source]¶: See db.Database.get_project_files().

get_project_versions(package)[source]¶: See db.Database.get_project_versions().

get_pypi_serial()[source]¶: See db.Database.get_pypi_serial().

get_search_index()[source]¶: See db.Database.get_search_index().

get_statistics()[source]¶: See db.Database.get_statistics().

get_version_files(package, version)[source]¶: See db.Database.get_version_files().

get_version_skip(package, version)[source]¶: See db.Database.get_version_skip().

get_versions_deleted(package)[source]¶: See db.Database.get_versions_deleted().

load_rewrites_pending()[source]¶: See db.Database.load_rewrites_pending().

log_build(build)[source]¶: See db.Database.log_build().

log_download(download)[source]¶: See db.Database.log_download().

log_json(json)[source]¶: See db.Database.log_json().

log_page(page)[source]¶: See db.Database.log_page().

log_project(project)[source]¶: See db.Database.log_project().

log_search(search)[source]¶: See db.Database.log_search().

package_marked_deleted(package)[source]¶: See db.Database.package_marked_deleted().

save_rewrites_pending(queue)[source]¶: See db.Database.save_rewrites_pending().

set_package_description(package, description)[source]¶: See db.Database.update_project_description().

set_pypi_serial(serial)[source]¶: See db.Database.set_pypi_serial().

skip_package(package, reason)[source]¶: See db.Database.skip_package().

skip_package_version(package, version, reason)[source]¶: See db.Database.skip_package_version().

test_package(package)[source]¶: See db.Database.test_package().

test_package_version(package, version)[source]¶: See db.Database.test_package_version().

unyank_version(package, version)[source]¶: See db.Database.unyank_version().

yank_version(package, version)[source]¶: See db.Database.yank_version().

piwheels.master.seraph ¶

Defines the Seraph task; see class for more details.

class piwheels.master.seraph.Seraph(config)[source]¶

This task is a simple load-sharing router for TheOracle tasks.

handle_back(queue)[source]¶: Receive a response from an instance of TheOracle on the back queue. Strip off the worker’s address frame and add it back to the available queue then send the response back to the client that made the original request.

handle_front(queue)[source]¶: If any workers are currently available, receive DbClient requests from the front queue and send it on to the worker including the client’s address frame.

piwheels.master.the_architect ¶

Defines TheArchitect task; see class for more details.

class piwheels.master.the_architect.TheArchitect(config)[source]¶

This task queries the backend database to determine which versions of packages have yet to be built (and aren’t marked to be skipped). It pushes the results to SlaveDriver to sort out.

close()[source]¶: Close all registered queues. This should be overridden to close any additional queues the task holds which aren’t registered.

quit()[source]¶: Overridden to cancel any existing long-running query.

update_build_queue()[source]¶

The architect simply runs the build queue query repeatedly, with a break of a minute between each execution.

All entries found within this limit are sorted into per-ABI queues and pushed to SlaveDriver which queues and dispatches jobs to build ABI-matched slaves as they become available.

piwheels.master.slave_driver ¶

Defines the SlaveDriver task; see class for more details.

class piwheels.master.slave_driver.SlaveDriver(config)[source]¶

This task handles interaction with the build slaves using the slave protocol. Interaction is driven by the slaves (i.e. the master doesn’t push jobs, rather the slaves request a job and the master replies with the next (package, version) tuple from the internal “builds” queue).

The task also incidentally interacts with several other queues: the internal “status” queue is sent details of every reply sent to a build slave (the main_loop() method passes this information on to any listening monitors). Also, the internal “indexes” queue is informed of any packages that need web page indexes re-building (as a result of a successful build).

close()[source]¶: Close all registered queues. This should be overridden to close any additional queues the task holds which aren’t registered.

do_built(slave)[source]¶

Handler for the build slave’s “BUILT” message, which is sent after an attempted package build succeeds or fails. The handler logs the result in the database and, if files have been generated by the build, informs the FileJuggler task to expect a file transfer before sending “SEND” back to the build slave with the required filename.

If no files were generated (e.g. in the case of a failed build, or a degenerate success), “DONE” is returned indicating that the build slave is free to discard all resources generated during the build and return to its idle state.

do_busy(slave)[source]¶

Handler for the build slave’s “BUSY” message, which is sent periodically during package builds. If the slave fails to respond with a BUSY ping for a duration longer than SlaveState.busy_timeout then the master will assume the slave has died and remove it from the internal state mapping (if the slave happens to resurrect itself later the master will simply treat it as a new build slave).

In response to “BUSY” the master can respond “CONT” to indicate the build should continue processing, or “DONE” to indicate that the build slave should immediately terminate and discard the build and return to “IDLE” state.

do_bye(slave)[source]¶

Handler for the build slave’s final “BYE” message upon shutdown. This removes the associated state from the internal slaves dict.

Parameters:	slave (SlaveState) – The object representing the current status of the build slave.

do_hello(slave)[source]¶

Handler for the build slave’s initial “HELLO” message. This associates the specified slave state with the slave’s address and returns “HELLO” with the master’s id for the slave (the id communicated back simply for consistency of logging; administrators can correlate master log messages with slave log messages when both have the same id number; we can’t use IP address for this as multiple slaves can run on one machine).

Parameters:	slave (SlaveState) – The object representing the current status of the build slave.

do_idle(slave)[source]¶

Handler for the build slave’s “IDLE” message (which is effectively the slave requesting work). If the master wants to terminate the slave, it sends back “BYE”. If the build queue (for the slave’s ABI) is empty or the task is currently paused, “SLEEP” is returned indicating the slave should wait a while and then try again.

If a job can be retrieved from the (ABI specific) build queue, then a “BUILD” message is sent back with the required package and version.

Parameters:	slave (SlaveState) – The object representing the current status of the build slave.

do_sent(slave)[source]¶

Handler for the build slave’s “SENT” message indicating that it’s finished sending the requested file to FileJuggler. The FsClient RPC mechanism is used to ask FileJuggler to verify the transfer against the stored hash and, if this is successful, a message is sent to TheScribe to regenerate the package’s index.

If further files remain to be transferred, another “SEND” message is returned to the build slave. Otherwise, “DONE” is sent to free all build resources.

If a transfer fails to verify, another “SEND” message with the same filename is returned to the build slave.

handle_build(queue)[source]¶

Refresh the ABI-specific queues of package versions waiting to be built. The queues are limited to 1000 packages per ABI, and are kept as lists ordered by release date. When a message arrives from TheArchitect it refreshes (replaces) all current queues. There is, however, still a duplication possibility as TheArchitect doesn’t know what packages are actively being built; this method handles filtering out such packages.

Even if the active builds fail (because build slaves crash, or the network dies) this doesn’t matter as a future re-run of the build queue query will return these packages again, and if no build slaves are actively working on them at that time they will then be retried.

handle_control(queue)[source]¶

Handle incoming requests to the internal control queue.

This class understands a couple of extra control messages unique to it, specifically “KILL” to tell a build slave to terminate, “SKIP” to tell a build slave to terminate its current build immmediately, and “HELLO” to cause all “HELLO” messages from build slaves to be replayed (for the benefit of a newly attached monitor process).

handle_delete(queue)[source]¶

Handle package or version deletion requests.

When the PyPI upstream deletes a version or package, the CloudGazer task requests that other tasks perform the deletion on its behalf. In the case of this task, this involves cancelling any pending builds of that package (version), and ignoring any builds involving that package (version) in the next queue update from TheArchitect.

handle_slave(queue)[source]¶

Handle requests from build slaves.

See the piw-slave chapter for an overview of the protocol for messages between build slaves and SlaveDriver. This method retrieves the message from the build slave, finds the associated SlaveState and updates it with the message, then calls the appropriate message handler. The handler will be expected to return a reply (in the usual form of a list of strings) or None if no reply should be sent (e.g. for a final “BYE” message).

kill_slave(slave_id)[source]¶: Additional task control method to trigger a “KILL” message to the internal control queue. See handle_control() for more information.

list_slaves()[source]¶: Additional task control method to trigger a “HELLO” message to the internal control queue. See quit() for more information.

remove_expired()[source]¶: Remove slaves which have exceeded their timeout.

skip_slave(slave_id)[source]¶: Additional task control method to trigger a “SKIP” message to the internal control queue. See handle_control() for more information.

sleep_slave(slave_id)[source]¶: Additional task control method to trigger a “SLEEP” message to the internal control queue. See handle_control() for more information.

wake_slave(slave_id)[source]¶: Additional task control method to trigger a “WAKE” message to the internal control queue. See handle_control() for more information.

piwheels.master.mr_chase ¶

piwheels.master.file_juggler ¶

Defines the FileJuggler task and the FsClient RPC class for interacting with it.

exception piwheels.master.file_juggler.TransferError[source]¶: Base class for errors raised during a file transfer.

exception piwheels.master.file_juggler.TransferIgnoreChunk[source]¶: Exception raised when a build slave sends CHUNK instead of HELLO as the first message (see FileJuggler.new_transfer()).

exception piwheels.master.file_juggler.TransferDone[source]¶: Exception raised when a transfer is complete. It may seem a little odd to use an exception for this, but it is “exceptional” behaviour to terminate the file transfer.

class piwheels.master.file_juggler.FileJuggler(config)[source]¶

This task handles file transfers from the build slaves. The specifics of the file transfer protocol are best understood from the implementation of the FileState class.

However, to detail how a file transfer begins: when a build slave has successfully completed a build it informs the master via the SlaveDriver task. That task replies with a “SEND” instruction to the slave (including a filename). The slave then initiates the transfer with a “HELLO” message to this task. Once transfers are complete the slave sends a “SENT” message to the SlaveDriver task which verifies the transfer and either retries it (when verification fails) or sends back “DONE” indicating the slave can wipe the source file.

current_transfer(transfer, msg, *args)[source]¶

Called for messages associated with an existing file transfer.

Usually this is “CHUNK” indicating another chunk of data. Rarely, it can be “HELLO” if the master has fallen silent and dropped tons of packets.

Parameters:	transfer (TransferState) – The object representing the state of the transfer. msg (str) – The message sent during the transfer. *args – All additional arguments; for “CHUNK” the first must be the file offset and the second the data to write to that offset.

do_expect(slave_id, file_state)[source]¶

Message sent by FsClient to inform file juggler that a build slave is about to start a file transfer. The message includes the full FileState. The state is stored in the pending map.

Parameters:	slave_id (int) – The identity of the build slave about to begin the transfer. file_state (list) – The details of the file to be transferred including the expected hash.

do_verify(slave_id, package)[source]¶

Message sent by FsClient to request that juggler verify a file transfer against the expected hash and, if it matches, rename the file into its final location.

Parameters:	slave_id (int) – The identity of the build slave that sent the file. package (str) – The name of the package that the file is to be committed to, if valid.

handle_file(queue)[source]¶

Handle incoming file-transfer messages from build slaves.

The file transfer protocol is in some ways very simple (see the chart in the piw-slave chapter for an overview of the message sequence) and in some ways rather complex (read the ZeroMQ guide chapter on file transfers for more detail on why multiple messages must be allowed in flight simultaneously).

The “normal” state for a file transfer is to be requesting and receiving chunks. Anything else, including redundant re-sends, and transfer completion is handled as an exceptional case.

handle_fs_request(queue)[source]¶: Handle incoming messages from FsClient instances.

new_transfer(msg, *args)[source]¶

Called for messages initiating a new file transfer.

The first message must be HELLO along with the id of the slave starting the transfer. The metadata for the transfer will be looked up in the pending list (which is written to by do_expect()).

Parameters:	msg (str) – The message sent to start the transfer (must be “HELLO”) *args – All additional arguments (expected to be an integer slave id).

once()[source]¶: This method is called once before the task loop starts. It the task needs to do some initialization or setup within the task thread, this is the place to do it.

class piwheels.master.file_juggler.FsClient(config, logger=None)[source]¶

RPC client class for talking to FileJuggler.

expect(slave_id, file_state)[source]¶: See FileJuggler.do_expect().

verify(slave_id, package)[source]¶: See FileJuggler.do_verify().

piwheels.master.big_brother ¶

Defines the BigBrother task; see class for more details.

class piwheels.master.big_brother.BigBrother(config)[source]¶

This task periodically queries the database and output file-system for various statistics like the number of packages known to the system, the number built, the number of packages built in the last hour, the remaining file-system space, etc. These statistics are written to the internal “status” queue which main_loop() uses to pass statistics to any listening monitors.

close()[source]¶: Close all registered queues. This should be overridden to close any additional queues the task holds which aren’t registered.

handle_control(queue)[source]¶

Handle incoming requests to the internal control queue.

This just adds handling for the custom STATS verb to replay the master stats history.

piwheels.master.the_secretary ¶

Defines the TheSecretary task; see class for more details.

class piwheels.master.the_secretary.TheSecretary(config)[source]¶

This task buffers requests for the scribe, for the purpose of consolidating multiple consecutive (duplicate) requests.

Requests to write the project page for a package (which is a relatively expensive operation in terms of database accesses) can come in thick and fast, particularly when a new version is being registered with lots of files. There’s little point in writing the project page 5 times in as many seconds, or writing the project page, then the index and project page immediately afterward. This class is used to buffer requests for up to a minute, allowing us to eliminate many of the duplicate requests.

close()[source]¶: Close all registered queues. This should be overridden to close any additional queues the task holds which aren’t registered.

handle_input(queue)[source]¶

Handle incoming write requests with buffering and de-dupe.

Some incoming requests (currently “HOME”, “SEARCH”, “DELPKG”, and “DELVER”) are passed directly through to TheScribe as these are either sufficiently rare (“HOME”, “SEARCH”) that no benefit is gained by buffering them or sufficiently urgent (“DELPKG”, “DELVER”) that they must be acted on immediately.

For other requests (“PROJECT” and “BOTH”), requests can come thick and fast in the case of multiple file registrations picked up by CloudGazer. In this case, requests are buffered for a minute and de-duplicated; e.g. if several requests are made to re-write the project page for package “foo” within that period, they will be combined into a single request. After the minute of buffering, the request is passed down to TheScribe.

handle_output()[source]¶

Passes buffered requests downstream.

This sub-task runs periodically to pluck things from the internal buffer that have reached the minute delay, and passes them downstream to TheScribe. The process stops when we run out of things that have expired.

once()[source]¶: This method is called once before the task loop starts. It the task needs to do some initialization or setup within the task thread, this is the place to do it.

piwheels.master.the_scribe ¶

piwheels.slave ¶

piwheels.slave.builder ¶

piwheels.initdb ¶

Contains the functions that make up the piw-initdb script.

piwheels.initdb.main(args=None)[source]¶: This is the main function for the piw-initdb script. It creates the piwheels database required by the master or, if it already exists, upgrades it to the current version of the application.

piwheels.initdb.detect_users(conn, test_user)[source]¶: Test that the user for conn is a cluster superuser (so we can drop and create anything we want in the database), and that test_user (which will be granted limited rights to various objects for the purposes of the piw-master script) exists and is not a cluster superuser.

piwheels.initdb.detect_version(conn)[source]¶: Detect the version of the database. This is typically done by reading the contents of the configuration table, but before that was added we can guess a couple of versions based on what tables exist (or don’t). Returns None if the database appears uninitialized, and raises RuntimeError is the version is so ancient we can’t do anything with it.

piwheels.initdb.get_connection(dsn)[source]¶: Return an SQLAlchemy connection to the specified dsn or raise RuntimeError if the database doesn’t exist (the administrator is expected to create the database before running this script).

piwheels.initdb.get_script(version=None)[source]¶: Generate the script to get the database from version (the result of detect_version()) to the current version of the software. If version is None, this is simply the contents of the sql/create_piwheels.sql script. Otherwise, it is a concatenation of various update scripts.

piwheels.initdb.parse_statements(script)[source]¶

This is an extremely crude statement splitter for PostgreSQL’s dialect of SQL. It understands --comments, "quoted identifiers", 'string literals' and $delim$ extended strings $delim$, but not E'\escaped strings' or /* C-style comments */. If you start using such things in the update scripts, you’ll need to extend this function to accommodate them.

It returns a generator which yields individiual statements from script, delimited by semi-colon terminators.

piwheels.importer ¶

piwheels.remove ¶

Contains the functions that implement the piw-remove script.

piwheels.remove.main(args=None)[source]¶: This is the main function for the piw-remove script. It uses MrChase to remove built packages from the system.

piwheels.remove.do_remove(config)[source]¶

Handles constructing and sending the “REMOVE” message to master.mr_chase.MrChase.

Parameters:	config – The configuration obtained from parsing the command line.

piwheels.transport ¶

This module augments the classes provided by pyzmq (the 0MQ Python bindings) to use CBOR encoding, and voluptuous for message validation. It also tweaks a few minor things like using seconds for timeouts.

class piwheels.transport.Context[source]¶: Wrapper for 0MQ zmq.Context. This extends the socket() method to include parameters for the socket’s protocol and logger.

class piwheels.transport.Socket(socket, protocol=None, logger=None)[source]¶

Wrapper for zmq.Socket. This extends 0MQ’s sockets to include a protocol which will be used to validate messages that are sent and received (via a voluptuous schema), and a logger which can be used to debug socket behaviour.

bind(address)[source]¶: Binds the socket to listen on the specified address.

close(linger=None)[source]¶: Closes the socket. If linger is specified, it is the number of seconds to wait for pending messages to be flushed.

connect(address)[source]¶: Connects the socket to the listening socket at address.

drain()[source]¶: Receives all pending messages in the queue and discards them. This is typically useful during shutdown routines or for testing.

poll(timeout=None, flags=<MagicMock name='mock.POLLIN' id='140103954655496'>)[source]¶

Polls the socket for pending data (by default, when flags is POLLIN). If no data is available after timeout seconds, returns False. Otherwise returns True.

If flags is POLLOUT instead, tests whether the socket has available slots for queueing new messages.

recv(flags=0)[source]¶: Receives the next message as a bytes string.

recv_addr_msg(flags=0)[source]¶: Receive a CBOR-encoded message (and associated data) along with the address it came from (represented as a bytes string).

recv_msg(flags=0)[source]¶

Receive a CBOR-encoded message, returning a tuple of the unicode message string and its associated data. This is the primary method used in piwheels for receving information into a task.

The message, and its associated data, will be validated agains the protocol associated with the socket on construction.

recv_multipart(flags=0)[source]¶: Receives a multi-part message, returning its content as a list of bytes strings.

send(buf, flags=0)[source]¶: Send buf (a bytes string).

send_addr_msg(addr, msg, data=NoData, flags=0)[source]¶: Send a CBOR-encoded message (and associated data) to addr, a bytes string.

send_msg(msg, data=NoData, flags=0)[source]¶

Send the unicode string msg with its associated data as a CBOR-encoded message. This is the primary method used in piwheels for sending information between tasks.

The message, and its associated data, must validate against the protocol associated with the socket on construction.

send_multipart(msg_parts, flags=0)[source]¶: Send msg_parts, a list of bytes strings as a multi-part message which can be received intact with recv_multipart().

subscribe(topic)[source]¶: Subscribes SUB type sockets to the specified topic (a string prefix).

unsubscribe(topic)[source]¶: Unsubscribes SUB type sockets from the specified topic (a string prefix).

hwm¶: The high-water mark of the socket, i.e. the number of messages that can be queued before the socket blocks (or drops, depending on the socket type) messages.

class piwheels.transport.Poller[source]¶

Wrapper for 0MQ zmq.Poller. This simply tweaks 0MQ’s poller to use seconds for timeouts, and to return a dict by default from poll().

poll(timeout=None)[source]¶: Poll all registered sockets for the events they were registered with, for timeout seconds. Returns a dictionary mapping sockets to events or an empty dictinoary if the timeout elapsed with no events occurring.

register(sock, flags=<MagicMock name='mock.POLLIN.__or__()' id='140103954656952'>)[source]¶: Register sock with the poller, watching for events as specified by flags (which defaults to POLLIN and POLLOUT events).

unregister(sock)[source]¶: Unregister sock from the poller. After this, calls to poll() will never return references to sock.

piwheels.tasks ¶

Implements the base classes (Task and its derivative PauseableTask) which form the basis of all the tasks in the piwheels master.

exception piwheels.tasks.TaskQuit[source]¶: Exception raised when the “QUIT” message is received by the internal control queue.

class piwheels.tasks.Task(config, control_protocol=Protocol(recv={'PAUSE': NoData, 'RESUME': NoData, 'QUIT': NoData}, send={}))[source]¶

The Task class is a Thread derivative which is the base for all tasks in the piwheels master. The run() method is overridden to perform a simple task loop which calls poll() once a cycle to react to any messages arriving into queues, and to dispatch any periodically executed methods.

Queues are associated with handlers via the register() method. Periodic methods are associated with an interval via the every() method. These should be called during initialization (don’t attempt to register handlers from within the thread itself).

Generally this shouldn’t be used as a base-class. Use one of the descendents that implements a pausing mechanism, NonStopTask, PauseableTask, or PausingTask.

close()[source]¶: Close all registered queues. This should be overridden to close any additional queues the task holds which aren’t registered.

every(interval, handler)[source]¶

Register handler to be called every interval periodically.

Parameters:	interval (timedelta) – The time interval between each run of handler. handler – The function or method to call periodically.

force(handler)[source]¶: Force handler to run next time its interval is polled.

handle_control(queue)[source]¶

Default handler for the internal control queue. In this base class it simply handles the “QUIT” message by raising TaskQuit (which the run() method will catch and use as a signal to end).

Messages other than QUIT, PAUSE and RESUME raise TaskControl which can be caught in descendents to implement custom control messages.

once()[source]¶: This method is called once before the task loop starts. It the task needs to do some initialization or setup within the task thread, this is the place to do it.

pause()[source]¶: Requests that the task pause itself. This is an idempotent method; it’s always safe to call repeatedly and even if the task isn’t pauseable it’ll simply be ignored.

poll(timeout=1)[source]¶: This method is called once per loop of the task’s run() method. It runs all periodic handlers, then polls all registered queues and calls their associated handlers if the poll is successful.

quit()[source]¶: Requests that the task terminate at its earliest convenience. To wait until the task has actually closed, call join() afterwards.

register(queue, handler, flags=<MagicMock name='mock.POLLIN' id='140103954655496'>)[source]¶

Register queue to be polled on each cycle of the task. Any messages with the relevant flags (defaults to POLLIN) will trigger the specified handler method which is expected to take a single argument which will be queue.

Parameters:	queue (transport.Socket) – The queue to poll. handler – The function or method to call when a message with matching flags arrives in queue. flags (int) – The flags to match in the queue poller (defaults to `POLLIN`).

resume()[source]¶: Requests that the task resume itself. This is an idempotent method; it’s safe to call repeatedly and even if the task isn’t pauseable it’ll simply be ignored.

run()[source]¶: This method is the main task loop. Override this to perform one-off startup processing within the task’s background thread, and to perform any finalization required.

socket(sock_type, protocol=None)[source]¶: Construct a socket and link it to the logger for this task. This is primarily useful for debugging purposes, but also ensures that the task will implicitly close and clean up the socket when it closes.

class piwheels.tasks.PauseableTask(config, control_protocol=Protocol(recv={'PAUSE': NoData, 'RESUME': NoData, 'QUIT': NoData}, send={}))[source]¶

Derivative of Task that implements a rudimentary pausing mechanism. When the “PAUSE” message is received on the internal control queue, the task will enter a loop which simply polls the control queue waiting for “RESUME” or “QUIT”. No other work will be done (Task.loop() and Task.poll() will not be called) until the task is resumed (or terminated).

If you need a more complex pausing implementation which can still do some work while paused (to drain incoming queues for instance), use PausingTask instead.

handle_control(queue)[source]¶

Default handler for the internal control queue. In this base class it simply handles the “QUIT” message by raising TaskQuit (which the run() method will catch and use as a signal to end).

Messages other than QUIT, PAUSE and RESUME raise TaskControl which can be caught in descendents to implement custom control messages.

piwheels.states ¶

This module defines several classes which permit interested tasks to track the state of build slaves (SlaveState), file transfers (TransferState), build attempts (BuildState), build artifacts (FileState) and various loggers.

class piwheels.states.FileState(filename, filesize, filehash, package_tag, package_version_tag, py_version_tag, abi_tag, platform_tag, dependencies, transferred=False)[source]¶

Represents the state of an individual build artifact (a package file, or wheel) including its filename, filesize, the SHA256 filehash, and various tags extracted from the build. Also tracks whether or not the file has been transferred.

Parameters:

filename (str) – The original filename of the build artifact.
filesize (int) – The size of the file in bytes.
filehash (str) – The SHA256 hash of the file contents.
package_tag (str) – The package tag extracted from the filename (first “-” separated component).
package_version_tag (str) – The package version tag extracted from the filename (second “-” separated component).
py_version_tag (str) – The python version tag extracted from the filename (third from last “-” separated component).
abi_tag (str) – The python ABI tag extracted from the filename (second from last “-” separated component).
platform_tag (str) – The platform tag extracted from the filename (last “-” separated component).
dependencies (set) – The set of dependencies that are required to use this particular wheel.
transferred (bool) – True if the file has been transferred from the build slave that generated it to the file server.

as_message()[source]¶: Convert the FileState object into a simpler list for serialization and transport.

classmethod from_message(value)[source]¶: Convert the output from as_message() back into a BuildState.

verified()[source]¶: Called to set transferred to True after a file transfer has been successfully verified.

class piwheels.states.BuildState(slave_id, package, version, abi_tag, status, duration, output, files, build_id=None)[source]¶

Represents the state of a package build including the package, version, status, build duration, and all the lines of output. The files attribute is a mapping containing details of each successfully built package file.

Parameters:

slave_id (int) – The master’s identifier for the build slave.
package (str) – The name of the package to build.
version (str) – The version number of the package to build.
abi_tag (str) – The ABI for which the build was attempted (must not be 'none').
status (bool) – True if the build succeeded, False if it failed.
duration (timedelta) – The amount of time (in seconds) it took to complete the build.
output (str) – The log output of the build.
files (dict) – A mapping of filenames to FileState objects for each artifact produced by the build.
build_id (int) – The integer identifier generated for the build by the database (None until the build has been inserted into the database).

as_message()[source]¶: Convert the BuildState, and its nested FileState objects into simpler lists for serialization and transport.

classmethod from_message(value)[source]¶: Convert the output from as_message() back into a BuildState.

logged(build_id)[source]¶: Called to fill in the build’s ID in the backend database.

files¶: A mapping of filename to FileState instances.

next_file¶: Returns the filename of the next file that needs transferring or None if all files have been transferred.

transfers_done¶: Returns True if all files have been transferred.

class piwheels.states.SlaveState(address, build_timeout, busy_timeout, native_py_version, native_abi, native_platform, label, os_name, os_version, board_revision, board_serial)[source]¶

Tracks the state of a build slave. The master updates this state with each request and reply sent to and received from the slave, and this class in turn manages the associated BuildState (accessible from build) and TransferState (accessible from transfer). The class also tracks the time a request was last seen from the build slave, and includes a kill() method.

Parameters:

address (bytes) –
The slave’s ephemeral 0MQ address.

Note

This is not the slave’s IP address; it’s a unique identifier generated on connection to the master’s ROUTER socket. It will be different each time the slave re-connects (due to timeout, reboot, etc).
timeout (int) – The number of seconds after which any build will be considered to have timed out (and the slave will be assumed crashed).
native_py_version (str) – The slave’s native Python version.
native_abi (str) – The slave’s native Python ABI.
native_platform (str) – The slave’s native platform.
label (str) – A label representing the slave.

class piwheels.states.TransferState(slave_id, file_state)[source]¶

Tracks the state of a file transfer. All file transfers are held in temporary locations until verify() indicates the transfer was successful, at which point they are atomically renamed into their final location.

The state is intimately tied to the file transfer protocol and includes methods to write a recevied chunk(), and to determine the next chunk to fetch(), as well as a property to determine when the transfer is done.

Parameters:	slave_id (str) – The ID number of the slave which built the file. file_state (FileState) – The details of the file to be transferred (filename, size, hash, etc.)

class piwheels.states.DownloadState[source]¶

Represents the state of the log entry for a download of a package wheel file, including its filename, the user’s host IP, access timestamp and information about the operating system and installer.

Parameters:

filename (str) – The filename of the downloaded wheel file.
host – The hostname or IP address of the user.
timestamp (datetime.datetime) – The timestamp at which the file was downloaded.
or None arch (str) – The architecture of the user’s computer system (usually armv6 or armv7).
distro_name (str or None) – The user’s operating system distribution name (e.g. Raspbian).
distro_version (str or None) – The version of the user’s operating system distribution.
os_name (str or None) – The name of the user’s operating system (e.g. Linux).
os_version (str or None) – The version of the user’s operating system (e.g. Linux kernel version).
py_name (str or None) – The Python implementation used (e.g. CPython).
py_version (str or None) – The Python version used (e.g. 3.7.3).
installer_name (str or None) – The name of the tool used to install the file (e.g. pip).
installer_version (str or None) – The version of the tool (e.g. pip) used to install the file.
setuptools_version (str or None) – The version of setuptools used.

class piwheels.states.SearchState[source]¶

Represents the state of the log entry for an instance of a package search, including the package name, user’s host IP, access timestamp and information about the operating system and installer.

Parameters:

package (str) – The name of the package searched for.
host (str) – The hostname or IP address of the user.
timestamp (datetime.datetime) – The timestamp at which the search occurred.
arch (str or None) – The architecture of the user’s computer system (usually armv6 or armv7).
distro_name (str or None) – The user’s operating system distribution name (e.g. Raspbian).
distro_version (str or None) – The version of the user’s operating system distribution.
os_name (str or None) – The name of the user’s operating system (e.g. Linux).
os_version (str or None) – The version of the user’s operating system (e.g. Linux kernel version).
py_name (str or None) – The Python implementation used (e.g. CPython).
py_version (str or None) – The Python version used (e.g. 3.7.3).
installer_name (str or None) – The name of the tool used (e.g. pip).
installer_version (str or None) – The version of the tool (e.g. pip) used.
setuptools_version (str or None) – The version of setuptools used.

class piwheels.states.ProjectState[source]¶

Represents the state of the log entry for an instance of project page hit, including the page name, the user’s host IP, access timestamp and the user’s user_agent.

Parameters:	package (str) – The name of the package searched for. host (str) – The hostname or IP address of the user. timestamp (`datetime.datetime`) – The timestamp at which the page was accessed. user_agent (str) – The user agent of the page request.

class piwheels.states.JSONState[source]¶

Represents the state of the log entry for an instance of project JSON download, including the page name, the user’s host IP, access timestamp and the user’s user_agent.

Parameters:	package (str) – The name of the package whose JSON file was accessed. host (str) – The hostname or IP address of the user. timestamp (datetime.datetime) – The timestamp at which the page was accessed. user_agent (str) – The user agent of the request.

class piwheels.states.PageState[source]¶

Represents the state of the log entry for an instance of web page hit, including the page name, the user’s host IP, access timestamp and the user’s user_agent.

Parameters:	page (str) – The name of the page accessed. host (str) – The IP address of the user. timestamp (datetime.datetime) – The timestamp at which the page was accessed. user_agent (str) – The user agent of the page request.

class piwheels.states.SlaveStats[source]¶

class piwheels.states.MasterStats[source]¶

piwheels.states.mkdir_override_symlink(pkg_dir)[source]¶: Make pkg_dir, replacing any existing symlink in its place. See the notes in TheScribe.write_package_index() for more information.

piwheels.ranges ¶

A set of utility routines for efficiently tracking byte ranges within a stream. These are used to track which chunks of a file have been received during file transfers from build slaves.

See FileJuggler for the usage of these functions.

piwheels.ranges.consolidate(ranges)[source]¶

Given a list of ranges in ascending order, this generator function returns the list with any overlapping ranges consolidated into individual ranges. For example:

>>> list(consolidate([range(0, 5), range(4, 10)]))
[range(0, 10)]
>>> list(consolidate([range(0, 5), range(5, 10)]))
[range(0, 10)]
>>> list(consolidate([range(0, 5), range(6, 10)]))
[range(0, 5), range(6, 10)]

piwheels.ranges.exclude(ranges, ex)[source]¶

Given a list of non-overlapping ranges in ascending order, and a range ex to exclude, this generator function returns ranges with all values covered by ex removed from any contained ranges. For example:

>>> list(exclude([range(10)], range(2)))
[range(2, 10)]
>>> list(exclude([range(10)], range(2, 4)))
[range(0, 2), range(4, 10)]

piwheels.ranges.intersect(range1, range2)[source]¶

Given two ranges range1 and range2 (which must both have a step of 1), returns the range formed by the intersection of the two ranges, or None if the ranges do not overlap. For example:

>>> intersect(range(10), range(5))
range(0, 5)
>>> intersect(range(10), range(10, 2))
>>> intersect(range(10), range(2, 5))
range(2, 5)

piwheels.ranges.split(ranges, i)[source]¶

Given a list of non-overlapping ranges in ascending order, this generator function returns the list with the range containing i split into two ranges, one ending at i and the other starting at i. If i is not contained in any of the ranges, then ranges is returned unchanged. For example:

>>> list(split([range(10)], 5))
[range(0, 5), range(5, 10)]
>>> list(split([range(10)], 0))
[range(0, 10)]
>>> list(split([range(10)], 20))
[range(0, 10)]

License¶

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

The following copyright and license applies to the included cbor2 module only (all files under piwheels/cbor2):

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

piwheels¶

Table of Contents¶

Overview¶

piw-master¶

Synopsis¶

Description¶

Deployment¶

Automatic start¶

Upgrades¶

piw-slave¶

Synopsis¶

Description¶

Deployment¶

Automatic start¶

piw-monitor¶

Synopsis¶

Description¶

Usage¶

Pause¶

Resume¶

Kill Slave¶

Terminate Master¶

Quit¶

piw-sense¶

Synopsis¶

Description¶

Usage¶

Layout¶

Navigation¶

piw-initdb¶

Synopsis¶

Description¶

Usage¶

piw-import¶

Synopsis¶

Description¶

Usage¶

piw-rebuild¶

Synopsis¶

Description¶

Usage¶

piw-remove¶

Synopsis¶

Description¶

Usage¶

piw-logger¶

Synopsis¶

Description¶

Usage¶

Development¶

Testing¶

Design¶

Tasks¶

Cloud Gazer¶

The Oracle¶

Seraph¶

The Architect¶

Slave Driver¶

Mr. Chase¶

File Juggler¶

Big Brother¶

The Scribe¶

The Secretary¶

Queues¶

Protocols¶

Slave Driver¶

Mr Chase (importing)¶

Mr Chase (removing)¶

Mr Chase (rebuilding)¶

File Juggler¶

Security¶

Module Reference¶

License¶

Indexes and Tables¶