3.3.14. Master-Worker API

This section describes the master-worker interface. It covers the communication protocol of the “classic” remote Worker. Notice there are other types of workers which behave a bit differently, such as Local Worker and Latent Worker.

3.3.14.1. Connection

The interface is based on Twisted’s Perspective Broker, which operates over TCP connections.

The worker connects to the master, using the parameters supplied to buildbot-worker create-worker. It uses a reconnecting process with an exponential backoff, and will automatically reconnect on disconnection.

Once connected, the worker authenticates with the Twisted Cred (newcred) mechanism, using the username and password supplied to buildbot-worker create-worker. The mind behind the worker is the worker bot instance (class buildbot_worker.pb.BotPb).

On the master side, the realm is implemented by buildbot.pbmanager.Dispatcher, which examines the username of incoming avatar requests. There are special cases for change and debug, which are not discussed here. For all other usernames, the botmaster is consulted, and if a worker with that name is configured, its buildbot.worker.Worker instance is returned as the perspective.

3.3.14.2. Workers

At this point, the master-side Worker object has a pointer to the remote worker-side Bot object in its self.worker, and the worker-side Bot object has a reference to the master-side Worker object in its self.perspective.

Bot methods

The worker-side Bot object has the following remote methods:

remote_getCommands

Returns a dictionary for all commands the worker recognizes: the key of the dictionary is the command name and the command version is the value.

remote_setBuilderList

Given a list of builders and their build directories, ensures that those builders, and only those builders, are running. This can be called after the initial connection is established, with a new list, to add or remove builders.

This method returns a dictionary of WorkerForBuilder objects - see below.

remote_print

Adds a message to the worker logfile.

remote_getWorkerInfo

Returns a dictionary with the contents of the worker’s info/ directory (i.e. file name is used as key and file contents as the value). This dictionary also contains the following keys:

environ

copy of the workers environment

system

OS the worker is running (extracted from Python’s os.name)

basedir

base directory where the worker is running

numcpus

number of CPUs on the worker, either as configured or as detected (since buildbot-worker version 0.9.0)

version

worker’s version (same as the result of remote_getVersion call)

worker_commands

worker supported commands (same as the result of remote_getCommands call)

remote_getVersion

Returns the worker’s version.

remote_shutdown

Shuts down the worker cleanly.

Worker methods

The master-side object has the following method:

perspective_keepalive

Does nothing - used to keep traffic flowing over the TCP connection

3.3.14.3. Setup

After the initial connection and trading of a mind (buildbot_worker.pb.BotPb) for an avatar (Worker), the master calls the Bot’s setBuilderList method to set up the proper builders on the worker side. This method returns a reference to each of the new worker-side WorkerForBuilderPb objects, described below. Each of these is handed to the corresponding master-side WorkerForBuilder object.

This immediately calls the remote setMaster method, and then the print method.

3.3.14.4. Pinging

To ping a remote Worker, the master calls its print method.

3.3.14.5. Building

When a build starts, the master calls the worker’s startBuild method. Each BuildStep instance will subsequently call the startCommand method, passing a reference to itself as the stepRef parameter. The startCommand method returns immediately, and the end of the command is signalled with a call to a method on the master-side BuildStep object.

3.3.14.6. Worker For Builders

Each worker has a set of builders which can run on it. These are represented by distinct classes on the master and worker, just like the Worker and Bot objects described above.

On the worker side, builders are represented as instances of the buildbot_worker.pb.WorkerForBuilderPb class. On the master side, they are represented by the buildbot.process.workerforbuilder.WorkerForBuilder class. The identical names are a source of confusion. The following will refer to these as the worker-side and master-side Worker For Builder classes. Each object keeps a reference to its opposite in self.remote.

Worker-Side WorkerForBuilderPb Methods

remote_setMaster

Provides a reference to the master-side Worker For Builder

remote_print

Adds a message to the worker logfile; used to check round-trip connectivity

remote_startBuild

Indicates that a build is about to start, and that any subsequent commands are part of that build

remote_startCommand

Invokes a command on the worker side

remote_interruptCommand

Interrupts the currently-running command

Master-side WorkerForBuilder Methods

The master side does not have any remotely-callable methods.

3.3.14.7. Commands

The actual work done by the worker is represented on the master side by a buildbot.process.remotecommand.RemoteCommand instance.

The command instance keeps a reference to the worker-side buildbot_worker.pb.WorkerForBuilderPb, and calls methods like remote_startCommand to start new commands. Once that method is called, the WorkerForBuilderPb instance keeps a reference to the command, and calls the following methods on it:

Master-Side RemoteCommand Methods

remote_update

Update information about the running command. See below for the format.

remote_complete

Signal that the command is complete, either successfully or with a Twisted failure.

3.3.14.8. Updates

Updates from the worker, sent via remote_update, are a list of individual update elements. Each update element is, in turn, a list of the form [data, 0], where the 0 is present for historical reasons. The data is a dictionary, with keys describing the contents. The updates are handled by remote_update.

Updates with different keys can be combined into a single dictionary or delivered sequentially as list elements, at the worker’s option.

To summarize, an updates parameter to remote_update might look like this:

[
    [ { 'header' : 'running command..' }, 0 ],
    [ { 'stdout' : 'abcd', 'stderr' : 'local modifications' }, 0 ],
    [ { 'log' : ( 'cmd.log', 'cmd invoked at 12:33 pm\n' ) }, 0 ],
    [ { 'rc' : 0 }, 0 ],
]

Defined Commands

The following commands are defined on the workers.

shell

Runs a shell command on the worker. This command takes the following arguments:

command

The command to run. If this is a string, it will be passed to the system shell as a string. Otherwise, it must be a list, which will be executed directly.

workdir

The directory in which to run the command, relative to the builder dir.

env

A dictionary of environment variables to augment or replace the existing environment on the worker. In this dictionary, PYTHONPATH is treated specially: it should be a list of path components, rather than a string, and will be prepended to the existing Python path.

initial_stdin

A string which will be written to the command’s standard input before it is closed.

want_stdout

If false, then no updates will be sent for stdout.

want_stderr

If false, then no updates will be sent for stderr.

usePTY

If true, the command should be run with a PTY (POSIX only). This defaults to False.

not_really

If true, skip execution and return an update with rc=0.

timeout

Maximum time without output before the command is killed.

maxTime

Maximum overall time from the start before the command is killed.

max_lines

Maximum overall produced lines by the command, then it is killed.

logfiles

A dictionary specifying logfiles other than stdio. Keys are the logfile names, and values give the workdir-relative filename of the logfile. Alternately, a value can be a dictionary; in this case, the dictionary must have a filename key specifying the filename, and can also have the following keys:

follow

Only follow the file from its current end-of-file, rather that starting from the beginning.

logEnviron

If false, the command’s environment will not be logged.

The shell command sends the following updates:

stdout

The data is a bytestring which represents a continuation of the stdout stream. Note that the bytestring boundaries are not necessarily aligned with newlines.

stderr

Similar to stdout, but for the error stream.

header

Similar to stdout, but containing data for a stream of Buildbot-specific metadata.

rc

The exit status of the command, where – in keeping with UNIX tradition – 0 indicates success and any nonzero value is considered a failure. No further updates should be sent after an rc.

failure_reason

Value is a string and describes additional scenarios when a process failed. The value of the failure_reason key can be one of the following:

• timeout if the command timed out due to time specified by the maxTime parameter being exceeded.

• timeout_without_output if the command timed out due to time specified by the timeout parameter being exceeded.

• max_lines_failure if the command is killed due to the number of lines specified by the max_lines parameter being exceeded.

log

This update contains data for a logfile other than stdio. The data associated with the update is a tuple of the log name and the data for that log. Note that non-stdio logs do not distinguish output, error, and header streams.

uploadFile

Upload a file from the worker to the master. The arguments are

workdir

Base directory for the filename, relative to the builder’s basedir.

workersrc

Name of the filename to read from, relative to the workdir.

writer

A remote reference to a writer object, described below.

maxsize

Maximum size, in bytes, of the file to write. The operation will fail if the file exceeds this size.

blocksize

The block size with which to transfer the file.

keepstamp

If true, preserve the file modified and accessed times.

The worker calls a few remote methods on the writer object. First, the write method is called with a bytestring containing data, until all of the data has been transmitted. Then, the worker calls the writer’s close, followed (if keepstamp is true) by a call to upload(atime, mtime).

This command sends rc and stderr updates, as defined for the shell command.

uploadDirectory

Similar to uploadFile, this command will upload an entire directory to the master, in the form of a tarball. It takes the following arguments:

workdir workersrc writer maxsize blocksize

See uploadFile for these arguments.

compress

Compression algorithm to use – one of None, 'bz2', or 'gz'.

The writer object is treated similarly to the uploadFile command, but after the file is closed, the worker calls the master’s unpack method with no arguments to extract the tarball.

This command sends rc and stderr updates, as defined for the shell command.

downloadFile

This command will download a file from the master to the worker. It takes the following arguments:

workdir

Base directory for the destination filename, relative to the builder basedir.

workerdest

Filename to write to, relative to the workdir.

reader

A remote reference to a reader object, described below.

maxsize

Maximum size of the file.

blocksize

The block size with which to transfer the file.

mode

Access mode for the new file.

The reader object’s read(maxsize) method will be called with a maximum size, which will return no more than that number of bytes as a bytestring. At EOF, it will return an empty string. Once EOF is received, the worker will call the remote close method.

This command sends rc and stderr updates, as defined for the shell command.

mkdir

This command will create a directory on the worker. It will also create any intervening directories required. It takes the following argument:

dir

Directory to create.

The mkdir command produces the same updates as shell.

rmdir

This command will remove a directory or file on the worker. It takes the following arguments:

dir

Directory to remove.

timeout maxTime

See shell above.

The rmdir command produces the same updates as shell.

cpdir

This command will copy a directory from one place to another place on the worker. It takes the following arguments:

fromdir

Source directory for the copy operation, relative to the builder’s basedir.

todir

Destination directory for the copy operation, relative to the builder’s basedir.

timeout maxTime

See shell above.

The cpdir command produces the same updates as shell.

stat

This command returns status information about a file or directory. It takes a single parameter, file, specifying the filename relative to the builder’s basedir.

It produces two status updates:

stat

The return value from Python’s os.stat.

rc

0 if the file is found, otherwise 1.

glob

This command finds all pathnames matching a specified pattern that uses shell-style wildcards. It takes a single parameter, path, specifying the pattern to pass to Python’s glob.glob function.

It produces two status updates:

files

The list of matching files returned from glob.glob

rc

0 if the glob.glob does not raise exception, otherwise 1.

listdir

This command reads the directory and returns the list with directory contents. It takes a single parameter, dir, specifying the directory relative to the builder’s basedir.

It produces two status updates:

files

The list of files in the directory returned from os.listdir

rc

0 if the os.listdir does not raise exception, otherwise 1.

rmfile

This command removes the file in the worker base directory. It takes a single parameter, path, specifying the file path relative to the builder’s basedir.

It produces one status update:

rc

0 if the os.remove does not raise exception, otherwise the corresponding errno.