12. 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
- piwheels.master.cloud_gazer
- piwheels.master.the_oracle
- piwheels.master.seraph
- piwheels.master.the_architect
- piwheels.master.slave_driver
- piwheels.master.mr_chase
- piwheels.master.file_juggler
- piwheels.master.big_brother
- piwheels.master.the_secretary
- piwheels.master.the_scribe
- piwheels.slave
- piwheels.slave.builder
- piwheels.initdb
- piwheels.importer
- piwheels.remove
- piwheels.transport
- piwheels.tasks
- piwheels.states
- piwheels.ranges
12.2. 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_version
(package, version)[source]¶ Remove the specified version of the specified package, along with all builds and files.
-
get_build_abis
(exclude_skipped=False)[source]¶ Return a set of ABIs. If exclude_skipped is
False
, return all ABIs from the build_abis table, otherwise return only active ABIs (not skipped).
-
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_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_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_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_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_search
(search)[source]¶ Log a search in the database, including data derived from JSON in pip’s user-agent.
-
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 ofRewritePendingRow
tuples.
-
set_package_description
(package, description)[source]¶ Update the description for package in the packages table.
-
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.
-
12.4. 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
, andCloudGazer
.Note that because database requests are notoriously variable in length the client RPC class (
DbClient
) doesn’t directly talk toTheOracle
. Rather, multiple instances ofTheOracle
are spawned andSeraph
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_delver
(package, version)[source]¶ Handler for “DELVER” message, sent by
DbClient
to delete a specific version of a package.
-
do_getabis
(exclude_skipped)[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 therewrites_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 therewrites_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.
-
-
class
piwheels.master.the_oracle.
DbClient
(config, logger=None)[source]¶ RPC client class for talking to
TheOracle
.-
add_package_name
(package, name, seen=datetime.datetime(1970, 1, 1, 0, 0, tzinfo=datetime.timezone.utc))[source]¶
-
set_package_description
(package, description)[source]¶ See
db.Database.update_project_description()
.
-
12.5. 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.
12.6. 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.
-
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.
-
12.7. 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
. TheFsClient
RPC mechanism is used to askFileJuggler
to verify the transfer against the stored hash and, if this is successful, a message is sent toTheScribe
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 asTheArchitect
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 fromTheArchitect
.
-
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 associatedSlaveState
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) orNone
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.
-
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.
-
12.9. 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 theSlaveDriver
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 fullFileState
. The state is stored in thepending
map.Parameters:
-
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:
-
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.
-
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 bydo_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).
-
12.10. 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.
12.11. 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 toTheScribe
.
-
12.15. 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). ReturnsNone
if the database appears uninitialized, and raisesRuntimeError
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 isNone
, this is simply the contents of thesql/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 notE'\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.
12.17. piwheels.remove¶
Contains the functions that implement the piw-remove script.
12.18. 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 thesocket()
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.-
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.
-
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='139920042552680'>)[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_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_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 withrecv_multipart()
.
-
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 adict
by default frompoll()
.-
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.
-
12.19. 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 aThread
derivative which is the base for all tasks in the piwheels master. Therun()
method is overridden to perform a simple task loop which callspoll()
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 theevery()
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
, orPausingTask
.-
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.
-
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 therun()
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='139920042552680'>)[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.
-
-
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()
andTask.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 therun()
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.
-
12.20. 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, requires_python, dependencies, transferred=False)[source]¶ Represents the state of an individual build artifact (a package file, or wheel) including its
filename
,filesize
, the SHA256filehash
, and various tags extracted from the build. Also tracks whether or not the file has beentransferred
.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).
- requires_python (str) – The
Requires-Python
specification for the file. - 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 aBuildState
.
-
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
, buildduration
, and all the lines ofoutput
. Thefiles
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 nestedFileState
objects into simpler lists for serialization and transport.
-
classmethod
from_message
(value)[source]¶ Convert the output from
as_message()
back into aBuildState
.
-
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 frombuild
) andTransferState
(accessible fromtransfer
). The class also tracks the time a request was last seen from the build slave, and includes akill()
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.
- address (bytes) –
-
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 tofetch()
, as well as a property to determine when the transfer isdone
.Parameters:
-
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’shost
IP, accesstimestamp
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’shost
IP, accesstimestamp
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’shost
IP, accesstimestamp
and the user’suser_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’shost
IP, accesstimestamp
and the user’suser_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’shost
IP, accesstimestamp
and the user’suser_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.
12.21. 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)]