18.3. select — Waiting for I/O completion¶
This module provides access to the select() and poll() functions
available in most operating systems, devpoll() available on
Solaris and derivatives, epoll() available on Linux 2.5+ and
kqueue() available on most BSD.
Note that on Windows, it only works for sockets; on other operating systems,
it also works for other file types (in particular, on Unix, it works on pipes).
It cannot be used on regular files to determine whether a file has grown since
it was last read.
Note
The selectors module allows high-level and efficient I/O
multiplexing, built upon the select module primitives. Users are
encouraged to use the selectors module instead, unless they want
precise control over the OS-level primitives used.
The module defines the following:
-
select.devpoll()¶ (Only supported on Solaris and derivatives.) Returns a
/dev/pollpolling object; see section /dev/poll Polling Objects below for the methods supported by devpoll objects.devpoll()objects are linked to the number of file descriptors allowed at the time of instantiation. If your program reduces this value,devpoll()will fail. If your program increases this value,devpoll()may return an incomplete list of active file descriptors.The new file descriptor is non-inheritable.
New in version 3.3.
Changed in version 3.4: The new file descriptor is now non-inheritable.
-
select.epoll(sizehint=-1, flags=0)¶ (Only supported on Linux 2.5.44 and newer.) Return an edge polling object, which can be used as Edge or Level Triggered interface for I/O events. sizehint and flags are deprecated and completely ignored.
See the Edge and Level Trigger Polling (epoll) Objects section below for the methods supported by epolling objects.
epollobjects support the context management protocol: when used in awithstatement, the new file descriptor is automatically closed at the end of the block.The new file descriptor is non-inheritable.
Changed in version 3.3: Added the flags parameter.
Changed in version 3.4: Support for the
withstatement was added. The new file descriptor is now non-inheritable.Deprecated since version 3.4: The flags parameter.
select.EPOLL_CLOEXECis used by default now. Useos.set_inheritable()to make the file descriptor inheritable.
-
select.poll()¶ (Not supported by all operating systems.) Returns a polling object, which supports registering and unregistering file descriptors, and then polling them for I/O events; see section Polling Objects below for the methods supported by polling objects.
-
select.kqueue()¶ (Only supported on BSD.) Returns a kernel queue object; see section Kqueue Objects below for the methods supported by kqueue objects.
The new file descriptor is non-inheritable.
Changed in version 3.4: The new file descriptor is now non-inheritable.
-
select.kevent(ident, filter=KQ_FILTER_READ, flags=KQ_EV_ADD, fflags=0, data=0, udata=0)¶ (Only supported on BSD.) Returns a kernel event object; see section Kevent Objects below for the methods supported by kevent objects.
-
select.select(rlist, wlist, xlist[, timeout])¶ This is a straightforward interface to the Unix
select()system call. The first three arguments are sequences of ‘waitable objects’: either integers representing file descriptors or objects with a parameterless method namedfileno()returning such an integer:- rlist: wait until ready for reading
- wlist: wait until ready for writing
- xlist: wait for an “exceptional condition” (see the manual page for what your system considers such a condition)
Empty sequences are allowed, but acceptance of three empty sequences is platform-dependent. (It is known to work on Unix but not on Windows.) The optional timeout argument specifies a time-out as a floating point number in seconds. When the timeout argument is omitted the function blocks until at least one file descriptor is ready. A time-out value of zero specifies a poll and never blocks.
The return value is a triple of lists of objects that are ready: subsets of the first three arguments. When the time-out is reached without a file descriptor becoming ready, three empty lists are returned.
Among the acceptable object types in the sequences are Python file objects (e.g.
sys.stdin, or objects returned byopen()oros.popen()), socket objects returned bysocket.socket(). You may also define a wrapper class yourself, as long as it has an appropriatefileno()method (that really returns a file descriptor, not just a random integer).Note
File objects on Windows are not acceptable, but sockets are. On Windows, the underlying
select()function is provided by the WinSock library, and does not handle file descriptors that don’t originate from WinSock.Changed in version 3.5: The function is now retried with a recomputed timeout when interrupted by a signal, except if the signal handler raises an exception (see PEP 475 for the rationale), instead of raising
InterruptedError.
-
select.PIPE_BUF¶ The minimum number of bytes which can be written without blocking to a pipe when the pipe has been reported as ready for writing by
select(),poll()or another interface in this module. This doesn’t apply to other kind of file-like objects such as sockets.This value is guaranteed by POSIX to be at least 512. Availability: Unix.
New in version 3.2.
18.3.1. /dev/poll Polling Objects¶
Solaris and derivatives have /dev/poll. While select() is
O(highest file descriptor) and poll() is O(number of file
descriptors), /dev/poll is O(active file descriptors).
/dev/poll behaviour is very close to the standard poll()
object.
-
devpoll.close()¶ Close the file descriptor of the polling object.
New in version 3.4.
-
devpoll.closed¶ Trueif the polling object is closed.New in version 3.4.
-
devpoll.fileno()¶ Return the file descriptor number of the polling object.
New in version 3.4.
-
devpoll.register(fd[, eventmask])¶ Register a file descriptor with the polling object. Future calls to the
poll()method will then check whether the file descriptor has any pending I/O events. fd can be either an integer, or an object with afileno()method that returns an integer. File objects implementfileno(), so they can also be used as the argument.eventmask is an optional bitmask describing the type of events you want to check for. The constants are the same that with
poll()object. The default value is a combination of the constantsPOLLIN,POLLPRI, andPOLLOUT.Warning
Registering a file descriptor that’s already registered is not an error, but the result is undefined. The appropriate action is to unregister or modify it first. This is an important difference compared with
poll().
-
devpoll.modify(fd[, eventmask])¶ This method does an
unregister()followed by aregister(). It is (a bit) more efficient that doing the same explicitly.
-
devpoll.unregister(fd)¶ Remove a file descriptor being tracked by a polling object. Just like the
register()method, fd can be an integer or an object with afileno()method that returns an integer.Attempting to remove a file descriptor that was never registered is safely ignored.
-
devpoll.poll([timeout])¶ Polls the set of registered file descriptors, and returns a possibly-empty list containing
(fd, event)2-tuples for the descriptors that have events or errors to report. fd is the file descriptor, and event is a bitmask with bits set for the reported events for that descriptor —POLLINfor waiting input,POLLOUTto indicate that the descriptor can be written to, and so forth. An empty list indicates that the call timed out and no file descriptors had any events to report. If timeout is given, it specifies the length of time in milliseconds which the system will wait for events before returning. If timeout is omitted, -1, orNone, the call will block until there is an event for this poll object.Changed in version 3.5: The function is now retried with a recomputed timeout when interrupted by a signal, except if the signal handler raises an exception (see PEP 475 for the rationale), instead of raising
InterruptedError.
18.3.2. Edge and Level Trigger Polling (epoll) Objects¶
http://linux.die.net/man/4/epoll
eventmask
Constant Meaning EPOLLINAvailable for read EPOLLOUTAvailable for write EPOLLPRIUrgent data for read EPOLLERRError condition happened on the assoc. fd EPOLLHUPHang up happened on the assoc. fd EPOLLETSet Edge Trigger behavior, the default is Level Trigger behavior EPOLLONESHOTSet one-shot behavior. After one event is pulled out, the fd is internally disabled EPOLLEXCLUSIVEWake only one epoll object when the associated fd has an event. The default (if this flag is not set) is to wake all epoll objects polling on a fd. EPOLLRDHUPStream socket peer closed connection or shut down writing half of connection. EPOLLRDNORMEquivalent to EPOLLINEPOLLRDBANDPriority data band can be read. EPOLLWRNORMEquivalent to EPOLLOUTEPOLLWRBANDPriority data may be written. EPOLLMSGIgnored.
-
epoll.close()¶ Close the control file descriptor of the epoll object.
-
epoll.closed¶ Trueif the epoll object is closed.
-
epoll.fileno()¶ Return the file descriptor number of the control fd.
-
epoll.fromfd(fd)¶ Create an epoll object from a given file descriptor.
-
epoll.register(fd[, eventmask])¶ Register a fd descriptor with the epoll object.
-
epoll.modify(fd, eventmask)¶ Modify a registered file descriptor.
-
epoll.unregister(fd)¶ Remove a registered file descriptor from the epoll object.
-
epoll.poll(timeout=-1, maxevents=-1)¶ Wait for events. timeout in seconds (float)
Changed in version 3.5: The function is now retried with a recomputed timeout when interrupted by a signal, except if the signal handler raises an exception (see PEP 475 for the rationale), instead of raising
InterruptedError.
18.3.3. Polling Objects¶
The poll() system call, supported on most Unix systems, provides better
scalability for network servers that service many, many clients at the same
time. poll() scales better because the system call only requires listing
the file descriptors of interest, while select() builds a bitmap, turns
on bits for the fds of interest, and then afterward the whole bitmap has to be
linearly scanned again. select() is O(highest file descriptor), while
poll() is O(number of file descriptors).
-
poll.register(fd[, eventmask])¶ Register a file descriptor with the polling object. Future calls to the
poll()method will then check whether the file descriptor has any pending I/O events. fd can be either an integer, or an object with afileno()method that returns an integer. File objects implementfileno(), so they can also be used as the argument.eventmask is an optional bitmask describing the type of events you want to check for, and can be a combination of the constants
POLLIN,POLLPRI, andPOLLOUT, described in the table below. If not specified, the default value used will check for all 3 types of events.Constant Meaning POLLINThere is data to read POLLPRIThere is urgent data to read POLLOUTReady for output: writing will not block POLLERRError condition of some sort POLLHUPHung up POLLRDHUPStream socket peer closed connection, or shut down writing half of connection POLLNVALInvalid request: descriptor not open Registering a file descriptor that’s already registered is not an error, and has the same effect as registering the descriptor exactly once.
-
poll.modify(fd, eventmask)¶ Modifies an already registered fd. This has the same effect as
register(fd, eventmask). Attempting to modify a file descriptor that was never registered causes anOSErrorexception with errnoENOENTto be raised.
-
poll.unregister(fd)¶ Remove a file descriptor being tracked by a polling object. Just like the
register()method, fd can be an integer or an object with afileno()method that returns an integer.Attempting to remove a file descriptor that was never registered causes a
KeyErrorexception to be raised.
-
poll.poll([timeout])¶ Polls the set of registered file descriptors, and returns a possibly-empty list containing
(fd, event)2-tuples for the descriptors that have events or errors to report. fd is the file descriptor, and event is a bitmask with bits set for the reported events for that descriptor —POLLINfor waiting input,POLLOUTto indicate that the descriptor can be written to, and so forth. An empty list indicates that the call timed out and no file descriptors had any events to report. If timeout is given, it specifies the length of time in milliseconds which the system will wait for events before returning. If timeout is omitted, negative, orNone, the call will block until there is an event for this poll object.Changed in version 3.5: The function is now retried with a recomputed timeout when interrupted by a signal, except if the signal handler raises an exception (see PEP 475 for the rationale), instead of raising
InterruptedError.
18.3.4. Kqueue Objects¶
-
kqueue.close()¶ Close the control file descriptor of the kqueue object.
-
kqueue.closed¶ Trueif the kqueue object is closed.
-
kqueue.fileno()¶ Return the file descriptor number of the control fd.
-
kqueue.fromfd(fd)¶ Create a kqueue object from a given file descriptor.
-
kqueue.control(changelist, max_events[, timeout=None]) → eventlist¶ Low level interface to kevent
- changelist must be an iterable of kevent object or
None - max_events must be 0 or a positive integer
- timeout in seconds (floats possible)
Changed in version 3.5: The function is now retried with a recomputed timeout when interrupted by a signal, except if the signal handler raises an exception (see PEP 475 for the rationale), instead of raising
InterruptedError.- changelist must be an iterable of kevent object or
18.3.5. Kevent Objects¶
https://www.freebsd.org/cgi/man.cgi?query=kqueue&sektion=2
-
kevent.ident¶ Value used to identify the event. The interpretation depends on the filter but it’s usually the file descriptor. In the constructor ident can either be an int or an object with a
fileno()method. kevent stores the integer internally.
-
kevent.filter¶ Name of the kernel filter.
Constant Meaning KQ_FILTER_READTakes a descriptor and returns whenever there is data available to read KQ_FILTER_WRITETakes a descriptor and returns whenever there is data available to write KQ_FILTER_AIOAIO requests KQ_FILTER_VNODEReturns when one or more of the requested events watched in fflag occurs KQ_FILTER_PROCWatch for events on a process id KQ_FILTER_NETDEVWatch for events on a network device [not available on Mac OS X] KQ_FILTER_SIGNALReturns whenever the watched signal is delivered to the process KQ_FILTER_TIMEREstablishes an arbitrary timer
-
kevent.flags¶ Filter action.
Constant Meaning KQ_EV_ADDAdds or modifies an event KQ_EV_DELETERemoves an event from the queue KQ_EV_ENABLEPermitscontrol() to returns the event KQ_EV_DISABLEDisablesevent KQ_EV_ONESHOTRemoves event after first occurrence KQ_EV_CLEARReset the state after an event is retrieved KQ_EV_SYSFLAGSinternal event KQ_EV_FLAG1internal event KQ_EV_EOFFilter specific EOF condition KQ_EV_ERRORSee return values
-
kevent.fflags¶ Filter specific flags.
KQ_FILTER_READandKQ_FILTER_WRITEfilter flags:Constant Meaning KQ_NOTE_LOWATlow water mark of a socket buffer KQ_FILTER_VNODEfilter flags:Constant Meaning KQ_NOTE_DELETEunlink() was called KQ_NOTE_WRITEa write occurred KQ_NOTE_EXTENDthe file was extended KQ_NOTE_ATTRIBan attribute was changed KQ_NOTE_LINKthe link count has changed KQ_NOTE_RENAMEthe file was renamed KQ_NOTE_REVOKEaccess to the file was revoked KQ_FILTER_PROCfilter flags:Constant Meaning KQ_NOTE_EXITthe process has exited KQ_NOTE_FORKthe process has called fork() KQ_NOTE_EXECthe process has executed a new process KQ_NOTE_PCTRLMASKinternal filter flag KQ_NOTE_PDATAMASKinternal filter flag KQ_NOTE_TRACKfollow a process across fork() KQ_NOTE_CHILDreturned on the child process for NOTE_TRACK KQ_NOTE_TRACKERRunable to attach to a child KQ_FILTER_NETDEVfilter flags (not available on Mac OS X):Constant Meaning KQ_NOTE_LINKUPlink is up KQ_NOTE_LINKDOWNlink is down KQ_NOTE_LINKINVlink state is invalid
-
kevent.data¶ Filter specific data.
-
kevent.udata¶ User defined value.
