The Event Loop and the Hub

The hub is a special greenlet created automatically to run the event loop.

The current hub can be retrieved with get_hub.

get_hub(*args, **kwargs)[source]

Return the hub for the current thread.

If a hub does not exist in the current thread, a new one is created of the type returned by get_hub_class().

Deprecated since version 1.3b1: The *args and **kwargs arguments are deprecated. They were only used when the hub was created, and so were non-deterministic—to be sure they were used, all callers had to pass them, or they were order-dependent. Use set_hub instead.

class Hub(loop=None, default=None)[source]

Bases: gevent._hub_primitives.WaitOperationsGreenlet

A greenlet that runs the event loop.

It is created automatically by get_hub().


Every time this greenlet (i.e., the event loop) is switched to, if the current greenlet has a switch_out method, it will be called. This allows a greenlet to take some cleanup actions before yielding control. This method should not call any gevent blocking functions.


Wait until the watcher (which must not be started) is ready.

The current greenlet will be unscheduled during this time.

cancel_wait(watcher, error, close_watcher=False)

Cancel an in-progress call to wait() by throwing the given error in the waiting greenlet.

Changed in version 1.3a1: Added the close_watcher parameter. If true, the watcher will be closed after the exception is thrown. The watcher should then be discarded. Closing the watcher is important to release native resources.

Changed in version 1.3a2: Allow the watcher to be None. No action is taken in that case.

the event loop object (`ILoop`) associated with this hub and thus
this native thread.

Destroy this hub and clean up its resources.

If you manually create hubs, you should call this method before disposing of the hub object reference.

handle_error(context, type, value, tb)[source]

Called by the event loop when an error occurs. The arguments type, value, and tb are the standard tuple returned by sys.exc_info().

Applications can set a property on the hub with this same signature to override the error handling provided by this class.

Errors that are system errors are passed to handle_system_error().

Parameters:context – If this is None, indicates a system error that should generally result in exiting the loop and being thrown to the parent greenlet.
handle_system_error(type, value)[source]

Called from handle_error when the exception type is determined to be a system error.

System errors cause the exception to be raised in the main greenlet (the parent of this hub).


Wait for the event loop to finish. Exits only when there are no more spawned greenlets, started servers, active timeouts or watchers.

If timeout is provided, wait no longer for the specified number of seconds.

Returns True if exited because the loop finished execution. Returns False if exited because of timeout expired.


Entry-point to running the loop. This method is called automatically when the hub greenlet is scheduled; do not call it directly.

Raises:gevent.exceptions.LoopExit – If the loop finishes running. This means that there are no other scheduled greenlets, and no active watchers or servers. In some situations, this indicates a programming error.
NOT_ERROR = (<class 'greenlet.GreenletExit'>, <class 'SystemExit'>)

Instances of these classes are not considered to be errors and do not get logged/printed when raised by the event loop.

SYSTEM_ERROR = (<class 'KeyboardInterrupt'>, <class 'SystemExit'>, <class 'SystemError'>)

If instances of these classes are raised into the event loop, they will be propagated out to the main greenlet (where they will usually be caught by Python itself)


The stream to which exceptions will be written. Defaults to sys.stderr unless assigned to.

New in version 1.2a1.


Is this the hub for the main thread?

New in version 1.3b1.

name = ''

A string giving the name of this hub. Useful for associating hubs with particular threads. Printed as part of the default repr.

New in version 1.3b1.


The DNS resolver that the socket functions will use.


The threadpool associated with this hub.

Usually this is a gevent.threadpool.ThreadPool, but you can customize that.

Use this object to schedule blocking (non-cooperative) operations in a different thread to prevent them from halting the event loop.

threadpool_size = 10

The size we use for our threadpool. Either use a subclass for this, or change it immediately after creating the hub.

The Event Loop

The current event loop can be obtained with get_hub().loop. All implementations of the loop provide a common minimum interface.

interface ILoop[source]

The common interface expected for all event loops.


This is an internal, low-level interface. It may change between minor versions of gevent.


The methods that create event loop watchers are io, timer, signal, idle, prepare, check, fork, async_, child, stat. These all return various types of IWatcher.

All of those methods have one or two common arguments. ref is a boolean saying whether the event loop is allowed to exit even if this watcher is still started. priority is event loop specific.


Floating point number of seconds giving (approximately) the minimum resolution of a timer (and hence the minimun value the sleep can sleep for). On libuv, this is fixed by the library, but on libev it is just a guess and the actual value is system dependent.


Boolean indicating whether this is the default loop

run(nowait=False, once=False)

Run the event loop.

This is usually called automatically by the hub greenlet, but in special cases (when the hub is not running) you can use this to control how the event loop runs (for example, to integrate it with another event loop).

run_callback(func, *args)

Run the func passing it args at the next opportune moment.

This is a way of handing control to the event loop and deferring an action.


now() -> float

Return the loop’s notion of the current time.

This may not necessarily be related to time.time() (it may have a different starting point), but it must be expressed in fractional seconds (the same units used by time.time()).

idle(ref=True, priority=None)

Create and return a watcher that fires when the event loop is idle.

io(fd, events, ref=True, priority=None)

Create and return a new IO watcher for the given fd.

events is a bitmask specifying which events to watch for. 1 means read, and 2 means write.

timer(after, repeat=0.0, ref=True, priority=None)

Create and return a timer watcher that will fire after after seconds.

If repeat is given, the timer will continue to fire every repeat seconds.

signal(signum, ref=True, priority=None)

Create and return a signal watcher for the signal signum, one of the constants defined in signal.

This is platform and event loop specific.


Clean up resources used by this loop.

If you create loops (especially loops that are not the default) you should call this method when you are done with the loop.


As an implementation note, the libev C loop implementation has a finalizer (__del__) that destroys the object, but the libuv and libev CFFI implementations do not. The C implementation may change.

child(pid, trace=0, ref=True)

Create a watcher that fires for events on the child with process ID pid.

This is platform specific and not available on Windows.

async_(ref=True, priority=None)

Create a watcher that fires when triggered, possibly from another thread.

Changed in version 1.3: This was previously just named async; for compatibility with Python 3.7 where async is a keyword it was renamed. On older versions of Python the old name is still around, but it will be removed in the future.

check(ref=True, priority=None)

Create and return a watcher that fires after the event loop polls for IO.

prepare(ref=True, priority=None)

Create and return a watcher that fires before the event loop polls for IO.


This method is not supported by libuv.

stat(path, interval=0.0, ref=True, priority=None)

Create a watcher that monitors the filesystem item at path.

If the operating system doesn’t support event notifications from the filesystem, poll for changes every interval seconds.

fork(ref=True, priority=None)

Create a watcher that fires when the process forks.

Availability: POSIX


Update the loop’s notion of the current time.

New in version 1.3: In the past, this available as update. This is still available as an alias but will be removed in the future.

interface IWatcher[source]

An event loop watcher.

These objects call their callback function when the event loop detects the event has happened.


You must call close() when you are done with this object to avoid leaking native resources.

start(callback, *args, **kwargs)

Have the event loop begin watching for this event.

When the event is detected, callback will be called with args.


Not all watchers accept **kwargs, and some watchers define special meanings for certain keyword args.


Have the event loop stop watching this event.

In the future you may call start() to begin watching again.


Dispose of any native resources associated with the watcher.

If we were active, stop.

Attempting to operate on this object after calling close is undefined. You should dispose of any references you have to it after calling this method.


class Waiter(hub=None)[source]

Bases: object

A low level communication utility for greenlets.

Waiter is a wrapper around greenlet’s switch() and throw() calls that makes them somewhat safer:

  • switching will occur only if the waiting greenlet is executing get() method currently;
  • any error raised in the greenlet is handled inside switch() and throw()
  • if switch()/throw() is called before the receiver calls get(), then Waiter will store the value/exception. The following get() will return the value/raise the exception.

The switch() and throw() methods must only be called from the Hub greenlet. The get() method must be called from a greenlet other than Hub.

>>> result = Waiter()
>>> timer = get_hub().loop.timer(0.1)
>>> timer.start(result.switch, 'hello from Waiter')
>>> result.get() # blocks for 0.1 seconds
'hello from Waiter'
>>> timer.close()

If switch is called before the greenlet gets a chance to call get() then Waiter stores the value.

>>> result = Waiter()
>>> timer = get_hub().loop.timer(0.1)
>>> timer.start(result.switch, 'hi from Waiter')
>>> sleep(0.2)
>>> result.get() # returns immediately without blocking
'hi from Waiter'
>>> timer.close()


This a limited and dangerous way to communicate between greenlets. It can easily leave a greenlet unscheduled forever if used incorrectly. Consider using safer classes such as gevent.event.Event, gevent.event.AsyncResult, or gevent.queue.Queue.


If a value/an exception is stored, return/raise it. Otherwise until switch() or throw() is called.


Return true if and only if it holds a value or an exception


Return true if and only if it is ready and holds a value


Switch to the greenlet if one’s available. Otherwise store the value.

Changed in version 1.3b1: The value is no longer optional.


Switch to the greenlet with the exception. If there’s no greenlet, store the exception.


Holds the exception info passed to throw() if throw() was called. Otherwise None.


class LoopExit[source]

Bases: Exception

Exception thrown when the hub finishes running ( would return).

In a normal application, this is never thrown or caught explicitly. The internal implementation of functions like gevent.hub.Hub.join() and gevent.joinall() may catch it, but user code generally should not.


Errors in application programming can also lead to this exception being raised. Some examples include (but are not limited too):

  • greenlets deadlocking on a lock;
  • using a socket or other gevent object with native thread affinity from a different thread

The following exceptions are not expected to be thrown and are not meant to be caught; if they are raised to user code it is generally a serious programming error or a bug in gevent, greenlet, or its event loop implementation. They are presented here for documentation purposes only.

class ConcurrentObjectUseError[source]

Bases: AssertionError

Raised when an object is used (waited on) by two greenlets independently, meaning the object was entered into a blocking state by one greenlet and then another while still blocking in the first one.

This is usually a programming error.

class BlockingSwitchOutError[source]

Bases: AssertionError

Raised when a gevent synchronous function is called from a low-level event loop callback.

This is usually a programming error.

class InvalidSwitchError[source]

Bases: AssertionError

Raised when the event loop returns control to a greenlet in an unexpected way.

This is usually a bug in gevent, greenlet, or the event loop.

Next page: gevent.core - (deprecated) event loop abstraction