Using Eliot with Twisted¶
Eliot provides a variety of APIs to support integration with the Twisted networking framework.
eliot.logwriter.ThreadedWriter is a logging destination that wraps a blocking destination and writes to it in a non-reactor thread.
This is useful because it keeps the Twisted reactor from blocking, e.g. if you’re writing to a log file and the hard drive is overloaded.
ThreadedWriter is a Twisted
Service and starting it will call
add_destinations for you and stopping it will call
remove_destination; there is no need to call those directly.
""" Output an Eliot message to a log file using the threaded log writer. """ from __future__ import unicode_literals, print_function from twisted.internet.task import react from eliot.logwriter import ThreadedWriter from eliot import log_message, FileDestination def main(reactor): print("Logging to example-eliot.log...") logWriter = ThreadedWriter( FileDestination(file=open("example-eliot.log", "ab")), reactor) # Manually start the service, which will add it as a # destination. Normally we'd register ThreadedWriter with the usual # Twisted Service/Application infrastructure. logWriter.startService() # Log a message: log_message(message_type="test", value="hello", another=1) # Manually stop the service. done = logWriter.stopService() return done if __name__ == '__main__': react(main, )
If you want log rotation you can pass in an
eliot.FileDestination wrapping one of the classes from twisted.python.logfile as the destination file.
If you wish you can direct Eliot logs to Twisted’s logging subsystem, if that is the primary logging system you’re using.
from eliot import add_destinations from eliot.twisted import TwistedDestination add_destinations(TwistedDestination())
If you’re using Twisted’s
trial program to run your tests you can redirect your Eliot logs to Twisted’s logs by calling
This function will automatically detect whether or not it is running under
If it is then you will be able to read your Eliot logs in
trial writes out logs by default.
If it is not running under
trial it will not do anything.
In addition calling it multiple times has the same effect as calling it once.
The way you use it is by putting it in your package’s
__init__.py: it will do the right thing and only redirect if you’re using
Take care if you are separately redirecting Twisted logs to Eliot; you should make sure not to call
redirectLogsForTrial in that case so as to prevent infinite loops.
eliot.writeFailure is the equivalent of
eliot.write_traceback, only for
from eliot import writeFailure class YourClass(object): def run(self): d = dosomething() d.addErrback(writeFailure)
Actions and inlineCallbacks¶
Eliot provides a decorator that is compatible with Twisted’s
inlineCallbacks but which also behaves well with Eliot’s actions.
twisted.internet.defer.inlineCallbacks in your code.
To understand why, consider the following example:
from eliot import start_action from twisted.internet.defer import inlineCallbacks @inlineCallbacks # don't use this in real code, use eliot.twisted.inline_callbacks def go(): with start_action(action_type=u"yourapp:subsystem:frob") as action: d = some_deferred_api() x = yield d action.log(message_type=u"some-report", x=x)
The action started by this generator remains active as
yield d gives up control to the
The next bit of code to run will be considered to be a child of
Since that code may be any arbitrary code that happens to get scheduled,
this is certainly wrong.
inlineCallbacks controller resumes the generator,
it will most likely do so with no active action at all.
This means that the log message following the yield will be recorded with no parent action,
also certainly wrong.
These problems are solved by using
eliot.twisted.inline_callbacks instead of
The behavior of the two decorators is identical except that Eliot’s version will preserve the generator’s action context and contain it within the generator.
This extends the
inlineCallbacks illusion of “synchronous” code to Eliot actions.
Actions and Deferreds¶
An additional set of APIs is available to help log actions when using Deferreds. To understand why, consider the following example:
from eliot import start_action def go(): action = start_action(action_type=u"yourapp:subsystem:frob") with action: d = Deferred() d.addCallback(gotResult, x=1) return d
This has two problems.
gotResult is not going to run in the context of the action.
Second, the action finishes once the
with block finishes, i.e. before
If we want
gotResult to be run in the context of the action and to delay the action finish we need to do some extra work, and manually wrapping all callbacks would be tedious.
To solve this problem you can use the
It grabs the action context when it is first created and provides the same API as
addCallbacks and friends), with the difference that added callbacks run in the context of the action.
When all callbacks have been added you can indicate that the action should finish after those callbacks have run by calling
As you would expect, if the
Deferred fires with a regular result that will result in success message.
Deferred fires with an errback that will result in failure message.
Finally, you can unwrap the
DeferredContext and access the wrapped
Deferred by accessing its
from eliot import start_action from eliot.twisted import DeferredContext def go(): with start_action(action_type=u"your_type").context() as action: d = DeferredContext(Deferred()) # gotResult(result, x=1) will be called in the context of the action: d.addCallback(gotResult, x=1) # After gotResult finishes, finish the action: d.addActionFinish() # Return the underlying Deferred: return d.result