Skip to content

TUTORIAL.md

Latest commit

 

History

History
604 lines (466 loc) · 17.6 KB

TUTORIAL.md

File metadata and controls

604 lines (466 loc) · 17.6 KB

beanstalkm Tutorial

Welcome, dear stranger, to a tour de force through beanstalkd's capabilities. Say hello to your fellow travel companion, the beanstalkm client library for Python. You'll get to know each other fairly well during this trip, so better start off on a friendly note. And now, let's go!

Contents

Getting Started

You'll need beanstalkd listening at port 14711 to follow along. So simply start it using: beanstalkd -l 127.0.0.1 -p 14711

Besides having beanstalkm installed, you'll typically also need PyYAML. If you insist, you can also use beanstalkm without PyYAML. For more details see Appendix A of this tutorial.

To use beanstalkm we have to import the library and set up a connection to an (already running) beanstalkd server:

>>> import beanstalkm
>>> beanstalk = beanstalkm.Client(host='127.0.0.1', port=14711)

If we leave out the host and/or port parameters, '127.0.0.1' and 11300 would be used as defaults, respectively. There is also a timeout parameter which determines how long, in seconds, the socket will wait for the server to respond. If it is None, then there will be no timeout; it defaults to the result of your system's socket.getdefaulttimeout().

API Map

beanstalkm exposes two layers:

  • High-level layer (Client / Message)

    • Client.put(message, tube=..., priority=..., delay=..., ttr=...) sends a message and returns a Message.
    • Client.reserve(timeout=None, drop=True) reserves a message (or returns None on timeout).
      • With drop=True (default), the reserved message is auto-deleted on receipt.
      • With drop=False, you manage the lifecycle yourself (delete/release/bury/...).
    • Message represents a job/message and provides methods like delete(), release(), bury(), kick(), touch(), stats().
    • Message envelope fields supported by this client include: body, created, sender, subscribe, channel, errors, token.

  • Low-level layer (beanstalk.queue.*)

    • beanstalk.queue is a thin command wrapper over the beanstalkd protocol, driven by api.json.
    • This is where tube management and introspection live: use/using, watch/watching/ignore, list_tubes, stats, stats_tube, kick, etc.

Client is a subclass of Connection. If you don't need the Client sugar, you can use Connection directly and access the low-level protocol API via Connection.queue.*:

>>> conn = beanstalkm.Connection(host='127.0.0.1', port=14711)
>>> conn.queue.use('default')
'default'
>>> conn.queue.using()
'default'

Basic Operation

Now that we have a connection set up, we can enqueue jobs:

>>> message = beanstalk.put('hey!')
>>> message.body
'hey!'

Or we can request jobs:

>>> beanstalk.queue.watch(beanstalkm.DEFAULT_TUBE)
>>> message = beanstalk.reserve(drop=False)  # default is drop=True (auto-delete)
>>> message.body
'hey!'

Once we are done with processing a job, we have to mark it as done, otherwise jobs are re-queued by beanstalkd after a "time to run" (120 seconds, per default) is surpassed. A job is marked as done, by calling delete:

>>> message.delete()

reserve blocks until a job is ready, possibly forever. If that is not desired, we can invoke reserve with a timeout (in seconds) how long we want to wait to receive a job. If such a reserve times out, it will return None:

>>> beanstalk.reserve(timeout=0) is None
True

If you use a timeout of 0, reserve will immediately return either a job or None.

Note that beanstalkm sends/receives JSON. The recommended message body is a JSON object (a Python dict).

Internally, beanstalkm can work with a structured "message envelope" (a JSON object containing fields like @context, body, created, sender, etc.; see message.json). That schema file is shipped with the library and is always loaded at import time.

Validation is optional: if you install jsonschema, then incoming envelopes may be validated against message.json. If jsonschema is not installed, validation is skipped and the client will still work, but you lose the extra guarantees.

Important subtlety: the shipped message.json schema requires body to be an object. If you enable validation and send a non-object body (like a string), validation will fail and the client may expose the full envelope dict instead of just the body. If you want stable message.body semantics across setups, prefer using a dict body.

>>> beanstalk.put(object())
Traceback (most recent call last):
...
TypeError: Object of type object is not JSON serializable

There is no restriction on what characters you can put in a job body, so they can be used to hold arbitrary binary data:

>>> _ = beanstalk.put({"data": "\x00\x01\xfe\xff"})
>>> beanstalk.queue.watch(beanstalkm.DEFAULT_TUBE)
>>> message = beanstalk.reserve(drop=False) ; print(repr(message.body["data"])) ; message.delete()
'\x00\x01\xfe\xff'

If you want to send images, put the image data as a string (for example a base64-encoded blob). If you want to send Unicode text, use normal str values (in Python 3, str is Unicode; in Python 2, encode unicode to a byte string).

Tube Management

A single beanstalkd server can provide many different queues, called "tubes" in beanstalkd. To see all available tubes:

>>> beanstalk.queue.list_tubes()  # requires PyYAML
['default', 'receive']

A beanstalkd client can choose one tube into which its job are put. This is the tube "used" by the client. To see what tube you are currently using:

>>> beanstalk.queue.using()
'default'

Unless told otherwise, a client uses the 'default' tube. If you want to use a different tube:

>>> beanstalk.queue.use('foo')
'foo'
>>> beanstalk.queue.using()
'foo'

If you decide to use a tube, that does not yet exist, the tube is automatically created by beanstalkd:

>>> beanstalk.queue.list_tubes()
['default', 'foo', 'receive']

Of course, you can always switch back to the default tube. Tubes that don't have any client using or watching, vanish automatically:

>>> beanstalk.queue.use('default')
'default'
>>> beanstalk.queue.using()
'default'
>>> beanstalk.queue.list_tubes()
['default', 'receive']

Further, a beanstalkd client can choose many tubes to reserve jobs from. These tubes are "watched" by the client. To see what tubes you are currently watching:

>>> beanstalk.queue.watching()
['default']

To watch an additional tube:

>>> beanstalk.queue.watch('bar')
2
>>> beanstalk.queue.watching()
['default', 'bar']

As before, tubes that do not yet exist are created automatically once you start watching them:

>>> beanstalk.queue.list_tubes()
['default', 'bar', 'receive']

To stop watching a tube:

>>> beanstalk.queue.ignore('bar')
1
>>> beanstalk.queue.watching()
['default']

You can't watch zero tubes. So if you try to ignore the last tube you are watching, beanstalkd will respond with NOT_IGNORED and beanstalkm will raise CommandFailed:

>>> from beanstalkm import CommandFailed
>>> try:
...     beanstalk.queue.ignore('default')
... except CommandFailed as e:
...     e.args[1]
'NOT_IGNORED'
>>> beanstalk.queue.watching()
['default']

To recap: each beanstalkd client manages two separate concerns: which tube newly created jobs are put into, and which tube(s) jobs are reserved from. Accordingly, there are two separate sets of functions for these concerns:

  • beanstalk.queue.use and beanstalk.queue.using affect where jobs are put when you use the low-level command API;
  • beanstalk.queue.watch and beanstalk.queue.watching control where jobs are reserved from.

At the higher-level API, you typically pass tube=... to Client.put(...).

Note that these concerns are fully orthogonal: for example, when you use a tube, it is not automatically watched. Neither does watching a tube affect the tube you are using.

Statistics

Beanstalkd accumulates various statistics at the server, tube and job level. Statistical details for a job can only be retrieved during the job's lifecycle. So let's create another job:

>>> message = beanstalk.put('ho?')
>>> message.uid > 0
True
>>> beanstalk.queue.watch(beanstalkm.DEFAULT_TUBE)
>>> message = beanstalk.reserve(drop=False)  # default is drop=True (auto-delete)

Now we retrieve job-level statistics:

>>> from pprint import pprint
>>> pprint(message.stats())                         # doctest: +ELLIPSIS
{'age': 0,
 ...
 'id': ...,
 ...
 'state': 'reserved',
 ...
 'tube': 'receive'}

After deleting a message, Message.stats() returns None because the message no longer has an ID:

>>> message.delete()
>>> message.stats() is None
True

Let's have a look at some numbers for the 'default' tube:

>>> pprint(beanstalk.queue.stats_tube('default'))     # doctest: +ELLIPSIS
{...
 'current-jobs-ready': 0,
 'current-jobs-reserved': 0,
 'current-jobs-urgent': 0,
 ...
 'name': 'default',
 ...}

Finally, there's an abundant amount of server-level statistics accessible via the Connection's stats method. We won't go into details here, but:

>>> pprint(beanstalk.queue.stats())                   # doctest: +ELLIPSIS
{...
 'current-connections': 1,
 'current-jobs-buried': 0,
 'current-jobs-delayed': 0,
 'current-jobs-ready': 0,
 'current-jobs-reserved': 0,
 'current-jobs-urgent': 0,
 ...}

Advanced Operation

In "Basic Operation" above, we discussed the typical lifecycle of a job:

 put            reserve               delete
-----> [READY] ---------> [RESERVED] --------> *poof*

(This picture was taken from beanstalkd's protocol documentation. It is
originally contained in `protocol.txt`, part of the beanstalkd
distribution.) #doctest:+SKIP

But besides ready and reserved, a job can also be delayed or buried. Along with those states come a few transitions, so the full picture looks like the following:

   put with delay               release with delay
  ----------------> [DELAYED] <------------.
                        |                   |
                        | (time passes)     |
                        |                   |
   put                  v     reserve       |       delete
  -----------------> [READY] ---------> [RESERVED] --------> *poof*
                       ^  ^                |  |
                       |   \  release      |  |
                       |    `-------------'   |
                       |                      |
                       | kick                 |
                       |                      |
                       |       bury           |
                    [BURIED] <---------------'
                       |
                       |  delete
                        `--------> *poof*

  (This picture was taken from beanstalkd's protocol documentation. It is
  originally contained in `protocol.txt`, part of the beanstalkd
  distribution.) #doctest:+SKIP

Now let's have a practical look at those new possibilities. For a start, we can create a job with a delay. Such a job will only be available for reservation once this delay passes:

>>> message = beanstalk.put('yes!', delay=1)
>>> message.uid > 0
True
>>> beanstalk.reserve(timeout=0) is None
True
>>> beanstalk.queue.watch(beanstalkm.DEFAULT_TUBE)
>>> message = beanstalk.reserve(timeout=1, drop=False)
>>> message.body
'yes!'

If we are not interested in a job anymore (e.g. after we failed to process it), we can simply release the job back to the tube it came from:

>>> message.release()
>>> message.stats()['state']
'ready'

Want to get rid of a job? Well, just "bury" it! A buried job is put aside and is therefore not available for reservation anymore:

>>> message = beanstalk.reserve(drop=False)
>>> message.bury()
>>> message.stats()['state']
'buried'
>>> beanstalk.reserve(timeout=0) is None
True

Buried jobs are maintained in a special FIFO-queue outside of the normal job processing lifecycle until they are kicked alive again:

>>> beanstalk.queue.kick(1)
1

You can request many jobs to be kicked alive at once, kick's return value will tell you how many jobs were actually kicked alive again:

>>> beanstalk.queue.kick(42)
0

If you still have a handle to a job (or know its job ID), you can also kick alive this particular job, overriding the FIFO operation of the burial queue:

>>> message = beanstalk.reserve(drop=False)
>>> message.bury()
>>> message.stats()['state']
'buried'
>>> message.kick()
>>> message.stats()['state']
'ready'

(NOTE: The kick-job command was introduced in beanstalkd v1.8.)

Inspecting Jobs

beanstalkd supports "peek" commands to inspect jobs without reserving them. beanstalkm exposes these commands at the low-level API as beanstalk.queue.peek(...), beanstalk.queue.peek_ready(), etc.

However, in the current beanstalkm implementation the peek* commands as defined in api.json are not wired up in a usable way and will raise UnexpectedResponse (the protocol status is not recognized by the client).

The examples below show the intended semantics from the beanstalkd protocol documentation, but they will currently fail when executed with this version of beanstalkm.

If you know the id of a job you're interested in, you can directly peek at the job:

>>> job = beanstalk.queue.peek(4)  # currently raises UnexpectedResponse
>>> job.body
...

If you are not interested in a particular job, but want to see jobs according to their state, beanstalkd also provides a few commands. To peek at the same job that would be returned by reserve -- the next ready job -- use peek-ready:

>>> job = beanstalk.queue.peek_ready()  # currently raises UnexpectedResponse
>>> job.body
...

You can also peek into the special queues for jobs that are delayed:

>>> job = beanstalk.queue.peek_delayed()  # currently raises UnexpectedResponse
>>> job.stats()['state']
'delayed'

... or buried:

>>> job = beanstalk.queue.peek_buried()  # currently raises UnexpectedResponse
>>> job.stats()['state']
'buried'

If you need to inspect a job, the supported workaround is to reserve it with drop=False, inspect its body, and then release() it back to the queue:

>>> beanstalk.queue.watch(beanstalkm.DEFAULT_TUBE)
>>> message = beanstalk.reserve(timeout=0, drop=False)
>>> if message is not None:
...     _ = message.body
...     message.release()

Job Priorities

Without job priorities, beanstalkd operates as a FIFO queue:

>>> _ = beanstalk.put({"v": "1"})
>>> _ = beanstalk.put({"v": "2"})
>>> beanstalk.queue.watch(beanstalkm.DEFAULT_TUBE)
>>> message = beanstalk.reserve(drop=False) ; print(message.body["v"]) ; message.delete()
1
>>> message = beanstalk.reserve(drop=False) ; print(message.body["v"]) ; message.delete()
2

If need arises, you can override this behaviour by giving different jobs different priorities. There are three hard facts to know about job priorities:

  1. Jobs with lower priority numbers are reserved before jobs with higher priority numbers.

  2. beanstalkd priorities are 32-bit unsigned integers (they range from 0 to 2**32 - 1).

  3. beanstalkm uses 2**31 as default job priority (beanstalkm.DEFAULT_PRIORITY).

To create a job with a custom priority, use the keyword-argument priority:

>>> _ = beanstalk.put({"v": "foo"}, priority=42)
>>> _ = beanstalk.put({"v": "bar"}, priority=21)
>>> _ = beanstalk.put({"v": "qux"}, priority=21)
>>> message = beanstalk.reserve(drop=False) ; print(message.body["v"]) ; message.delete()
bar
>>> message = beanstalk.reserve(drop=False) ; print(message.body["v"]) ; message.delete()
qux
>>> message = beanstalk.reserve(drop=False) ; print(message.body["v"]) ; message.delete()
foo

Note how 'bar' and 'qux' left the queue before 'foo', even though they were enqueued well after 'foo'. Note also that within the same priority jobs are still handled in a FIFO manner.

Fin!

>>> beanstalk.close()

That's it, for now. We've left a few capabilities untouched (touch and time-to-run). But if you've really read through all of the above, send me a message and tell me what you think of it. And then go get yourself a treat. You certainly deserve it.

Appendix A: beanstalkm and YAML

As beanstalkd uses YAML for diagnostic information (like the results of beanstalk.queue.stats() or beanstalk.queue.list_tubes()), you'll typically need PyYAML. Depending on your performance needs, you may want to supplement that with the libyaml C extension.

beanstalkm will automatically use PyYAML if it is installed. If PyYAML is not available, beanstalkm will log an error and YAML-returning commands will not be parsed.

Important: there is no supported "raw YAML" mode. Without PyYAML, YAML response bodies are not consumed by the client, which can desynchronize the connection. So in practice PyYAML is required if you intend to call any YAML-returning methods.

The current public API does not expose a supported "raw YAML" mode, so in practice PyYAML is required for these methods to be useful:

  • beanstalk.queue.list_tubes()
  • beanstalk.queue.watching()
  • beanstalk.queue.stats()
  • beanstalk.queue.stats_tube(name)
  • Message.stats()