Fivebeans: различия между версиями

Материал из support.qbpro.ru
imported>Supportadmin
imported>Supportadmin
Нет описания правки
 
(не показано 25 промежуточных версий этого же участника)
Строка 1: Строка 1:
beanstalkd client & worker daemon for node
beanstalkd client & worker daemon for node


[https://www.npmjs.org/package/fivebeans оригинал]
[https://www.npmjs.org/package/fivebeans оригинал][https://github.com/ceejbot/fivebeans GitHub]
 
==FiveBeansClient==
 
Heavily inspired by node-beanstalk-client, which is a perfectly usable client but somewhat dusty. I wanted more complete support of the beanstalkd protocol in a project written in plain javascript.
 
All client method names are the same case & spelling as the beanstalk text command, with hyphens replaced by underscore. The single exception is delete, which is renamed to destroy().
 
For complete details on the beanstalkd commands, see its protocol documentation.


===Creating a client===
===Creating a client===
Строка 14: Строка 22:
The client is not usable until you call its connect() method. Here's an example of setting up a client:
The client is not usable until you call its connect() method. Here's an example of setting up a client:


<nowiki>var fivebeans = require('fivebeans');
<code>
 
var fivebeans = require('fivebeans');
var client = new fivebeans.client('10.0.1.1', 11300);
var client = new fivebeans.client('10.0.1.1', 11300);
client
client
     .on('connect', function()
     .on('connect', function()
     {
     {
Строка 30: Строка 38:
         // underlying connection has closed
         // underlying connection has closed
     })
     })
     .connect();</nowiki>
     .connect();
</code>


===Producing jobs===
===Producing jobs===
====use====
<code>client.use(tube, function(err, tubename) {});</code>
Use the specified tube. Reponds with the name of the tube being used.
====list_tube_used====
<code>client.list_tube_used(function(err, tubename) {});</code>
Responds with the name of the tube currently being used by the client.
====put====
<code>client.put(priority, delay, ttr, payload, function(err, jobid) {});</code>
Submit a job with the specified priority (smaller integers are higher priority), delay in seconds, and allowed time-to-run in seconds. The payload contains the job data the server will return to clients reserving jobs; it can be either a Buffer object or a string. No processing is done on the data. Responds with the id of the newly-created job.
====peek_ready====
<code>client.peek_ready(function(err, jobid, payload) {});</code>
Peek at the data for the job at the top of the ready queue of the tube currently in use. Responds with the job id and payload of the next job, or 'NOT_FOUND' if there are no qualifying jobs in the tube. The payload is a Buffer object.
====peek_delayed====
<code>client.peek_delayed(function(err, jobid, payload) {});</code>
Peek at the data for the delayed job with the shortest delay in the tube currently in use. Responds with the job id and payload of the next job, or 'NOT_FOUND' in err if there are no qualifying jobs in the tube. The payload is a Buffer object.
====peek_buried====
<code>client.peek_buried(function(err, jobid, payload) {});</code>
Peek at the data for the next buried job in the tube currently in use. Responds with the job id and payload of the next job, or 'NOT_FOUND' in err if there are no qualifying jobs in the tube. The payload is a Buffer object.
===Consuming jobs===
===Consuming jobs===
====watch====
<code>client.watch(tube, function(err, numwatched) {});</code>
Watch the named tube. Responds with the number of tubes currently watched by the client.
====ignore====
<code>client.ignore(tube, function(err, numwatched) {});</code>
Ignore the named tube. Responds with the number of tubes currently watched by the client.
====list_tubes_watched====
<code>client.list_tubes_watched(function(err, tubelist) {});</code>
Responds with an array containing the names of the tubes currently watched by the client.
====reserve====
<code>client.reserve(function(err, jobid, payload) {});</code>
Reserve a job. Responds with the id and the job data. The payload is a Buffer object.
====reserve_with_timeout====
<code>client.reserve_with_timeout(seconds, function(err, jobid, payload) {});</code>
Reserve a job, waiting the specified number of seconds before timing out. err contains the string "TIMED_OUT" if the specified time elapsed before a job became available. Payload is a buffer.
====touch====
<code>client.touch(jobid, function(err) {});</code>
Inform the server that the client is still processing a job, thus requesting more time to work on it.
====destroy====
<code>client.destroy(jobid, function(err) {});</code>
Delete the specified job. Responds with null if successful, a string error otherwise. This is the only method not named identically to its beanstalkd counterpart, because delete is a reserved word in Javascript.
====release====
<code>client.release(jobid, priority, delay, function(err) {});</code>
Release the specified job and assign it the given priority and delay (in seconds). Responds with null if successful, a string error otherwise.
====bury====
<code>client.bury(jobid, priority, function(err) {});</code>
Bury the specified job and assign it the given priority. Responds with null if successful, a string error otherwise.
====kick====
<code>client.kick(maxToKick, function(err, numkicked) {});</code>
Kick at most maxToKick delayed and buried jobs back into the active queue. Responds with the number of jobs kicked.
====kick_job====
<code>client.kick_job(jobID, function(err) {});</code>
Kick the specified job id. Responds with NOT_FOUND if the job was not found. Supported in beanstalkd versions >= 1.6.
===Server statistics===
===Server statistics===
====peek====
<code>client.peek(id, function(err, jobid, payload) {});</code>
Peek at the data for the specified job. Payload is a Buffer object.
====pause_tube====
<code>client.pause_tube(tubename, delay, function(err) {});</code>
Pause the named tube for the given number of seconds. No new jobs may be reserved from the tube while it is paused.
====list_tubes====
<code>client.list_tubes(function(err, tubenames) {});</code>
List all the existing tubes. Responds with an array of tube names.
====stats_job====
<code>client.stats_job(jobid, function(err, response) {});</code>
Request statistics for the specified job. Responds with a hash containing information about the job. See the beanstalkd documentation for a complete list of stats.
====stats_tube====
<code>client.stats_tube(tubename, function(err, response) {});</code>
Request statistics for the specified tube. Responds with a hash containing information about the tube. See the beanstalkd documentation for a complete list of stats.
====stats====
<code>client.stats(function(err, response) {});</code>
Request statistics for the beanstalkd server. Responds with a hash containing information about the server. See the beanstalkd documentation for a complete list of stats.
==FiveBeansWorker==
==FiveBeansWorker==
Inspired by node-beanstalk-worker but updated & rewritten to work with jobs queued by Stalker.
The worker pulls jobs off the queue & passes them to matching handlers. It deletes successful jobs & buries unsuccessful ones. It continues processing past all recoverable errors, though it emits events on error.
===API===
===API===
====constructor====
====constructor====
<code>new FiveBeansWorker(options)</code>
Returns a new worker object. options is a hash containing the following keys:
*'''id''': how this worker should identify itself in log events
*'''host''': beanstalkd host
*'''port''': beanstalkd port
*'''handlers''': hash with handler objects, with handler types as keys
*'''ignoreDefault''': true if this worker should ignore the default tube timeout: timeout parameter used with on reserve_with_timeout, defaults to 10 (in seconds)
====start====
<code>start(tubelist)</code>
Connect the worker to the beanstalkd server & make it watch the specified tubes. Emits the 'started' event when it is complete.
====stop====
<code>stop()</code>
Finish processing the current job then close the client. Emits the 'stopped' event when complete.
====watch====
<code>watch(tubelist, callback)</code>
Begin watching the tubes named in the list.
====ignore====
<code>ignore(tubelist, callback)</code>
Ignore the tubes named in the list.
===Events===
===Events===
The worker is intended to continue processing jobs through most errors. Its response to exceptions encountered when processing jobs is to bury the job and emit an event that can be logged or handled somewhere else.
*'''error''': Emitted on error in the underlying client. Payload is the error object. Execution is halted. You must listen for this event.
*'''close''': Emitted on close in the underlying client. No payload.
*'''started''': Worker has started processing. No payload.
*'''stopped''': Worker has stopped processing. No payload.
*'''info''': The worker has taken some action that you might want to log. The payload is an object with information about the action, with two fields:
<code>{
    clientid: 'id-of-worker',
    message: 'a logging-style description of the action'
}</code>
This event is the tattered remnants of what used to be built-in logging, and it might go away.
*'''warning''': The worker has encountered an error condition that will not stop processing, but that you might wish to act upon or log. The payload is an object with information about the error. Fields guaranteed to be present are:
<code>{
    clientid: 'id-of-worker',
    message: 'the context of the error',
    error: errorObject
}</code>
Some errors might have additional fields providing context, such as a job id.
'''job.reserved''': The worker has reserved a job. The payload is the job id.
'''job.handled''': The worker has completed processing a job. The payload is an object with information about the job.
<code>{
    id: job id,
    type: job type,
    elapsed: elapsed time in ms,
    action: [ 'success' | 'release' | 'bury' | custom error message ]
}</code>
'''job.deleted''': The worker has deleted a job. The payload is the job id.
'''job.buried''': The worker has buried a job. The payload is the job id.
===Jobs===
===Jobs===
Each job must be a JSON-serialized object with two fields:
*'''type''': type string matching a handler
*'''payload''': job data, in whatever format the job defines
The worker looks up a handler using the given type string and calls work() on the job payload.
The job may also be a JSON array containing two items:
<code>[ tubename, jobdata ]</code>
Where the second item is an object as specified above. This is for compatibility with the Stalker library, which wraps the job data this way.
===Handlers===
===Handlers===
Handler modules must export a single function that returns an object. The object must have a field called 'type' with a brief descriptive string. It must also expose a function called work() with this signature:
<code>work(jobdata, callback(action, delay))</code>
*'''jobdata''': job payload
*'''action''': 'success' | 'release' | 'bury' | custom error message
*'''delay''': seconds to delay if the job is released; otherwise unused
If the action is "success", the job is deleted. If it is "release", the job is released with the specified delay. If it is "bury", the job is buried. All other actions are treated as errors & the job is buried in response.
Here's a simple handler example.
<code>module.exports = function()
{
    function EmitKeysHandler()
    {
        this.type = 'emitkeys';
    }
    EmitKeysHandler.prototype.work = function(payload, callback)
    {
        var keys = Object.keys(payload);
        for (var i = 0; i < keys.length; i++)
            console.log(keys[i]);
        callback('success');
    }
    var handler = new EmitKeysHandler();
    return handler;
};</code>
The examples directory has another sample handler.
===Example===
===Example===
This example starts a worker capable of handling the emitkeys example from above.
<code>var Beanworker = require('fivebeans').worker;
var options =
{
    id: 'worker_4',
    host: '127.0.0.1',
    port: 11300,
    handlers:
    {
        emitkeys: require('./emitkeyshandler')()
    },
    ignoreDefault: true
}
var worker = new Beanworker(options);
worker.start(['high', 'medium', 'low']);</code>
==FiveBeansRunner==
==FiveBeansRunner==
A wrapper that runs a single beanstalkd worker as a daemon. Responds to the USR2 signal by reloading the configuration and restarting the worker. Handles SIGINT, SIGHUP, and SIGQUIT by completing processing on the current job then stopping.
Example use:
<nowiki>var fivebeans = require('fivebeans');
var runner = new fivebeans.runner('worker_id_1', '/path/to/config.yml');
runner.go();</nowiki>
===bin/beanworker===
===bin/beanworker===
The above code plus yargs wrapped in a node shell script for your convenience.
bin/beanworker --id=[ID] --config=[config.yml]
Creates a runner for a worker with the specified ID & configured with the specified yaml file.
Here's the complete source:
<nowiki>#!/usr/bin/env node
var argv = require('yargs')
    .usage('Usage: beanworker --id=[ID] --config=[config.yml]')
    .default('id', 'defaultID')
    .demand(['config'])
    .argv;
var FiveBeans = require('fivebeans');
var runner = new FiveBeans.runner(argv.id, argv.config);
runner.go();</nowiki>
===Configuration file===
===Configuration file===
Here's an example yaml configuration:
<nowiki>beanstalkd:
    host: "127.0.0.1"
    port: 11300
watch:
    - 'circle'
    - 'picadilly'
    - 'northern'
    - 'central'
handlers:
    - "./handlers/holborn.js"
    - "./handlers/greenpark.js"
    - "./handlers/knightsbridge.js"
ignoreDefault: true
beanstalkd: where to connect
watch: a list of tubes to watch.
handlers: a list of handler files to require
ignoreDefault: true if this worker should ignore the default tube
</nowiki>
If the handler paths don't start with / the current working directory will be prepended to them before they are required.
Why yaml not json? Because when I originally wrote this, it was in support of a ruby service, and yaml is the native config format over in that land. I continue using it because it's more readable than json and easier for humans to type.

Текущая версия от 20:59, 11 мая 2014

beanstalkd client & worker daemon for node

оригиналGitHub

FiveBeansClient

Heavily inspired by node-beanstalk-client, which is a perfectly usable client but somewhat dusty. I wanted more complete support of the beanstalkd protocol in a project written in plain javascript.

All client method names are the same case & spelling as the beanstalk text command, with hyphens replaced by underscore. The single exception is delete, which is renamed to destroy().

For complete details on the beanstalkd commands, see its protocol documentation.

Creating a client

The client constructor takes two arguments:

  • host: The address of the beanstalkd server. Defaults to 127.0.0.1.
  • port: Port to connect to. Defaults to 11300.

The client emits three events that you should listen for: connect, error, and close.

The client is not usable until you call its connect() method. Here's an example of setting up a client:

var fivebeans = require('fivebeans');
var client = new fivebeans.client('10.0.1.1', 11300);
client
   .on('connect', function()
   {
       // client can now be used
   })
   .on('error', function(err)
   {
       // connection failure    
   })
   .on('close', function()
   {
       // underlying connection has closed
   })
   .connect();

Producing jobs

use

client.use(tube, function(err, tubename) {});

Use the specified tube. Reponds with the name of the tube being used.

list_tube_used

client.list_tube_used(function(err, tubename) {});

Responds with the name of the tube currently being used by the client.

put

client.put(priority, delay, ttr, payload, function(err, jobid) {});

Submit a job with the specified priority (smaller integers are higher priority), delay in seconds, and allowed time-to-run in seconds. The payload contains the job data the server will return to clients reserving jobs; it can be either a Buffer object or a string. No processing is done on the data. Responds with the id of the newly-created job.

peek_ready

client.peek_ready(function(err, jobid, payload) {});

Peek at the data for the job at the top of the ready queue of the tube currently in use. Responds with the job id and payload of the next job, or 'NOT_FOUND' if there are no qualifying jobs in the tube. The payload is a Buffer object.

peek_delayed

client.peek_delayed(function(err, jobid, payload) {});

Peek at the data for the delayed job with the shortest delay in the tube currently in use. Responds with the job id and payload of the next job, or 'NOT_FOUND' in err if there are no qualifying jobs in the tube. The payload is a Buffer object.

peek_buried

client.peek_buried(function(err, jobid, payload) {});

Peek at the data for the next buried job in the tube currently in use. Responds with the job id and payload of the next job, or 'NOT_FOUND' in err if there are no qualifying jobs in the tube. The payload is a Buffer object.

Consuming jobs

watch

client.watch(tube, function(err, numwatched) {});

Watch the named tube. Responds with the number of tubes currently watched by the client.

ignore

client.ignore(tube, function(err, numwatched) {});

Ignore the named tube. Responds with the number of tubes currently watched by the client.

list_tubes_watched

client.list_tubes_watched(function(err, tubelist) {});

Responds with an array containing the names of the tubes currently watched by the client.

reserve

client.reserve(function(err, jobid, payload) {});

Reserve a job. Responds with the id and the job data. The payload is a Buffer object.

reserve_with_timeout

client.reserve_with_timeout(seconds, function(err, jobid, payload) {});

Reserve a job, waiting the specified number of seconds before timing out. err contains the string "TIMED_OUT" if the specified time elapsed before a job became available. Payload is a buffer.

touch

client.touch(jobid, function(err) {});

Inform the server that the client is still processing a job, thus requesting more time to work on it.

destroy

client.destroy(jobid, function(err) {});

Delete the specified job. Responds with null if successful, a string error otherwise. This is the only method not named identically to its beanstalkd counterpart, because delete is a reserved word in Javascript.

release

client.release(jobid, priority, delay, function(err) {});

Release the specified job and assign it the given priority and delay (in seconds). Responds with null if successful, a string error otherwise.

bury

client.bury(jobid, priority, function(err) {});

Bury the specified job and assign it the given priority. Responds with null if successful, a string error otherwise.

kick

client.kick(maxToKick, function(err, numkicked) {});

Kick at most maxToKick delayed and buried jobs back into the active queue. Responds with the number of jobs kicked.

kick_job

client.kick_job(jobID, function(err) {});

Kick the specified job id. Responds with NOT_FOUND if the job was not found. Supported in beanstalkd versions >= 1.6.

Server statistics

peek

client.peek(id, function(err, jobid, payload) {});

Peek at the data for the specified job. Payload is a Buffer object.

pause_tube

client.pause_tube(tubename, delay, function(err) {});

Pause the named tube for the given number of seconds. No new jobs may be reserved from the tube while it is paused.

list_tubes

client.list_tubes(function(err, tubenames) {});

List all the existing tubes. Responds with an array of tube names.

stats_job

client.stats_job(jobid, function(err, response) {});

Request statistics for the specified job. Responds with a hash containing information about the job. See the beanstalkd documentation for a complete list of stats.

stats_tube

client.stats_tube(tubename, function(err, response) {});

Request statistics for the specified tube. Responds with a hash containing information about the tube. See the beanstalkd documentation for a complete list of stats.

stats

client.stats(function(err, response) {});

Request statistics for the beanstalkd server. Responds with a hash containing information about the server. See the beanstalkd documentation for a complete list of stats.

FiveBeansWorker

Inspired by node-beanstalk-worker but updated & rewritten to work with jobs queued by Stalker.

The worker pulls jobs off the queue & passes them to matching handlers. It deletes successful jobs & buries unsuccessful ones. It continues processing past all recoverable errors, though it emits events on error.

API

constructor

new FiveBeansWorker(options)

Returns a new worker object. options is a hash containing the following keys:

  • id: how this worker should identify itself in log events
  • host: beanstalkd host
  • port: beanstalkd port
  • handlers: hash with handler objects, with handler types as keys
  • ignoreDefault: true if this worker should ignore the default tube timeout: timeout parameter used with on reserve_with_timeout, defaults to 10 (in seconds)

start

start(tubelist)

Connect the worker to the beanstalkd server & make it watch the specified tubes. Emits the 'started' event when it is complete.

stop

stop()

Finish processing the current job then close the client. Emits the 'stopped' event when complete.

watch

watch(tubelist, callback)

Begin watching the tubes named in the list.

ignore

ignore(tubelist, callback)

Ignore the tubes named in the list.

Events

The worker is intended to continue processing jobs through most errors. Its response to exceptions encountered when processing jobs is to bury the job and emit an event that can be logged or handled somewhere else.

  • error: Emitted on error in the underlying client. Payload is the error object. Execution is halted. You must listen for this event.
  • close: Emitted on close in the underlying client. No payload.
  • started: Worker has started processing. No payload.
  • stopped: Worker has stopped processing. No payload.
  • info: The worker has taken some action that you might want to log. The payload is an object with information about the action, with two fields:
{
   clientid: 'id-of-worker',
   message: 'a logging-style description of the action'
}

This event is the tattered remnants of what used to be built-in logging, and it might go away.

  • warning: The worker has encountered an error condition that will not stop processing, but that you might wish to act upon or log. The payload is an object with information about the error. Fields guaranteed to be present are:
{
   clientid: 'id-of-worker',
   message: 'the context of the error',
   error: errorObject
}

Some errors might have additional fields providing context, such as a job id.

job.reserved: The worker has reserved a job. The payload is the job id.

job.handled: The worker has completed processing a job. The payload is an object with information about the job.

{
   id: job id,
   type: job type,
   elapsed: elapsed time in ms,
   action: [ 'success' | 'release' | 'bury' | custom error message ]
}

job.deleted: The worker has deleted a job. The payload is the job id.

job.buried: The worker has buried a job. The payload is the job id.

Jobs

Each job must be a JSON-serialized object with two fields:

  • type: type string matching a handler
  • payload: job data, in whatever format the job defines

The worker looks up a handler using the given type string and calls work() on the job payload.

The job may also be a JSON array containing two items:

[ tubename, jobdata ]

Where the second item is an object as specified above. This is for compatibility with the Stalker library, which wraps the job data this way.

Handlers

Handler modules must export a single function that returns an object. The object must have a field called 'type' with a brief descriptive string. It must also expose a function called work() with this signature:

work(jobdata, callback(action, delay))

  • jobdata: job payload
  • action: 'success' | 'release' | 'bury' | custom error message
  • delay: seconds to delay if the job is released; otherwise unused

If the action is "success", the job is deleted. If it is "release", the job is released with the specified delay. If it is "bury", the job is buried. All other actions are treated as errors & the job is buried in response.

Here's a simple handler example.

module.exports = function()
{
   function EmitKeysHandler()
   {
       this.type = 'emitkeys';
   }
   EmitKeysHandler.prototype.work = function(payload, callback)
   {
       var keys = Object.keys(payload);
       for (var i = 0; i < keys.length; i++)
           console.log(keys[i]);
       callback('success');
   }
   var handler = new EmitKeysHandler();
   return handler;
};

The examples directory has another sample handler.

Example

This example starts a worker capable of handling the emitkeys example from above.

var Beanworker = require('fivebeans').worker;
var options =
{
   id: 'worker_4',
   host: '127.0.0.1',
   port: 11300,
   handlers:
   {
       emitkeys: require('./emitkeyshandler')()
   },
   ignoreDefault: true
}
var worker = new Beanworker(options);
worker.start(['high', 'medium', 'low']);

FiveBeansRunner

A wrapper that runs a single beanstalkd worker as a daemon. Responds to the USR2 signal by reloading the configuration and restarting the worker. Handles SIGINT, SIGHUP, and SIGQUIT by completing processing on the current job then stopping.

Example use:

var fivebeans = require('fivebeans');
var runner = new fivebeans.runner('worker_id_1', '/path/to/config.yml');
runner.go();

bin/beanworker

The above code plus yargs wrapped in a node shell script for your convenience.

bin/beanworker --id=[ID] --config=[config.yml]

Creates a runner for a worker with the specified ID & configured with the specified yaml file.

Here's the complete source:

#!/usr/bin/env node

var argv = require('yargs')
    .usage('Usage: beanworker --id=[ID] --config=[config.yml]')
    .default('id', 'defaultID')
    .demand(['config'])
    .argv;

var FiveBeans = require('fivebeans');

var runner = new FiveBeans.runner(argv.id, argv.config);
runner.go();

Configuration file

Here's an example yaml configuration:

beanstalkd:
    host: "127.0.0.1"
    port: 11300
watch:
    - 'circle'
    - 'picadilly'
    - 'northern'
    - 'central'
handlers:
    - "./handlers/holborn.js"
    - "./handlers/greenpark.js"
    - "./handlers/knightsbridge.js"
ignoreDefault: true
beanstalkd: where to connect
watch: a list of tubes to watch.
handlers: a list of handler files to require
ignoreDefault: true if this worker should ignore the default tube

If the handler paths don't start with / the current working directory will be prepended to them before they are required.

Why yaml not json? Because when I originally wrote this, it was in support of a ruby service, and yaml is the native config format over in that land. I continue using it because it's more readable than json and easier for humans to type.