qless 0.9.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- data/Gemfile +8 -0
- data/HISTORY.md +168 -0
- data/README.md +571 -0
- data/Rakefile +28 -0
- data/bin/qless-campfire +106 -0
- data/bin/qless-growl +99 -0
- data/bin/qless-web +23 -0
- data/lib/qless.rb +185 -0
- data/lib/qless/config.rb +31 -0
- data/lib/qless/job.rb +259 -0
- data/lib/qless/job_reservers/ordered.rb +23 -0
- data/lib/qless/job_reservers/round_robin.rb +34 -0
- data/lib/qless/lua.rb +25 -0
- data/lib/qless/qless-core/cancel.lua +71 -0
- data/lib/qless/qless-core/complete.lua +218 -0
- data/lib/qless/qless-core/config.lua +44 -0
- data/lib/qless/qless-core/depends.lua +65 -0
- data/lib/qless/qless-core/fail.lua +107 -0
- data/lib/qless/qless-core/failed.lua +83 -0
- data/lib/qless/qless-core/get.lua +37 -0
- data/lib/qless/qless-core/heartbeat.lua +50 -0
- data/lib/qless/qless-core/jobs.lua +41 -0
- data/lib/qless/qless-core/peek.lua +155 -0
- data/lib/qless/qless-core/pop.lua +278 -0
- data/lib/qless/qless-core/priority.lua +32 -0
- data/lib/qless/qless-core/put.lua +156 -0
- data/lib/qless/qless-core/queues.lua +58 -0
- data/lib/qless/qless-core/recur.lua +181 -0
- data/lib/qless/qless-core/retry.lua +73 -0
- data/lib/qless/qless-core/ruby/lib/qless-core.rb +1 -0
- data/lib/qless/qless-core/ruby/lib/qless/core.rb +13 -0
- data/lib/qless/qless-core/ruby/lib/qless/core/version.rb +5 -0
- data/lib/qless/qless-core/ruby/spec/qless_core_spec.rb +13 -0
- data/lib/qless/qless-core/stats.lua +92 -0
- data/lib/qless/qless-core/tag.lua +100 -0
- data/lib/qless/qless-core/track.lua +79 -0
- data/lib/qless/qless-core/workers.lua +69 -0
- data/lib/qless/queue.rb +141 -0
- data/lib/qless/server.rb +411 -0
- data/lib/qless/tasks.rb +10 -0
- data/lib/qless/version.rb +3 -0
- data/lib/qless/worker.rb +195 -0
- metadata +239 -0
data/Gemfile
ADDED
data/HISTORY.md
ADDED
@@ -0,0 +1,168 @@
|
|
1
|
+
qless
|
2
|
+
=====
|
3
|
+
|
4
|
+
My hope for qless is that it will make certain aspects of pipeline management will be made
|
5
|
+
easier. For the moment, this is a stream of consciousness document meant to capture the
|
6
|
+
features that have been occurring to me lately. After these, I have some initial thoughts
|
7
|
+
on the implementation, concluding with the outstanding __questions__ I have.
|
8
|
+
|
9
|
+
I welcome input on any of this.
|
10
|
+
|
11
|
+
Context
|
12
|
+
-------
|
13
|
+
|
14
|
+
This is a subject that has been on my mind in particular in three contexts:
|
15
|
+
|
16
|
+
1. `custom crawl` -- queue management has always been an annoyance, and it's reaching the
|
17
|
+
breaking point for me
|
18
|
+
1. `freshscape` -- I'm going to be encountering very similar problems like these in freshscape,
|
19
|
+
and I'd like to be able to avoid some of the difficulties I've encountered.
|
20
|
+
1. `general` -- There are a lot of contexts in which such a system would be useful.
|
21
|
+
__Update__ Myron pointed out that in fact `resque` is built on a simple protocol,
|
22
|
+
where each job is a JSON blob with two keys: `id` and `args`. That makes me feel
|
23
|
+
like this is on the right track!
|
24
|
+
|
25
|
+
Feature Requests
|
26
|
+
----------------
|
27
|
+
|
28
|
+
Some of the features that I'm really after include
|
29
|
+
|
30
|
+
1. __Jobs should not get dropped on the floor__ -- This has been a problem for certain
|
31
|
+
projects, including our custom crawler. In this case, jobs had a propensity for
|
32
|
+
getting lost in the shuffle.
|
33
|
+
1. __Job stats should be available__ -- It would be nice to be able to track summary statistics
|
34
|
+
in one place. Perhaps about the number currently in each stage, waiting for each stage,
|
35
|
+
time spent in each stage, number of retries, etc.
|
36
|
+
1. __Job movement should be atomic__ -- One of the problems we've encountered with using
|
37
|
+
Redis is that it's been hard to keep items moving from one queue to another in an atomic
|
38
|
+
way. This has the unfortunate effect of making it difficult to trust the queues to hold
|
39
|
+
any real meaning. For example, the queues use both a list and a hash to track items, and
|
40
|
+
the lengths of the two often get out of sync.
|
41
|
+
1. __Retry logic__ -- For this, I believe we need the ability to support some automatic
|
42
|
+
retry logic. This should be configurable, and based on the stage
|
43
|
+
1. __Data lookups should be easy__ -- It's been difficult to quickly identify a work item and
|
44
|
+
get information on its state. We've usually had to rely on an external storage for this.
|
45
|
+
1. __Manual requeuing__ -- We should be able to safely and atomically move items from one
|
46
|
+
queue into another. We've had problems with race conditions in the past.
|
47
|
+
1. __Priority__ -- Jobs should be describable with priority as well. On occasion we've had
|
48
|
+
to push items through more quickly than others, and it would be great if the underlying
|
49
|
+
system supported that.
|
50
|
+
1. __Tagging / Tracking__ -- It would be nice to be able to mark certain jobs with tags, and
|
51
|
+
as work items that we'd like to track. It should then be possible to get a summary along
|
52
|
+
the lines of "This is the state of the jobs you said you were interested in." I have a
|
53
|
+
system for this set up for my personally, and it has been _extremely_ useful.
|
54
|
+
1. __The system should be reliable and highly available__ -- We're trusting this system to
|
55
|
+
be a single point of failure, and as such it needs to be robust and dependable.
|
56
|
+
1. __High Performance__ -- We should be able to expect this system to support a large number
|
57
|
+
of jobs in a short amount of time. For some context, we need custom crawler to support
|
58
|
+
about 50k state transitions in a day, but my goal is to support millions of transitions
|
59
|
+
in a day, and my bonus goal is 10 million or so transitions in a day.
|
60
|
+
1. __Scheduled Work__ -- We should be able to schedule work items to be enqueued as some
|
61
|
+
specified time.
|
62
|
+
1. __UI__ -- It would be nice to have a useful interface providing insight into the state of
|
63
|
+
the pipeline(s).
|
64
|
+
1. __Namespaces__ -- It might be nice to be able to segment the jobs into namespaces based on
|
65
|
+
project, stage, type, etc. It shouldn't have any explicit meaning outside of partitioning
|
66
|
+
the work space.
|
67
|
+
1. __Language Agnosticism__ -- The lingua franca for this should be something supported by a
|
68
|
+
large number of languages, and the interface should likewise be supported by a large number
|
69
|
+
of languages. In this way, I'd like it to be possible that a job is handled by one language
|
70
|
+
in one stage, and conceivably another in a different stage.
|
71
|
+
1. __Clients Should be Easy to Write__ -- I don't want to put too much burden on the authors
|
72
|
+
of various clients, because I think this helps a project to gain adoption. But, the
|
73
|
+
out-of-the-box feature set should be compelling.
|
74
|
+
|
75
|
+
Thoughts / Recommendations
|
76
|
+
--------------------------
|
77
|
+
|
78
|
+
1. `Redis` as the storage engine. It's been heavily battle-tested, and it's highly available,
|
79
|
+
supports much of the data structures we'd need for these features (atomicity, priority,
|
80
|
+
robustness, performance, replicatable, good at saving state). To boot, it's widely available.
|
81
|
+
1. `JSON` as the lingua franca for communication of work units. Every language I've encountered
|
82
|
+
has strong support for it, it's expressive, and it's human readable.
|
83
|
+
|
84
|
+
Until recently, I had been imagining an HTTP server sitting in front of Redis. Mostly because
|
85
|
+
I figured that would be one way to make clients easy to write -- if all the logic is pinned
|
86
|
+
up in the server. That said, it's just a second moving part upon which to rely. And Myron
|
87
|
+
made the very compelling case for having the clients maintain the state and rely solely on
|
88
|
+
Redis. However, Redis 2.6 provides us with a way to get the best of both worlds -- clients
|
89
|
+
that are easy to write and yet smart enough to do everything themselves. This mechanism is
|
90
|
+
stored Lua scripts.
|
91
|
+
|
92
|
+
Redis has very tight integration with scripting language `Lua`, and the two major selling points
|
93
|
+
for us on this point are:
|
94
|
+
|
95
|
+
1. Atomicity _and_ performance -- Lua scripts are guaranteed to be the only thing running
|
96
|
+
on the Redis instance, and so it makes certain locking mechanisms significantly easier.
|
97
|
+
And since it's running on the actual redis instance, we can enjoy great performance.
|
98
|
+
1. All clients can use the same scripts -- Lua scripts for redis are loaded into the instance,
|
99
|
+
and then can be identified by a hash. But the language invoking them is irrelevant. As such,
|
100
|
+
the burden can still be placed on the original implementation, clients can be easy to
|
101
|
+
write, but still be smart enough to manage the queues themselves.
|
102
|
+
|
103
|
+
One other added benefit is that when using Lua, redis imports a C implementation of a JSON
|
104
|
+
parser and makes it available from Lua scripts. This is just icing on the cake.
|
105
|
+
|
106
|
+
Planned Features / Organization
|
107
|
+
===============================
|
108
|
+
|
109
|
+
All the smarts are essentially going to go into a collection of Lua scripts to be stored and
|
110
|
+
run on Redis. In addition to these Lua scripts, I'd like to provide a simple web interface
|
111
|
+
to provide pretty access to some of this functionality.
|
112
|
+
|
113
|
+
Round 1
|
114
|
+
-------
|
115
|
+
|
116
|
+
1. __Workers must heartbeat jobs__ -- When a worker is given a job, it is given an exclusive
|
117
|
+
lock, and no other worker will get that job so long as it continues to heartbeat. The
|
118
|
+
service keeps track of which locks are going to expire, and will give the work to another
|
119
|
+
worker if the original worker fails to check in. The expiry time is provided every time
|
120
|
+
work is given to a worker, and an updated time is provided when heartbeat-ing. If the
|
121
|
+
lock has been given to another worker, the heartbeat will return `false`.
|
122
|
+
1. __Stats A-Plenty__ -- Stats will be kept of when a job was enqueued for a stage, when it
|
123
|
+
was popped of to be worked on, and when it was completed. In addition, summary statistics
|
124
|
+
will be kept for all the stages.
|
125
|
+
1. __Job Data Stored Temporarily__ -- The data for each job will be stored temporarily. It's
|
126
|
+
yet to be determined exactly what the expiration policy will be, (either the last _k_
|
127
|
+
jobs or the last _x_ amount of time). But still, all the data about a job will be available
|
128
|
+
1. __Atomic Requeueing__ -- If a work item is moved from one queue to another, it is moved.
|
129
|
+
If a worker is in the middle of processing it, its heartbeat will not be renewed, and it
|
130
|
+
will not be allowed to complete.
|
131
|
+
1. __Scheduling / Priority__ -- Jobs can be scheduled to become active at a certain time.
|
132
|
+
This does not mean that the job will be worked on at that time, though. It simply means
|
133
|
+
that after a given scheduled time, it will be considered a candidate, and it will still
|
134
|
+
be subject to the normal priority rules. Priority is always given to an active job with
|
135
|
+
the lowest priority score.
|
136
|
+
1. __Tracking__ -- Jobs can be marked as being of particular interest, and their progress
|
137
|
+
can be tracked accordingly.
|
138
|
+
1. __Web App__ -- A simple web app would be nice
|
139
|
+
|
140
|
+
Round 2
|
141
|
+
-------
|
142
|
+
|
143
|
+
1. __Retry logic__ -- For this, I believe we need the ability to support some automatic
|
144
|
+
retry logic. This should be configurable, and based on the stage
|
145
|
+
1. __Tagging__ -- Jobs should be able to be tagged with certain meaningful flags, like a
|
146
|
+
version number for the software used to process it, or the workers used to process it.
|
147
|
+
|
148
|
+
Questions
|
149
|
+
=========
|
150
|
+
|
151
|
+
1. __Implicit Queue Creation__ -- Each queue needs some configuration, like the heartbeat rate,
|
152
|
+
the time to live for a job, etc. And not only that, but there might be additional more complicated
|
153
|
+
configuration (flow of control). So, which of these should be supported and which not?
|
154
|
+
|
155
|
+
1. Static queue definition -- at a very minimum, we should be able to configure some ahead of time
|
156
|
+
1. Dynamic queue creation -- should there just be another endpoint that allows queues to be added?
|
157
|
+
If so, should these queues then be saved to persist?
|
158
|
+
1. Implicit queue creation -- if we push to a non-existent queue, should we get a warning?
|
159
|
+
An error? Should the queue just be created with some sort of default values?
|
160
|
+
|
161
|
+
On the one hand, I would like to make the system very flexible and amenable to sort of
|
162
|
+
ad-hoc queues, but on the other hand, there may be non-default-able configuration values
|
163
|
+
for queues.
|
164
|
+
|
165
|
+
1. __Job Data Storage__ -- How long should we keep the data about jobs around? We'd like to be
|
166
|
+
able to get information about a job, but those should probably be expired. Should expiration
|
167
|
+
policy be set to hold jobs for a certain amount of time? Should this window be configured for
|
168
|
+
simply the last _k_ jobs?
|
data/README.md
ADDED
@@ -0,0 +1,571 @@
|
|
1
|
+
qless
|
2
|
+
=====
|
3
|
+
|
4
|
+
Qless is a powerful `Redis`-based job queueing system inspired by
|
5
|
+
[resque](https://github.com/defunkt/resque#readme),
|
6
|
+
but built on a collection of Lua scripts, maintained in the
|
7
|
+
[qless-core](https://github.com/seomoz/qless-core) repo.
|
8
|
+
|
9
|
+
Philosophy and Nomenclature
|
10
|
+
===========================
|
11
|
+
A `job` is a unit of work identified by a job id or `jid`. A `queue` can contain
|
12
|
+
several jobs that are scheduled to be run at a certain time, several jobs that are
|
13
|
+
waiting to run, and jobs that are currently running. A `worker` is a process on a
|
14
|
+
host, identified uniquely, that asks for jobs from the queue, performs some process
|
15
|
+
associated with that job, and then marks it as complete. When it's completed, it
|
16
|
+
can be put into another queue.
|
17
|
+
|
18
|
+
Jobs can only be in one queue at a time. That queue is whatever queue they were last
|
19
|
+
put in. So if a worker is working on a job, and you move it, the worker's request to
|
20
|
+
complete the job will be ignored.
|
21
|
+
|
22
|
+
A job can be `canceled`, which means it disappears into the ether, and we'll never
|
23
|
+
pay it any mind every again. A job can be `dropped`, which is when a worker fails
|
24
|
+
to heartbeat or complete the job in a timely fashion, or a job can be `failed`,
|
25
|
+
which is when a host recognizes some systematically problematic state about the
|
26
|
+
job. A worker should only fail a job if the error is likely not a transient one;
|
27
|
+
otherwise, that worker should just drop it and let the system reclaim it.
|
28
|
+
|
29
|
+
Features
|
30
|
+
========
|
31
|
+
|
32
|
+
1. __Jobs don't get dropped on the floor__ -- Sometimes workers drop jobs. Qless
|
33
|
+
automatically picks them back up and gives them to another worker
|
34
|
+
1. __Tagging / Tracking__ -- Some jobs are more interesting than others. Track those
|
35
|
+
jobs to get updates on their progress. Tag jobs with meaningful identifiers to
|
36
|
+
find them quickly in the UI.
|
37
|
+
1. __Job Dependencies__ -- One job might need to wait for another job to complete
|
38
|
+
1. __Stats__ -- `qless` automatically keeps statistics about how long jobs wait
|
39
|
+
to be processed and how long they take to be processed. Currently, we keep
|
40
|
+
track of the count, mean, standard deviation, and a histogram of these times.
|
41
|
+
1. __Job data is stored temporarily__ -- Job info sticks around for a configurable
|
42
|
+
amount of time so you can still look back on a job's history, data, etc.
|
43
|
+
1. __Priority__ -- Jobs with the same priority get popped in the order they were
|
44
|
+
inserted; a higher priority means that it gets popped faster
|
45
|
+
1. __Retry logic__ -- Every job has a number of retries associated with it, which are
|
46
|
+
renewed when it is put into a new queue or completed. If a job is repeatedly
|
47
|
+
dropped, then it is presumed to be problematic, and is automatically failed.
|
48
|
+
1. __Web App__ -- With the advent of a Ruby client, there is a Sinatra-based web
|
49
|
+
app that gives you control over certain operational issues
|
50
|
+
1. __Scheduled Work__ -- Until a job waits for a specified delay (defaults to 0),
|
51
|
+
jobs cannot be popped by workers
|
52
|
+
1. __Recurring Jobs__ -- Scheduling's all well and good, but we also support
|
53
|
+
jobs that need to recur periodically.
|
54
|
+
1. __Notifications__ -- Tracked jobs emit events on pubsub channels as they get
|
55
|
+
completed, failed, put, popped, etc. Use these events to get notified of
|
56
|
+
progress on jobs you're interested in.
|
57
|
+
|
58
|
+
Enqueing Jobs
|
59
|
+
=============
|
60
|
+
First things first, require `qless` and create a client. The client accepts all the
|
61
|
+
same arguments that you'd use when constructing a redis client.
|
62
|
+
|
63
|
+
``` ruby
|
64
|
+
require 'qless'
|
65
|
+
|
66
|
+
# Connect to localhost
|
67
|
+
client = Qless::Client.new
|
68
|
+
# Connect to somewhere else
|
69
|
+
client = Qless::Client.new(:host => 'foo.bar.com', :port => 1234)
|
70
|
+
```
|
71
|
+
|
72
|
+
Jobs should be classes or modules that define a `perform` method, which
|
73
|
+
must accept a single `job` argument:
|
74
|
+
|
75
|
+
``` ruby
|
76
|
+
class MyJobClass
|
77
|
+
def self.perform(job)
|
78
|
+
# job is an instance of `Qless::Job` and provides access to
|
79
|
+
# job.data, a means to cancel the job (job.cancel), and more.
|
80
|
+
end
|
81
|
+
end
|
82
|
+
```
|
83
|
+
|
84
|
+
Now you can access a queue, and add a job to that queue.
|
85
|
+
|
86
|
+
``` ruby
|
87
|
+
# This references a new or existing queue 'testing'
|
88
|
+
queue = client.queues['testing']
|
89
|
+
# Let's add a job, with some data. Returns Job ID
|
90
|
+
queue.put(MyJobClass, :hello => 'howdy')
|
91
|
+
# => "0c53b0404c56012f69fa482a1427ab7d"
|
92
|
+
# Now we can ask for a job
|
93
|
+
job = queue.pop
|
94
|
+
# => <Qless::Job 0c53b0404c56012f69fa482a1427ab7d (MyJobClass / testing)>
|
95
|
+
# And we can do the work associated with it!
|
96
|
+
job.perform
|
97
|
+
```
|
98
|
+
|
99
|
+
The job data must be serializable to JSON, and it is recommended
|
100
|
+
that you use a hash for it. See below for a list of the supported job options.
|
101
|
+
|
102
|
+
The argument returned by `queue.put` is the job ID, or jid. Every Qless
|
103
|
+
job has a unique jid, and it provides a means to interact with an
|
104
|
+
existing job:
|
105
|
+
|
106
|
+
``` ruby
|
107
|
+
# find an existing job by it's jid
|
108
|
+
job = client.jobs[jid]
|
109
|
+
|
110
|
+
# Query it to find out details about it:
|
111
|
+
job.klass # => the class of the job
|
112
|
+
job.queue # => the queue the job is in
|
113
|
+
job.data # => the data for the job
|
114
|
+
job.history # => the history of what has happened to the job sofar
|
115
|
+
job.dependencies # => the jids of other jobs that must complete before this one
|
116
|
+
job.dependents # => the jids of other jobs that depend on this one
|
117
|
+
job.priority # => the priority of this job
|
118
|
+
job.tags # => array of tags for this job
|
119
|
+
job.original_retries # => the number of times the job is allowed to be retried
|
120
|
+
job.retries_left # => the number of retries left
|
121
|
+
|
122
|
+
# You can also change the job in various ways:
|
123
|
+
job.move("some_other_queue") # move it to a new queue
|
124
|
+
job.cancel # cancel the job
|
125
|
+
job.tag("foo") # add a tag
|
126
|
+
job.untag("foo") # remove a tag
|
127
|
+
```
|
128
|
+
|
129
|
+
Running A Worker
|
130
|
+
================
|
131
|
+
|
132
|
+
The Qless ruby worker was heavily inspired by Resque's worker,
|
133
|
+
but thanks to the power of the qless-core lua scripts, it is
|
134
|
+
*much* simpler and you are welcome to write your own (e.g. if
|
135
|
+
you'd rather save memory by not forking the worker for each job).
|
136
|
+
|
137
|
+
As with resque...
|
138
|
+
|
139
|
+
* The worker forks a child process for each job in order to provide
|
140
|
+
resilience against memory leaks. Pass the `RUN_AS_SINGLE_PROCESS`
|
141
|
+
environment variable to force Qless to not fork the child process.
|
142
|
+
Single process mode should only be used in some test/dev
|
143
|
+
environments.
|
144
|
+
* The worker updates its procline with its status so you can see
|
145
|
+
what workers are doing using `ps`.
|
146
|
+
* The worker registers signal handlers so that you can control it
|
147
|
+
by sending it signals.
|
148
|
+
* The worker is given a list of queues to pop jobs off of.
|
149
|
+
* The worker logs out put based on `VERBOSE` or `VVERBOSE` (very
|
150
|
+
verbose) environment variables.
|
151
|
+
* Qless ships with a rake task (`qless:work`) for running workers.
|
152
|
+
It runs `qless:setup` before starting the main work loop so that
|
153
|
+
users can load their environment in that task.
|
154
|
+
* The sleep interval (for when there is no jobs available) can be
|
155
|
+
configured with the `INTERVAL` environment variable.
|
156
|
+
|
157
|
+
Resque uses queues for its notion of priority. In contrast, qless
|
158
|
+
has priority support built-in. Thus, the worker supports two strategies
|
159
|
+
for what order to pop jobs off the queues: ordered and round-robin.
|
160
|
+
The ordered reserver will keep popping jobs off the first queue until
|
161
|
+
it is empty, before trying to pop job off the second queue. The
|
162
|
+
round-robin reserver will pop a job off the first queue, then the second
|
163
|
+
queue, and so on. You could also easily implement your own.
|
164
|
+
|
165
|
+
To start a worker, load the qless rake tasks in your Rakefile, and
|
166
|
+
define a `qless:setup` task:
|
167
|
+
|
168
|
+
``` ruby
|
169
|
+
require 'qless/tasks'
|
170
|
+
namespace :qless do
|
171
|
+
task :setup do
|
172
|
+
require 'my_app/environment' # to ensure all job classes are loaded
|
173
|
+
|
174
|
+
# Set options via environment variables
|
175
|
+
# The only required option is QUEUES; the
|
176
|
+
# rest have reasonable defaults.
|
177
|
+
ENV['REDIS_URL'] ||= 'redis://some-host:7000/3'
|
178
|
+
ENV['QUEUES'] ||= 'fizz,buzz'
|
179
|
+
ENV['JOB_RESERVER'] ||= 'Ordered'
|
180
|
+
ENV['INTERVAL'] ||= '10' # 10 seconds
|
181
|
+
ENV['VERBOSE'] ||= 'true'
|
182
|
+
end
|
183
|
+
end
|
184
|
+
```
|
185
|
+
|
186
|
+
Then run the `qless:work` rake task:
|
187
|
+
|
188
|
+
```
|
189
|
+
rake qless:work
|
190
|
+
```
|
191
|
+
|
192
|
+
The following signals are supported:
|
193
|
+
|
194
|
+
* TERM: Shutdown immediately, stop processing jobs.
|
195
|
+
* INT: Shutdown immediately, stop processing jobs.
|
196
|
+
* QUIT: Shutdown after the current job has finished processing.
|
197
|
+
* USR1: Kill the forked child immediately, continue processing jobs.
|
198
|
+
* USR2: Don't process any new jobs
|
199
|
+
* CONT: Start processing jobs again after a USR2
|
200
|
+
|
201
|
+
You should send these to the master process, not the child.
|
202
|
+
|
203
|
+
Workers also support middleware modules that can be used to inject
|
204
|
+
logic before, after or around the processing of a single job in
|
205
|
+
the child process. This can be useful, for example, when you need to
|
206
|
+
re-establish a connection to your database in each job.
|
207
|
+
|
208
|
+
Define a module with an `around_perform` method that calls `super` where you
|
209
|
+
want the job to be processed:
|
210
|
+
|
211
|
+
``` ruby
|
212
|
+
module ReEstablishDBConnection
|
213
|
+
def around_perform(job)
|
214
|
+
MyORM.establish_connection
|
215
|
+
super
|
216
|
+
end
|
217
|
+
end
|
218
|
+
```
|
219
|
+
|
220
|
+
Then, mix-it into the worker class. You can mix-in as many
|
221
|
+
middleware modules as you like:
|
222
|
+
|
223
|
+
``` ruby
|
224
|
+
require 'qless/worker'
|
225
|
+
Qless::Worker.class_eval do
|
226
|
+
include ReEstablishDBConnection
|
227
|
+
include SomeOtherAwesomeMiddleware
|
228
|
+
end
|
229
|
+
```
|
230
|
+
|
231
|
+
Web Interface
|
232
|
+
=============
|
233
|
+
|
234
|
+
Qless ships with a resque-inspired web app that lets you easily
|
235
|
+
deal with failures and see what it is processing. If you're project
|
236
|
+
has a rack-based ruby web app, we recommend you mount Qless's web app
|
237
|
+
in it. Here's how you can do that with `Rack::Builder` in your `config.ru`:
|
238
|
+
|
239
|
+
``` ruby
|
240
|
+
Qless::Server.client = Qless::Client.new(:host => "some-host", :port => 7000)
|
241
|
+
|
242
|
+
Rack::Builder.new do
|
243
|
+
use SomeMiddleware
|
244
|
+
|
245
|
+
map('/some-other-app') { run Apps::Something.new }
|
246
|
+
map('/qless') { run Qless::Server.new }
|
247
|
+
end
|
248
|
+
```
|
249
|
+
|
250
|
+
For an app using Rails 3+, check the router documentation for how to mount
|
251
|
+
rack apps.
|
252
|
+
|
253
|
+
Job Dependencies
|
254
|
+
================
|
255
|
+
Let's say you have one job that depends on another, but the task definitions are
|
256
|
+
fundamentally different. You need to bake a turkey, and you need to make stuffing,
|
257
|
+
but you can't make the turkey until the stuffing is made:
|
258
|
+
|
259
|
+
``` ruby
|
260
|
+
queue = client.queues['cook']
|
261
|
+
stuffing_jid = queue.put(MakeStuffing, {:lots => 'of butter'})
|
262
|
+
turkey_jid = queue.put(MakeTurkey , {:with => 'stuffing'}, :depends=>[stuffing_jid])
|
263
|
+
```
|
264
|
+
|
265
|
+
When the stuffing job completes, the turkey job is unlocked and free to be processed.
|
266
|
+
|
267
|
+
Priority
|
268
|
+
========
|
269
|
+
Some jobs need to get popped sooner than others. Whether it's a trouble ticket, or
|
270
|
+
debugging, you can do this pretty easily when you put a job in a queue:
|
271
|
+
|
272
|
+
``` ruby
|
273
|
+
queue.put(MyJobClass, {:foo => 'bar'}, :priority => 10)
|
274
|
+
```
|
275
|
+
|
276
|
+
What happens when you want to adjust a job's priority while it's still waiting in
|
277
|
+
a queue?
|
278
|
+
|
279
|
+
``` ruby
|
280
|
+
job = client.jobs['0c53b0404c56012f69fa482a1427ab7d']
|
281
|
+
job.priority = 10
|
282
|
+
# Now this will get popped before any job of lower priority
|
283
|
+
```
|
284
|
+
|
285
|
+
Scheduled Jobs
|
286
|
+
==============
|
287
|
+
If you don't want a job to be run right away but some time in the future, you can
|
288
|
+
specify a delay:
|
289
|
+
|
290
|
+
``` ruby
|
291
|
+
# Run at least 10 minutes from now
|
292
|
+
queue.put(MyJobClass, {:foo => 'bar'}, :delay => 600)
|
293
|
+
```
|
294
|
+
|
295
|
+
This doesn't guarantee that job will be run exactly at 10 minutes. You can accomplish
|
296
|
+
this by changing the job's priority so that once 10 minutes has elapsed, it's put before
|
297
|
+
lesser-priority jobs:
|
298
|
+
|
299
|
+
``` ruby
|
300
|
+
# Run in 10 minutes
|
301
|
+
queue.put(MyJobClass, {:foo => 'bar'}, :delay => 600, :priority => 100)
|
302
|
+
```
|
303
|
+
|
304
|
+
Recurring Jobs
|
305
|
+
==============
|
306
|
+
Sometimes it's not enough simply to schedule one job, but you want to run jobs regularly.
|
307
|
+
In particular, maybe you have some batch operation that needs to get run once an hour and
|
308
|
+
you don't care what worker runs it. Recurring jobs are specified much like other jobs:
|
309
|
+
|
310
|
+
``` ruby
|
311
|
+
# Run every hour
|
312
|
+
queue.recur(MyJobClass, {:widget => 'warble'}, 3600)
|
313
|
+
# => 22ac75008a8011e182b24cf9ab3a8f3b
|
314
|
+
```
|
315
|
+
|
316
|
+
You can even access them in much the same way as you would normal jobs:
|
317
|
+
|
318
|
+
``` ruby
|
319
|
+
job = client.jobs['22ac75008a8011e182b24cf9ab3a8f3b']
|
320
|
+
# => < Qless::RecurringJob 22ac75008a8011e182b24cf9ab3a8f3b >
|
321
|
+
```
|
322
|
+
|
323
|
+
Changing the interval at which it runs after the fact is trivial:
|
324
|
+
|
325
|
+
``` ruby
|
326
|
+
# I think I only need it to run once every two hours
|
327
|
+
job.interval = 7200
|
328
|
+
```
|
329
|
+
|
330
|
+
If you want it to run every hour on the hour, but it's 2:37 right now, you can specify
|
331
|
+
an offset which is how long it should wait before popping the first job:
|
332
|
+
|
333
|
+
``` ruby
|
334
|
+
# 23 minutes of waiting until it should go
|
335
|
+
queue.recur(MyJobClass, {:howdy => 'hello'}, 3600, :offset => 23 * 60)
|
336
|
+
```
|
337
|
+
|
338
|
+
Recurring jobs also have priority, a configurable number of retries, and tags. These
|
339
|
+
settings don't apply to the recurring jobs, but rather the jobs that they create. In the
|
340
|
+
case where more than one interval passes before a worker tries to pop the job, __more than
|
341
|
+
one job is created__. The thinking is that while it's completely client-managed, the state
|
342
|
+
should not be dependent on how often workers are trying to pop jobs.
|
343
|
+
|
344
|
+
``` ruby
|
345
|
+
# Recur every minute
|
346
|
+
queue.recur(MyJobClass, {:lots => 'of jobs'}, 60)
|
347
|
+
# Wait 5 minutes
|
348
|
+
queue.pop(10).length
|
349
|
+
# => 5 jobs got popped
|
350
|
+
```
|
351
|
+
|
352
|
+
Configuration Options
|
353
|
+
=====================
|
354
|
+
You can get and set global (read: in the context of the same Redis instance) configuration
|
355
|
+
to change the behavior for heartbeating, and so forth. There aren't a tremendous number
|
356
|
+
of configuration options, but an important one is how long job data is kept around. Job
|
357
|
+
data is expired after it has been completed for `jobs-history` seconds, but is limited to
|
358
|
+
the last `jobs-history-count` completed jobs. These default to 50k jobs, and 30 days, but
|
359
|
+
depending on volume, your needs may change. To only keep the last 500 jobs for up to 7 days:
|
360
|
+
|
361
|
+
``` ruby
|
362
|
+
client.config['jobs-history'] = 7 * 86400
|
363
|
+
client.config['jobs-history-count'] = 500
|
364
|
+
```
|
365
|
+
|
366
|
+
Tagging / Tracking
|
367
|
+
==================
|
368
|
+
In qless, 'tracking' means flagging a job as important. Tracked jobs have a tab reserved
|
369
|
+
for them in the web interface, and they also emit subscribable events as they make progress
|
370
|
+
(more on that below). You can flag a job from the web interface, or the corresponding code:
|
371
|
+
|
372
|
+
``` ruby
|
373
|
+
client.jobs['b1882e009a3d11e192d0b174d751779d'].track
|
374
|
+
```
|
375
|
+
|
376
|
+
Jobs can be tagged with strings which are indexed for quick searches. For example, jobs
|
377
|
+
might be associated with customer accounts, or some other key that makes sense for your
|
378
|
+
project.
|
379
|
+
|
380
|
+
``` ruby
|
381
|
+
queue.put(MyJobClass, {:tags => 'aplenty'}, :tags => ['12345', 'foo', 'bar'])
|
382
|
+
```
|
383
|
+
|
384
|
+
This makes them searchable in the web interface, or from code:
|
385
|
+
|
386
|
+
``` ruby
|
387
|
+
jids = client.jobs.tagged('foo')
|
388
|
+
```
|
389
|
+
|
390
|
+
You can add or remove tags at will, too:
|
391
|
+
|
392
|
+
``` ruby
|
393
|
+
job = client.jobs['b1882e009a3d11e192d0b174d751779d']
|
394
|
+
job.tag('howdy', 'hello')
|
395
|
+
job.untag('foo', 'bar')
|
396
|
+
```
|
397
|
+
|
398
|
+
Notifications
|
399
|
+
=============
|
400
|
+
Tracked jobs emit events on specific pubsub channels as things happen to them. Whether
|
401
|
+
it's getting popped off of a queue, completed by a worker, etc. A good example of how
|
402
|
+
to make use of this is in the `qless-campfire` or `qless-growl`. The jist of it goes like
|
403
|
+
this, though:
|
404
|
+
|
405
|
+
``` ruby
|
406
|
+
client.events do |on|
|
407
|
+
on.canceled { |jid| puts "#{jid} canceled" }
|
408
|
+
on.stalled { |jid| puts "#{jid} stalled" }
|
409
|
+
on.track { |jid| puts "tracking #{jid}" }
|
410
|
+
on.untrack { |jid| puts "untracking #{jid}" }
|
411
|
+
on.completed { |jid| puts "#{jid} completed" }
|
412
|
+
on.failed { |jid| puts "#{jid} failed" }
|
413
|
+
on.popped { |jid| puts "#{jid} popped" }
|
414
|
+
on.put { |jid| puts "#{jid} put" }
|
415
|
+
end
|
416
|
+
```
|
417
|
+
|
418
|
+
Those familiar with redis pubsub will note that a redis connection can only be used
|
419
|
+
for pubsub-y commands once listening. For this reason, invoking `client.events` actually
|
420
|
+
creates a second connection so that `client` can still be used as it normally would be:
|
421
|
+
|
422
|
+
``` ruby
|
423
|
+
client.events do |on|
|
424
|
+
on.failed do |jid|
|
425
|
+
puts "#{jid} failed in #{client.jobs[jid].queue_name}"
|
426
|
+
end
|
427
|
+
end
|
428
|
+
```
|
429
|
+
|
430
|
+
Heartbeating
|
431
|
+
============
|
432
|
+
When a worker is given a job, it is given an exclusive lock to that job. That means
|
433
|
+
that job won't be given to any other worker, so long as the worker checks in with
|
434
|
+
progress on the job. By default, jobs have to either report back progress every 60
|
435
|
+
seconds, or complete it, but that's a configurable option. For longer jobs, this
|
436
|
+
may not make sense.
|
437
|
+
|
438
|
+
``` ruby
|
439
|
+
# Hooray! We've got a piece of work!
|
440
|
+
job = queue.pop
|
441
|
+
# How long until I have to check in?
|
442
|
+
job.ttl
|
443
|
+
# => 59
|
444
|
+
# Hey! I'm still working on it!
|
445
|
+
job.heartbeat
|
446
|
+
# => 1331326141.0
|
447
|
+
# Ok, I've got some more time. Oh! Now I'm done!
|
448
|
+
job.complete
|
449
|
+
```
|
450
|
+
|
451
|
+
If you want to increase the heartbeat in all queues,
|
452
|
+
|
453
|
+
``` ruby
|
454
|
+
# Now jobs get 10 minutes to check in
|
455
|
+
client.config['heartbeat'] = 600
|
456
|
+
# But the testing queue doesn't get as long.
|
457
|
+
client.queues['testing'].heartbeat = 300
|
458
|
+
```
|
459
|
+
|
460
|
+
When choosing a heartbeat interval, realize that this is the amount of time that
|
461
|
+
can pass before qless realizes if a job has been dropped. At the same time, you don't
|
462
|
+
want to burden qless with heartbeating every 10 seconds if your job is expected to
|
463
|
+
take several hours.
|
464
|
+
|
465
|
+
An idiom you're encouraged to use for long-running jobs that want to check in their
|
466
|
+
progress periodically:
|
467
|
+
|
468
|
+
``` ruby
|
469
|
+
# Wait until we have 5 minutes left on the heartbeat, and if we find that
|
470
|
+
# we've lost our lock on a job, then honorable fall on our sword
|
471
|
+
if (job.ttl < 300) && !job.heartbeat
|
472
|
+
return / die / exit
|
473
|
+
end
|
474
|
+
```
|
475
|
+
|
476
|
+
Stats
|
477
|
+
=====
|
478
|
+
One nice feature of `qless` is that you can get statistics about usage. Stats are
|
479
|
+
aggregated by day, so when you want stats about a queue, you need to say what queue
|
480
|
+
and what day you're talking about. By default, you just get the stats for today.
|
481
|
+
These stats include information about the mean job wait time, standard deviation,
|
482
|
+
and histogram. This same data is also provided for job completion:
|
483
|
+
|
484
|
+
``` ruby
|
485
|
+
# So, how're we doing today?
|
486
|
+
stats = client.stats.get('testing')
|
487
|
+
# => { 'run' => {'mean' => ..., }, 'wait' => {'mean' => ..., }}
|
488
|
+
```
|
489
|
+
|
490
|
+
Time
|
491
|
+
====
|
492
|
+
It's important to note that Redis doesn't allow access to the system time if you're
|
493
|
+
going to be making any manipulations to data (which our scripts do). And yet, we
|
494
|
+
have heartbeating. This means that the clients actually send the current time when
|
495
|
+
making most requests, and for consistency's sake, means that your workers must be
|
496
|
+
relatively synchronized. This doesn't mean down to the tens of milliseconds, but if
|
497
|
+
you're experiencing appreciable clock drift, you should investigate NTP. For what it's
|
498
|
+
worth, this hasn't been a problem for us, but most of our jobs have heartbeat intervals
|
499
|
+
of 30 minutes or more.
|
500
|
+
|
501
|
+
Ensuring Job Uniqueness
|
502
|
+
=======================
|
503
|
+
|
504
|
+
As mentioned above, Jobs are uniquely identied by an id--their jid.
|
505
|
+
Qless will generate a UUID for each enqueued job or you can specify
|
506
|
+
one manually:
|
507
|
+
|
508
|
+
``` ruby
|
509
|
+
queue.put(MyJobClass, { :hello => 'howdy' }, :jid => 'my-job-jid')
|
510
|
+
```
|
511
|
+
|
512
|
+
This can be useful when you want to ensure a job's uniqueness: simply
|
513
|
+
create a jid that is a function of the Job's class and data, it'll
|
514
|
+
guaranteed that Qless won't have multiple jobs with the same class
|
515
|
+
and data.
|
516
|
+
|
517
|
+
Setting Default Job Options
|
518
|
+
===========================
|
519
|
+
|
520
|
+
`Qless::Queue#put` accepts a number of job options (see above for their
|
521
|
+
semantics):
|
522
|
+
|
523
|
+
* jid
|
524
|
+
* delay
|
525
|
+
* priority
|
526
|
+
* tags
|
527
|
+
* retries
|
528
|
+
* depends
|
529
|
+
|
530
|
+
When enqueueing the same kind of job with the same args in multiple
|
531
|
+
places it's a pain to have to declare the job options every time.
|
532
|
+
Instead, you can define default job options directly on the job class:
|
533
|
+
|
534
|
+
``` ruby
|
535
|
+
class MyJobClass
|
536
|
+
def self.default_job_options(data)
|
537
|
+
{ :priority => 10, :delay => 100 }
|
538
|
+
end
|
539
|
+
end
|
540
|
+
|
541
|
+
queue.put(MyJobClass, { :some => "data" }, :delay => 10)
|
542
|
+
```
|
543
|
+
|
544
|
+
Individual jobs can still specify options, so in this example,
|
545
|
+
the job would be enqueued with a priority of 10 and a delay of 10.
|
546
|
+
|
547
|
+
Testing Jobs
|
548
|
+
============
|
549
|
+
When unit testing your jobs, you will probably want to avoid the
|
550
|
+
overhead of round-tripping them through redis. You can of course
|
551
|
+
use a mock job object and pass it to your job class's `perform`
|
552
|
+
method. Alternately, if you want a real full-fledged `Qless::Job`
|
553
|
+
instance without round-tripping it through Redis, use `Qless::Job.build`:
|
554
|
+
|
555
|
+
``` ruby
|
556
|
+
describe MyJobClass do
|
557
|
+
let(:client) { Qless::Client.new }
|
558
|
+
let(:job) { Qless::Job.build(client, MyJobClass, :data => { "some" => "data" }) }
|
559
|
+
|
560
|
+
it 'does something' do
|
561
|
+
MyJobClass.perform(job)
|
562
|
+
# make an assertion about what happened
|
563
|
+
end
|
564
|
+
end
|
565
|
+
```
|
566
|
+
|
567
|
+
The options hash passed to `Qless::Job.build` supports all the same
|
568
|
+
options a normal job supports. See
|
569
|
+
[the source](https://github.com/seomoz/qless/blob/master/lib/qless/job.rb)
|
570
|
+
for a full list.
|
571
|
+
|