A persistent and flexible background jobs library for go.
Version: 0.1.0
Jobs is powered by redis and supports the following features:
- A job can encapsulate any arbitrary functionality. A job can do anything which can be done in a go function.
- A job can be one-off (only executed once) or recurring (scheduled to execute at a specific interval).
- A job can be retried a specified number of times if it fails.
- A job is persistent, with protections against power loss and other worst case scenarios. (See the Guarantees section below)
- Work on jobs can be spread amongst any number of concurrent workers accross any number of machines.
- Every job will be executed at least once, and in all but the most extreme cases will be executed exactly once. (See the Guarantees section below)
- You can query the database to find out e.g. the number of jobs that are currently executing or how long a particular job took to execute.
- Any job that permanently fails will have its error captured and stored.
Jobs is intended to be used in web applications. It is useful for cases where you need to execute some long-running code, but you don't want your users to wait for the code to execute before rendering a response. A good example is sending a welcome email to your users after they sign up. You can use Jobs to schedule the email to be sent asynchronously, and render a response to your user without waiting for the email to be sent. You could use a goroutine to accomplish the same thing, but in the event of a server restart or power loss, the email might never be sent. Jobs guarantees that the email will be sent at some time, and allows you to spread the work between different machines.
Jobs must be organized into discrete types. Here's an example of how to register a job which sends a welcome email to users:
// We'll specify that we want the job to be retried 3 times before finally failing
welcomeEmailJobs, err := jobs.RegisterType("welcomeEmail", 3, func(user *User) {
msg := fmt.Sprintf("Hello, %s! Thanks for signing up for foo.com.", user.Name)
if err := emails.Send(user.EmailAddress, msg); err != nil {
// Panics will be captured by a worker, triggering up to 3 retries
panic(err)
}
})After registering a job type, you can schedule a job using the Schedule or ScheduleRecurring methods like so:
// The priority argument lets you choose how important the job is. Higher
// priority jobs will be executed first.
job, err := welcomeEmailJobs.Schedule(100, time.Now(), &User{EmailAddress: "[email protected]"})
if err != nil {
// Handle err
}You can use the Job object returned by Schedule or ScheduleRecurring to check on the status of the job or cancel it manually.
You can schedule any number of worker pools accross any number of machines, provided every machine agrees on the definition of the job types. If you want, you can start a worker pool on the same machines that are scheduling jobs, or you can have each worker pool running on a designated machine. It is technically safe to start multiple pools on a single machine, but typically you should only start one pool per machine.
To create a new pool with the default configuration, just pass in nil:
pool := jobs.NewPool(nil)You can also specify a different configuration by passing in *PoolConfig. Any zero values in the config you pass in will fallback to the default values. So here's how you could start a pool with 10 workers and a batch size of 10, while letting the other options remain the default.
pool := jobs.NewPool(&jobs.PoolConfig{
NumWorkers: 10,
BatchSize: 10,
})After you have created a pool, you can start it with the Start method. Once started, the pool will continuously query the database for new jobs and delegate those jobs to workers. Any program that calls Pool.Start() should also wait for the workers to finish before exiting. You can do so by wrapping Close and Wait in a defer statement. Typical usage looks something like this:
func main() {
pool := jobs.NewPool(nil)
defer func() {
pool.Close()
if err := pool.Wait(); err != nil {
// Handle err
}
}
if err := pool.Start(); err != nil {
// Handle err
}
}You can also call Close and Wait at any time to manually stop the pool from executing new jobs. In this case, any jobs that are currently being executed will still finish.
Since jobs is powered by redis, there is a chance that you can lose data with the default redis configuration. To get the best persistence guarantees, you should set redis to use both AOF and RDB persistence modes and set fsync to "always". With these settings, redis is more or less as persistent as a database like postgres. If want better performance and are okay with a greater (but still very small) chance of losing data, you can set fsync to "everysec".
Read more about redis persistence.
Jobs is carefully written using redis transactions and lua scripting so that all database changes are atomic. If redis crashes in the middle of a transaction or script execution, it is possible that your AOF file can become corrupted. If this happens, redis will refuse to start until the AOF file is fixed. It is relatively easy to fix the problem with the redis-check-aof tool, which will remove the partial transaction from the AOF file. In effect, this guarantees that modifications of the database are atomic, even in the event of a power loss or hard reset, with the caveat that you may need to break out the redis-check-aof tool in the worst case scenario.
Read more about redis transactions and scripts.
Jobs gaurantees that a job will be executed at least once, provided it has been persisted on disk. (See the section on Persistence directly above). A job can only picked up by one pool at a time becuase a pool atomically pops (gets and immediately moves) the next available jobs from the database. A job can only be executed by one worker at a time becuase the jobs are delegated to workers via a shared channel. Each worker pool checks on the health of all the other poools when it starts. If a pool crashes or is otherwise disconnected, any jobs it had grabbed from the database that did not yet finish will be requeued and picked up by a different pool.
There are two known ways that a job may be executed more than once:
- If there is a power failure or hard reset while a worker is in the middle of executing a job, the job may be stuck in a half-executed state. Since there is no way to know how much of the job was successfully completed, the job will be requeued and picked up by a different pool, where it may be partially or fully executed more than once.
- If a pool becomes disconnected, it will be considered stale and its jobs will be requeued and reclaimed by a different pool. However, if the stale pool is able to partly or fully execute jobs without a reliable internet connection, any jobs belonging to the stale pool might be executed more than once. You can increase the StaleTimeout parameter for a pool to make this scenario less likely.
Jobs is licensed under the MIT License. See the LICENSE file for more information.