GitHub - zalgonoise/micron: This is my cron (micron); there are others like it but this one is mine. A cron library in Go.

micron

This is my cron (micron); there are others like it but this one is mine. A cron-scheduler library in Go

Concept

cron is a Go library that allows adding cron-like scheduler(s) to Go apps, compatible with Unix's cron time/date strings, to execute actions within the context of the app.

By itself, cron is a fantastic tool released in the mid-70's, written in C, where the user defines a specification in a crontab, a file listing jobs with a reference of a time/date specification and a (Unix) command to execute.

Within Go, it should provide the same set of features as the original binary, but served in a library as a (blocking / background) pluggable service. This means full compatibility with cron strings for scheduling, support for multiple executable types or functions, and support for multiple job scheduling (with different time/date specifications). Additionally, the library extends this functionality to support definition of seconds' frequency in cron-strings, context support, error returns, and full observability coverage (with added support for metrics, logs and traces decorators).

Motivation

In a work environment, we see cron many times, in different scenarios. From cron installations in bare-metal Linux servers to one-shot containers configured in Kubernetes deployments. Some use-cases are very simple, others complex, but the point is that is a tool used for nearly 50 years at the time of writing.

But the original tool is a C binary that executes a Unix command. If we want to explore schedulers for Go applications (e.g. a script executed every # hours), this means that the app needs to be compiled as a binary, and then to configure a cron-job to execute the app in a command.

While this is fine, it raises the question -- what if I want to include it within the application? This should make a lot of sense to you if you're a fan of SQLite like me.

There were already two libraries with different implementations, in Go:

robfig/cron with 12k GitHub stars
go-co-op/gocron with 4.3k GitHub stars

Don't get me wrong -- there is nothing inherently wrong about these implementations; I went through them carefully both for insight and to understand what could I explore differently. A very obvious change would be a more "modern" approach including a newer Go version (as these required Go 1.12 and 1.16 respectively); which by itself includes log/slog and all the other observability-related decorators that also leverage context.Context.

Another more obvious exploration path would be the parser logic, as I could use my generic lexer and generic parser in order to potentially improve it.

Lastly I could try to split the cron service's components to be more configurable even in future iterations, once I had decided on the general API for the library. There was enough ground to explore and to give it a go. :)

A personal project that I have for a Steam CLI app is currently using this cron library to regularly check for discounts in the Steam Store, for certain products on a certain frequency, as configured by the user.

Usage

Using cron is as layered and modular as you want it to be. This chapter describes how to use the library effectively.

Getting `cron`

You're able to fetch cron as a Go module by importing it in your project and running go mod tidy:

package main

import (
	"fmt"
	"context"
	
	"github.com/zalgonoise/micron"
	"github.com/zalgonoise/micron/executor"
)

func main() {
	fn := func(context.Context) error {
		fmt.Println("done!")

		return nil
	}
	
	c, err := micron.New(micron.WithJob("my-job", "* * * * *", executor.Runnable(fn)))
	// ...
}

Cron Runtime

The runtime is the component that will control (like the name implies) how the module runs -- that is, controlling the flow of job selection and execution. The runtime will allow cron to be executed as a goroutine, as its Runtime.Run method has no returns, and errors are channeled via its Runtime.Err method (which returns an error channel). The actual runtime of the cron is still managed with a context.Context that is provided when calling Runtime.Run -- which can impose a cancellation or timeout strategy.

Just like the simple example above, creating a cron runtime starts with the cron.New constructor function.

This function only has a variadic parameter for cfg.Option[cron.Config]. This allows full modularity on the way you build your cron runtime, to be as simple or as detailed as you want it to be -- provided that it complies with the minimum requirements to create one; to supply either:

a selector.Selector
or, a (set of) executor.Runner. This can be supplied as executor.Runnable as well.

func New(options ...cfg.Option[*Config]) (Runtime, error)

Below is a table with all the options available for creating a cron runtime:

Function	Input Parameters	Description
`WithSelector`	`sel selector.Selector`	Configures the with the input `selector.Selector`.
`WithJob`	`id string`, `cron string`, `runners ...executor.Runner`	Adds a new `executor.Executor` to the `Runtime` configuration from the input ID, cron string and set of `executor.Runner`.
`WithErrorBufferSize`	`size int`	Defines the capacity of the error channel that the `Runtime` exposes in its `Runtime.Err` method.
`WithMetrics`	`m cron.Metrics`	Configures the `Runtime` with the input metrics registry.
`WithLogger`	`logger *slog.Logger`	Configures the `Runtime` with the input logger.
`WithLogHandler`	`handler slog.Handler`	Configures the `Runtime` with logging using the input log handler.
`WithTrace`	`tracer trace.Tracer`	Configures the `Runtime` with the input trace.Tracer.

The simplest possible cron runtime could be the result of a call to cron.New with a single cron.WithJob option. This creates all the components that a cron runtime needs with the most minimal setup. It creates the underlying selector and executors.

Otherwise, the caller must use the WithSelector option, and configure a selector.Selector manually when doing so. This results in more boilerplate to get the runtime set up, but provides deeper control on how the cron should be composed. The next chapter covers what is a selector.Selector and how to create one.

Cron Selector

This component is responsible for picking up the next job to execute, according to their schedule frequency. For this, the Selector is configured with a set of executor.Executor, which in turn will expose a Next method. With this information, the Selector cycles through its executor.Executor and picks up the next task(s) to run.

While the Selector calls the executor.Executor's Exec method, the actual waiting is within the executor.Executor's logic.

You're able to create a Selector through its constructor function:

func New(options ...cfg.Option[*Config]) (Selector, error)

Below is a table with all the options available for creating a cron job selector:

Function	Input Parameters	Description
`WithExecutors`	`executors ...executor.Executor`	Configures the `Selector` with the input `executor.Executor`(s).
`WithBlock`		Configures the `Selector` to block (wait) for the underlying `executor.Executor`(s) to complete the task.
`WithTimeout`	`dur time.Duration`	Configures a (non-blocking) `Selector` to wait a certain duration before detaching of the executable task, before continuing to select the next one.
`WithMetrics`	`m selector.Metrics`	Configures the `Selector` with the input metrics registry.
`WithLogger`	`logger *slog.Logger`	Configures the `Selector` with the input logger.
`WithLogHandler`	`handler slog.Handler`	Configures the `Selector` with logging using the input log handler.
`WithTrace`	`tracer trace.Tracer`	Configures the `Selector` with the input trace.Tracer.

There is a catch to the Selector, which is the actual job's execution time. While the Selector cycles through its executor.Executor list, it will execute the task while waiting for it to return with or without an error. This may cause issues when a given running task takes too long to complete when there are other, very frequent tasks. If there is a situation where the long-running task overlaps the execution time for another scheduled job, that job's execution is potentially skipped -- as the next task would only be picked up and waited for once the long-running one exits.

For this reason, there are two implementations of Selector:

A blocking one, that waits for every job to run and return an error, accurately returning the correct outcome in its Next call. This implementation is great for fast and snappy jobs, or less frequent / non-overlapping schedules and executions. There is less resource overhead to it, and the error returns are fully accurate with the actual outcome.
A non-blocking one, that waits for a job to raise an error in a goroutine, with a set timeout (either set by the caller or a default one). This implementation is great if the jobs are too frequent and / or the tasks too long, when it risks skipping executions due to stuck long-running tasks. It relies more heavily on having configured Observability at least on the executor.Executor level to underline those events (which get detached from the Selector after timing out).

It is important to have a good idea of how your cron jobs will execute and how often, or simply ensure that there is at least logging enabled for the configured executor.Executor(s).

Cron Executor

Like the name implies, the Executor is the component that actually executes the job, on its next scheduled time.

The Executor is composed of a cron schedule and a (set of) Runner(s). Also, the Executor stores an ID that is used to identify this particular job.

Having these 3 components in mind, it's natural that the Executor exposes three methods:

Exec - runs the task when on its scheduled time.
Next - calls the underlying schedule.Scheduler Next method.
ID - returns the ID.

Considering that the Executor holds a specific schedule.Scheduler, it is also responsible for managing any waiting time before the job is executed. The strategy employed by the Executable type is one that calculates the duration until the next job, and sleeps until that time is reached (instead of, for example, calling the schedule.Scheduler Next method every second).

To create an Executor, you can use the New function that serves as a constructor. Note that the minimum requirements to creating an Executor are to include both a schedule.Scheduler with the WithScheduler option (or a cron string, using the WithSchedule option), and at least one Runner with the WithRunners option.

The Runner itself is an interface with a single method (Run), that takes in a context.Context and returns an error. If your implementation is so simple that you have it as a function and don't need to create a type for this Runner, then you can use the Runnable type instead, which is a type alias to a function of the same signature, but implements Runner by calling itself as a function, in its Run method.

Creating an Executor is as easy as calling its constructor function:

func New(id string, options ...cfg.Option[*Config]) (Executor, error)

Below is a table with all the options available for creating a cron job executor:

Function	Input Parameters	Description
`WithRunners`	`runners ...Runner`	Configures the `Executor` with the input `Runner`(s).
`WithScheduler`	`sched schedule.Scheduler`	Configures the `Executor` with the input `schedule.Scheduler`.
`WithSchedule`	`cron string`	Configures the `Executor` with a `schedule.Scheduler` using the input cron string.
`WithLocation`	`loc *time.Location`	Configures the `Executor` with a `schedule.Scheduler` using the input `time.Location`.
`WithMetrics`	`m executor.Metrics`	Configures the `Executor` with the input metrics registry.
`WithLogger`	`logger *slog.Logger`	Configures the `Executor` with the input logger.
`WithLogHandler`	`handler slog.Handler`	Configures the `Executor` with logging using the input log handler.
`WithTrace`	`tracer trace.Tracer`	Configures the `Executor` with the input trace.Tracer.

Cron Scheduler

The Scheduler is responsible for keeping schedule state (for example, derived from a cron string), and calculating the next job's execution time, with the context of the input timestamp. As such, the Scheduler interface only exposes one method, Next which is responsible of making such calculations.

The default implementation of Scheduler, CronSchedule, will be created from parsing a cron string, and is nothing but a data structure with a cronlex.Schedule bounded to a time.Location.

While the CronSchedule leverages different schedule elements with cronlex.Resolver interfaces, the Scheduler uses these values as a difference from the input timestamp, to create a new date with a time.Date() call. This call merely adds the difference until the next job to the current time, on different elements of the timestamp, with added logic to calculate weekdays if set.

Fortunately, Go's time package is super solid and allows date overflows, calculating them accordingly. This makes the logic of the base implementation a total breeze, and simple enough to be pulled off as opposed to ticking every second, checking for new jobs.

You're able to create a Scheduler by calling its constructor function, with the mandatory minimum of supplying a cron string through its WithSchedule option.

func New(options ...cfg.Option[Config]) (Scheduler, error)

Below is a table with all the options available for creating a cron job scheduler:

Function	Input Parameters	Description
`WithSchedule`	`cron string`	Configures the `Scheduler` with the input cron string.
`WithLocation`	`loc *time.Location`	Configures the `Scheduler` with the input `time.Location`.
`WithMetrics`	`m executor.Metrics`	Configures the `Scheduler` with the input metrics registry.
`WithLogger`	`logger *slog.Logger`	Configures the `Scheduler` with the input logger.
`WithLogHandler`	`handler slog.Handler`	Configures the `Scheduler` with logging using the input log handler.
`WithTrace`	`tracer trace.Tracer`	Configures the `Scheduler` with the input trace.Tracer.

Cron Schedule

Schedule is a data structure that holds a set of Resolvers, for each node, segment or unit of the schedule. This implementation focuses on the cron string specification with added support for a seconds definition (instead of the usual minutes, hours, days-of-the-month, months and weekdays). Each of these elements are Resolvers interfaces that will calculate the difference until the target value(s) is reached. More information on Resolvers in its own section.

The Schedule only holds the state of a parsed cron string, and its elements are made public so that implementations of Scheduler can leverage it to calculate the job's next execution time.

To create a new Schedule type, you're able to use the Parse function, that consumes the input cron string, returning a Schedule and an error if raised. More details on the actual parsing of the string in its own section.

Once created, the elements of Schedule are accessed directly, where the caller can use the Resolver interface:

type Schedule struct {
	Sec      Resolver
	Min      Resolver
	Hour     Resolver
	DayMonth Resolver
	Month    Resolver
	DayWeek  Resolver
}

Schedule Resolver

This component calculates the difference between the input value and the (set of) value(s) it is configured to be triggered on, also given a certain maximum value for that Resolver's range.

This difference is useful on a Scheduler, where time.Date() sums the input time to the difference until the next execution. As such, given each node in a Schedule, it is possible to derive the next time per node, following this logic.

Take a minute as an example. It is an element that spans from 0 to 59; and consider all elements start at zero. A minutes Resolver is configured with a maximum value of 59, and this example is configured to trigger at minute 15. If the input time is 2:03 PM, the Resolver returns a difference of 12. When the Scheduler gets this information, it adds up the 12 minutes to the input time, and returns the resulting datetime value.

This also makes the Resolver very flexible, in tandem with the corresponding Schedule and Scheduler, providing customizable levels of precision. The seconds Resolver in Schedule is an example of this.

The implementations of Resolver can be found in the resolve package. These are derived from parsing a cron string and assigned automatically when calling the Parse function.

To explore the different implementations, it's good to have in mind the modularity in cron schedules:

it supports "every value" sequences when using a star (*)
it supports range value sequences when using a dash (-, e.g. 0-15)
it supports step value sequences when using a slash (/, e.g. */3)
it supports separated range value elements when using commas (,, e.g. 0,20,25,30)
it supports a combination of range values and steps (e.g. 0-15/3)
it supports overrides, for certain configurations (e.g. @weekly)

Having this in mind, this could technically be achieved with a single Resolver type (that you will find for step values), but to maximize performance and reduce complexity where it is not needed, there are four types of Resolver:

`Everytime` Resolver

type Everytime struct{}

The Everytime Resolver always returns zero, meaning that it resolves to the current time always (trigger now type of action). This is the default Resolver whenever a star (*) node is found, for example.

`FixedSchedule` Resolver

type FixedSchedule struct {
	Max int
	At  int
}

The FixedSchedule Resolver resolves at a specific value within the entire range. It is especially useful when a node only contains a non-star (*) alphanumeric value, such as the cron string 0 0 * * *, which has both minutes and hours as FixedSchedule Resolvers.

`RangeSchedule` Resolver

type RangeSchedule struct {
    Max  int
    From int
    To   int
}

The RangeSchedule Resolver resolves within a portion of the entire range. It is used when a range is provided without any gaps. This means that the schedule element does not contain any slashes (/) or commas (,), and is only a range with a dash (-), such as the cron string 0-15 * * * *, which contains a RangeSchedule Resolver for its minutes' node.

`StepSchedule` Resolver

type StepSchedule struct {
    Max   int
    Steps []int
}

The StepSchedule Resolver is the most complex of all -- that could potentially serve all the other implementations, however it can be the most resource-expensive of all Resolvers.

This implementation stores the defined values for the schedule and returns the difference of the closest value ahead of it. This involves scanning all the steps in the sequence as it requires looking into values that are less than the input, e.g., when the input is value 57, the maximum is 59, and the closest step is 0, with a difference of 3.

Doing this for complex cron setups can be more complex, and that is the major reason for having several other implementations. Regardless, if your cron string is one that involves many individual steps separated by commas (,), or contains a given frequency delimited by a slash (/), it surely will have an underlying StepSchedule Resolver to resolve that / those node(s).

The StepSchedule Resolver is the only Resolvers which exposes a constructor, with the NewStepSchedule function, that takes from and to values, a maximum, and a certain frequency (which should be 1 if no custom frequency is desired). Any further additions to the Steps in the StepSchedule Resolver, should be added to the data structure, separately:

func NewStepSchedule(from, to, maximum, frequency int) StepSchedule

In the Schedule Parser section, we explore how its processor will create the Schedule types following some rules, when working with the abstract syntax tree from parsing the cron string.

Schedule Parser

To consume a cron string and make something meaningful of it, we must parse it. This step is most important to nail down accurately as it is the main source of user input within this library's logic. The executed jobs are of the caller's responsibility as they pass into it whatever they want. But having a correct understanding of the input schedule as well as calculating the times for the jobs' execution is fundamentally most important.

As mentioned before, this package exposes a Parse function that consumes a cron string returning a Schedule and an error:

func Parse(cron string) (s Schedule, err error)

This is a process broken down in three phases that can be explored individually, having into consideration that the lexer and parser components work in tandem:

A lexer, that consumes individual bytes from the cron string, emitting meaningful tokens about what they represent. The tokens also hold the value (bytes) that represent them.
A parser, which consumes the tokens as they are emitted from the lexer, and progressively builds an abstract syntax tree that is the cron schedule and its different nodes.
A processor, which consumes the finalized abstract syntax tree created by the parser, validates its contents and creates the appropriate Schedule.

The implementations of the underlying lexer and parser logic are taken from Go's standard library, from its go/token and text/template packages. There is also a fantastic talk from Rob Pike not only describing how to write a lexer and parser as state machines in Go, but also carrying the viewer through the details of the text/template implementation. With this in mind, I had released the same general concept of a lexer and parser, but supporting generic types, allowing consumers of the library to write their parsers for anything, with any type (both for input, tokens and output). These implementations can be further explored below:

Taking this as a kick-off point, it's only required to implement 3 (major) functions for the lexer, parser and processor phases of this pipeline. These functions are broken down with their own individual handler functions, as required.

Introducing the Token type, we can see how different symbols are set to trigger different tokens, while having a few general ones like TokenAlphaNum for numbers and characters, TokenEOF when the end of the string is reached, etc.

Starting with the lexer, it exposes a StateFunc function. This is the starting point for the state machine that consumes individual bytes. This function basically emits a token for a given character (if supported), except for alphanumeric bytes (that are collected and a single token emitted with them), and at the end of the string.

At the same time that the lexer is emitting these tokens, the parser's ParseFunc function is consuming them to build an abstract syntax tree. This tree has a root (top-level) node that is branched for how many nodes are there in a cron string -- say, * * * * * contains 5 nodes, * * * * * * contains 6 nodes, and @weekly contains 1 node.

If these nodes are more complex than a single value, then the node will store the value while aggregating all following (chained) values by their symbol as children. These symbol nodes must contain a value as child (a comma cannot be left by itself at the end of a node). Take the following node in a cron string: 0-15/3,20,25. The node in the tree should look like:

- (root)
  +- alphanum --> value: 0
    +- dash   --> value: 15
    +- slash  --> value: 3
    +- comma  --> value: 20
    +- comma  --> value: 25

Having created the cron's abstract syntax tree we arrive to the last phase, the ProcessFunc function. It starts off by validating the contents in the abstract syntax tree to ensure there are no unsupported values like greater than the maximum, etc.

Once ensured it is valid, the function checks how many nodes are children of the root node in the tree, with support for 3 types of lengths:

1 child node means this should be an exception (like @weekly).
5 child nodes represent a classic cron string ranging from minutes to weekdays (e.g. * * * * *).
6 child nodes represent an extended cron string supporting seconds to weekdays (e.g. * * * * * *).

Handling the exceptions is very simple as the function only switches on the supported values looking for a match. The switch statement is the fastest algorithm to perform this check.

A classic cron string with 5 nodes will still have a seconds Resolver in its Schedule, by configuring it as a FixedSchedule type, triggering at value 0 with a maximum of 59 (for seconds).

Generally, the Resolver types for each node from both classic and extended cron strings are built by checking if it's a star (*) or alphanumeric node, creating the appropriate Resolver. Note that step values are sorted and compacted before being returned, for optimal efficiency when being used.

Lastly, considering the weekday support for 0 and 7 as Sundays, if the weekday Resolver is a StepSchedule type, it is normalized as a 0 value.

Example

A working example is the Steam CLI app mentioned in the Motivation section above. This application exposes some commands, one of them being monitor. This file provides some insight on how the cron service is set up from a main.go / script-like approach.

You can also take a look at its runner.go file, that implements the executor.Runner interface.

Disclaimer

This is not a one-size-fits-all solution! Please take your time to evaluate it for your own needs with due diligence. While having a library for this and a library for that is pretty nice, it could potentially be only overhead hindering the true potential of your app! Be sure to read the code that you are using to be a better judge if it is a good fit for your project. With that in mind, I hope you enjoy this library. Feel free to contribute by filing either an issue or a pull request.

Name		Name	Last commit message	Last commit date
Latest commit History 68 Commits
executor		executor
log		log
metrics		metrics
schedule		schedule
selector		selector
tracing		tracing
.gitignore		.gitignore
.golangci.yml		.golangci.yml
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
cron.go		cron.go
cron_config.go		cron_config.go
cron_integration_test.go		cron_integration_test.go
cron_test.go		cron_test.go
cron_with_logs.go		cron_with_logs.go
cron_with_metrics.go		cron_with_metrics.go
cron_with_trace.go		cron_with_trace.go
go.mod		go.mod
go.sum		go.sum
tools.go		tools.go

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

micron

This is my cron (micron); there are others like it but this one is mine. A cron-scheduler library in Go

Concept

Motivation

Usage

Getting `cron`

Cron Runtime

Cron Selector

Cron Executor

Cron Scheduler

Cron Schedule

Schedule Resolver

`Everytime` Resolver

`FixedSchedule` Resolver

`RangeSchedule` Resolver

`StepSchedule` Resolver

Schedule Parser

Example

Disclaimer

About

Releases 4

Packages

Languages

License

zalgonoise/micron

Folders and files

Latest commit

History

Repository files navigation

micron

This is my cron (micron); there are others like it but this one is mine. A cron-scheduler library in Go

Concept

Motivation

Usage

Getting cron

Cron Runtime

Cron Selector

Cron Executor

Cron Scheduler

Cron Schedule

Schedule Resolver

Everytime Resolver

FixedSchedule Resolver

RangeSchedule Resolver

StepSchedule Resolver

Schedule Parser

Example

Disclaimer

About

Resources

License

Stars

Watchers

Forks

Releases 4

Packages 0

Languages

Getting `cron`

`Everytime` Resolver

`FixedSchedule` Resolver

`RangeSchedule` Resolver

`StepSchedule` Resolver

Packages