ent

Simple data entities for Go

Ent, short for "entities", is data persistence for Go, using plain Go structs.

Features:

• No extra "description" files needed: just add tags to your Go structs.

• Automatically versioned ents.

• Automatic unique and non-unique secondary indexes (e.g. look up an account by email) including compound indexes.

• Transactional edits — a change either fully succeeds or not at all.

• Multiple storage backends with a small API for using custom storage.

• CouchDB-like get-modify-put — when "putting" it back, if the ent has changed by someone else since you loaded it, there will be a version conflict error and no changes will be made.

• Uses code generation instead of reflection to minimize magic — read the go fmt-formatted generated code to understand exactly what is happening.

• Generated code is included in your documentation (e.g. via go doc)

Tutorial

Ent uses Go structs. By adding ent.EntBase as the first embedded field in a struct, you have made it into an ent! You can find a complete example of this code in examples/tutorial.

Let's start by defining an "Account" struct type:

//go:generate entgen
package main

import "github.com/rsms/ent"

type AccountKind int32

const (
  AccountRestricted = AccountKind(iota)
  AccountMember
  AccountAdmin
)

type Account struct {
  ent.EntBase `account` // type name, used to store these kinds of ents
  name        string
  displayName string      `ent:"alias"`   // use a different field name for storage
  email       string      `ent:",unique"` // maintain a unique index for this field
  kind        AccountKind `ent:",index"`  // maintain a (non-unique) index for this field
}

When this file changes we should run entgen (or go generate if we have a //go:generate comment in the file.) Doing so causes a few methods, functions and data to be generated for the Account type, which makes it a fully-functional data entity.

You can either run entgen from source or go modules, or install it with go get github.com/rsms/ent/entgen. See #entgen for more details.

We can now create, load, query, update and delete accounts. But let's start by making one and just printing it:

func main() {
  a := &Account{
    name:  "Jane",
    alias: "j-town",
    email: "[email protected]",
    kind:  AccountMember,
  }
  println(a.String())
}

If we build & run this we should see the following output:

{
  _ver:  "0",
  _id:   "0",
  name:  "Jane",
  alias: "j-town",
  email: "[email protected]",
  kind:  1
}

Notice a couple of things:

• The String method returns a JSON-like representation

• There are two implicit fields: _ver and _id. These represent the current version and id of an ent, respectively. These values can be accessed via the Version() and Id() methods. A value of 0 (zero) means "not yet assigned".

• The displayName field is called alias; renamed by the ent field tag.

• Field order matches our struct definition.

Now let's store this account in a database. This is really what ent is about — data persistence. We start this example by creating a place to store ents, a storage. Here we use an in-memory storage implementation mem.EntStorage but there are other kinds, like Redis.

import "github.com/rsms/ent/mem"

  // add to our main function:
  estore := mem.NewEntStorage()
  if err := a.Create(estore); err != nil {
    panic(err)
  }
  println(a.String())

Our account is now saved. The output from the last print statement now contains non-zero version & id:

{
  _ver:  "1",
  _id:   "1",
  name:  "Jane",
  ...

Another part of the program can load the ent:

  a, _ = LoadAccountById(estore, 1)
  fmt.Printf("account #1: %v\n", a) // { _ver: "1", _id: "1", name: "Jane", ...

Notice that we started using Go's "fmt" package. If you are following along, import "fmt".

Since we specified unique on the email field, we can look up ents by email in addition to id:

  b, _ := LoadAccountByEmail(estore, "[email protected]")
  fmt.Printf(`account with email "[email protected]": %v\n`, b)
  // { _ver: "1", _id: "1", name: "Jane", ...

  _, err := LoadAccountByEmail(estore, "[email protected]")
  fmt.Printf("error from lookup of non-existing email: %v\n", err)

If we just need to check if an ent exists, or we only need to know the id, we can use the Find... functions instead of the Load... functions:

  id, _ := FindAccountByEmail(estore, "[email protected]")
  fmt.Printf(`id of account with email "[email protected]": %v\n`, id) // 1

These functions were generated for us by entgen. The Find...ByFIELD and Load...ByFIELD functions performs a lookup on a secondary index ("email" in the example above.)

In our struct definition we declared that we wanted the kind field to be indexed, which means there are also functions for looking up accounts by kind. Indexes which are not unique, i.e. indexes declared with the "index" field tag rather than the "unique" tag, returns a list of ents. To make this example more interesting, let's create a few more ents to play with:

  (&Account{email: "[email protected]", kind: AccountMember}).Create(estore)
  (&Account{email: "[email protected]", kind: AccountAdmin}).Create(estore)
  (&Account{email: "[email protected]", kind: AccountRestricted}).Create(estore)

And let's try querying for different kinds of users:

  accounts, _ := LoadAccountByKind(estore, AccountMember, 0)
  fmt.Printf("'member' accounts: %+v\n", accounts)

  accounts, _ = LoadAccountByKind(estore, AccountAdmin, 0)
  fmt.Printf("'admin' accounts: %+v\n", accounts)

We should see "Jane" and "robin" listed for AccountMember and "thor" for AccountAdmin.

Non-unique indexes as we just explored does not imply any constraints on ents. But unique indexes do — it's kind of the whole point with a unique index :-) When we create or update an ent with a change to a unique index we may get an error in case there is a conflict. For example, let's try creating a new account that uses the same email address as Jane's account:

  err = (&Account{email: "[email protected]"}).Create(estore)
  fmt.Printf("error (duplicate email): %v\n", err)
  // unique index conflict account.email with ent #1

The same would happen if we tried to update an account to use an already-used email value:

  a, _ = LoadAccountByEmail(estore, "[email protected]")
  a.SetEmail("[email protected]")
  fmt.Printf("error (duplicate email): %v\n", a.Save())
  // unique index conflict account.email with ent #1

However if we change the email of Jane's account, we can reuse Jane's old email address:

  a, _ = LoadAccountById(estore, 1)
  a.SetEmail("[email protected]")
  a.Save()

  a, _ = LoadAccountByEmail(estore, "[email protected]")
  a.SetEmail("[email protected]")
  fmt.Printf("no error: %v\n", a.Save())

The ent system maintains these indexes automatically and updates them in a transactional manner: a Create or Save call either fully succeeds, including index changes, or has no effect at all. This promise is declared by the ent system but actually fulfilled by the particular storage used. Both of the storage implementations that comes with ent are fully transactional (mem and redis.)

Changes to ents are tracked with versioning. Every update to an ent increments its version. The version is used when updating an ent:

When we say "make X changes to ent Y of version Z" the storage...

• Checks the current version against the expected version Z.

• If the ent's version is Z in the storage then the changes are applied and its version is incremented to Z+1.

• However if someone else made a change to the ent after we loaded it, the version in storage won't match Z and we get a ErrVersionConflict error.

Example of a version conflict:

  a1, _ := LoadAccountById(estore, 1)
  a2, _ := LoadAccountById(estore, 1)
  // make a change to copy a1 and save it
  a1.SetName("Jenn")
  a1.Save()
  // make a change to copy a2 and save it
  a2.SetName("Jeannie")
  fmt.Printf("version conflict error: %v\n", a2.Save())

To resolve a conflict we either need to discard our change to a2 or load the current version and reapply our changes. If simply replacing values is not what we want, we could load a second copy and merge our new values with the current ones. In this example we simply retry our edit on the most recent version:

  a2.Reload() // load msot current values from storage
  a2.SetName("Jeannie")
  fmt.Printf("save now works (no error): %v\n", a2.Save())

In some ways this approach resembles "compare and swap" memory operations:

atomically:
  if currentValue is expectedValue:
    setValue(newValue)

This versioning approach was inspired by CouchDB.

entgen

entgen is a program that parses go packages and generates ent code for all ent-enabled struct types.

You can either run entgen from source or go modules, or install it with go get github.com/rsms/ent/entgen. To install a specific version run go get github.com/rsms/ent/[email protected] replacing the "X"es with the version you want.

Synopsis

Usage: entgen [options] [<srcdir> ...]
options:
  -debug
      Debug logging (implies -v)
  -entpkg string
      Import path of ent package (default "github.com/rsms/ent")
  -filter string
      Only process go struct types which name matches the provided
      regular expression
  -h, -help
      Show help and exit
  -nofmt
      Disable "gofmt" formatting of generated code
  -o string
      Filename of generated go code, relative to <srcdir>.
      Use "-" for stdout. (default "ents.gen.go")
  -v
      Verbose logging
  -version
      Print "entgen 0.1.0" and exit