generate uuid v7 #4681

dkuku · 2025-11-19T19:37:50Z

Add uuid v7 generator.
Also the formatter updated the files - I think it's because it was not touched for long time.
example uuid generated by some random page from google:
019a9da1-53b4-7f77-9609-140ad8880bd2
and one generated with ecto:

iex(2)> Ecto.UUID.generate(:v7)
"019a9da1-568e-7981-9bc3-4ba057de8778"

greg-rychlewski · 2025-11-20T10:41:03Z

I wonder if we want to have a 2 arity function to let users control the sub millisecond monotonicity guarantee. For example from what I know Postgres and Maria DB use 12 extra timestamp bits to include the nano seconds. Or maybe we want to consider just using nanoseconds in the 1 arity function to be more inline with these databases

josevalim · 2025-11-20T13:02:41Z

lib/ecto/uuid.ex

  """
  @spec generate() :: t
-  def generate(), do: encode(bingenerate())
+  def generate(), do: encode(bingenerate(@default_version))


Suggested change

def generate(), do: encode(bingenerate(@default_version))

def generate(version \\ @default_version), do: encode(bingenerate(version))

And similar below.

josevalim · 2025-11-20T13:05:29Z

lib/ecto/schema.ex

      section for more info). Primary keys are automatically set up for embedded
      schemas as well, defaulting to `{:id,  :binary_id, autogenerate: true}`.
+      This will generate the default UUID v4. You can use UUID v7 instead by setting
+      the primary key to `{:id, :binary_id, autogenerate: {Ecto.UUID, :generate, [:v7]}}`


Let's also document this in Ecto.UUID's module doc.

josevalim · 2025-11-20T13:19:25Z

@greg-rychlewski in such cases, we can provide a :v7_nano in the future?

lib/ecto/uuid.ex

warmwaffles · 2025-11-20T17:13:05Z

I know that @ryanwinchester implemented / improved uuid v7 for :uniq, so his input may be valuable here.

ryanwinchester · 2025-11-20T17:36:27Z

I know that @ryanwinchester implemented / improved uuid v7 for :uniq, so his input may be valuable here.

In uniq it's a naive implementation of just ms timestamps, before the RFC had section 6.2.

I wonder if we want to have a 2 arity function to let users control the sub millisecond monotonicity guarantee. For example from what I know Postgres and Maria DB use 12 extra timestamp bits to include the nano seconds. Or maybe we want to consider just using nanoseconds in the 1 arity function to be more inline with these databases

Yes. In postgres at least, it's monotonic and stepped by a certain minimum number of bits as per section 6.2, method 3. I created a separate UUIDv7 library that does this as well.

I think if we're adding UUIDv7 to Ecto, it would be useful if we did this to match the way the most used database(s) do it. Or at least the option to.

josevalim · 2025-11-20T17:46:35Z

I think if we're adding UUIDv7 to Ecto, it would be useful if we did this to match the way the most used database(s) do it. Or at least the option to.

@ryanwinchester I agree. The question is if we do it by default or opt-in. Do you have any thoughts on the matter?

In any case, I think we can merge this now and you could contribute your nano implementation next. Would that work for you? Or would prefer for it to be baked in right now?

ryanwinchester · 2025-11-20T17:56:12Z

I think if we're adding UUIDv7 to Ecto, it would be useful if we did this to match the way the most used database(s) do it. Or at least the option to.

@ryanwinchester I agree. The question is if we do it by default or opt-in. Do you have any thoughts on the matter?

Personally, I would prefer it to be default as it is in Postgres, but I don't know what most people would prefer.

In any case, I think we can merge this now and you could contribute your nano implementation next. Would that work for you? Or would prefer for it to be baked in right now?

Maybe instead of version \\ @default_version, we could make it opts \\ [] with a :version field and merge it, leaving it more open to changing without breaking changes?

i.e. we could have version: 7 (or version: :v7) with only millisecond precision be the default, and the increased precision version could still be the same :version, but with other options to specify which method of extra precision you prefer. Just a thought

(there are 3 optional methods for increased precision + monotonicity in the RFC, and I can contribute method 3)

josevalim · 2025-11-20T17:58:38Z

Maybe instead of version \ @default_version, we could make it opts \ [] with a :version field and merge it, but leaving it more open to changing without breaking changes?

Yes, regardless of what we do here, I think options would be better indeed, thank you. @dkuku, can you please adjust accordingly?

Also, can you please go ahead and change Ecto.Schema to make it so that, if autogenerate: list_of_options is given, it automatically becomes {mod, :autogenerate, [list_of_options]}? This should keep it flexible and ergonomic to everyone.

dkuku · 2025-11-20T19:32:37Z

Also, can you please go ahead and change Ecto.Schema to make it so that, if autogenerate: list_of_options is given, it automatically becomes {mod, :autogenerate, [list_of_options]}? This should keep it flexible and ergonomic to everyone.

It started to work with type instead of mod, I also added a new callback but I'm not sure if it's necessary ?

dkuku · 2025-11-20T19:54:06Z

lib/ecto/type.ex

  to `field` with the `:autogenerate` flag.
  """
  @callback autogenerate() :: term()
+  @callback autogenerate(term) :: term()


it could be a Keyword but I see the functions in this module mostly operate on terms so I keep it consistent

We can skip this change I think... since theoretically with the MFA you can call it anything or even have higher arity?

ryanwinchester · 2025-11-21T00:26:30Z

lib/ecto/uuid.ex

  @spec generate() :: t
-  def generate(), do: encode(bingenerate())
+  @spec generate(options) :: t
+  def generate(opts \\ [@default_options]), do: encode(bingenerate(opts))


probably don't need the @default_options and could just use an empty list since you supply a default version with Keyword.get/3

ryanwinchester · 2025-11-21T00:31:11Z

lib/ecto/uuid.ex

  @spec bingenerate() :: raw
-  def bingenerate() do
+  def bingenerate(), do: bingenerate(@default_options)
+
+  @doc """
+  Generates a uuid with the given options in binary format.
+  """
+  @spec bingenerate(options) :: raw
+  def bingenerate(opts) do
+    case Keyword.get(opts, :version, @default_version) do
+      4 -> bingenerate_v4()
+      7 -> bingenerate_v7()
+      _ -> raise ArgumentError, "unknown UUID version: #{inspect(opts[:version])}"
+    end
+  end


same idea as generate, we don't need the default options

Suggested change

@spec bingenerate() :: raw

def bingenerate() do

def bingenerate(), do: bingenerate(@default_options)

@doc """

Generates a uuid with the given options in binary format.

"""

@spec bingenerate(options) :: raw

def bingenerate(opts) do

case Keyword.get(opts, :version, @default_version) do

4 -> bingenerate_v4()

7 -> bingenerate_v7()

_ -> raise ArgumentError, "unknown UUID version: #{inspect(opts[:version])}"

end

end

@doc """

Generates a uuid with the given options in binary format.

"""

@spec bingenerate(options) :: raw

def bingenerate(opts \\ []) do

case Keyword.get(opts, :version, @default_version) do

4 -> bingenerate_v4()

7 -> bingenerate_v7()

version -> raise ArgumentError, "unknown UUID version: #{inspect(version)}"

end

end

josevalim

Please address @ryanwinchester's feedback and we can ship it!

Thank you all! ❤️

josevalim · 2025-11-21T14:51:57Z

💚 💙 💜 💛 ❤️

warmwaffles · 2025-11-21T16:05:40Z

Thank you @ryanwinchester and @dkuku, looking forward to moving to ecto for generating uuidv7 instead of relying on uniq

dkuku added 2 commits November 19, 2025 20:33

generate uuid v7

c76f398

update schema docs

6245180

josevalim reviewed Nov 20, 2025

View reviewed changes

lib/ecto/uuid.ex Show resolved Hide resolved

cr suggestions

04e34b8

dkuku requested a review from josevalim November 20, 2025 16:33

typo

d47aa86

revert to system_time

c3977d3

josevalim approved these changes Nov 20, 2025

View reviewed changes

dkuku added 2 commits November 20, 2025 20:27

use options for uuid type

475af88

allow autogenerate to acccept a list of options

3f2a6ef

dkuku added 2 commits November 20, 2025 20:44

restore formatting

8563e5b

update docs

5479ee2

dkuku commented Nov 20, 2025

View reviewed changes

update type in uuid

39de72d

dkuku requested review from josevalim and ryanwinchester November 20, 2025 20:00

ryanwinchester reviewed Nov 21, 2025

View reviewed changes

Remove additional autogenerate callback

5e5ab80

josevalim reviewed Nov 21, 2025

View reviewed changes

cr suggestions

8dd203a

josevalim merged commit e6da70d into elixir-ecto:master Nov 21, 2025
7 checks passed

ryanwinchester mentioned this pull request Nov 23, 2025

UUIDv7 monotonicity and increased clock precision #4683

Merged

	def generate(), do: encode(bingenerate(@default_version))
	def generate(version \\ @default_version), do: encode(bingenerate(version))

generate uuid v7 #4681

generate uuid v7 #4681

Conversation

dkuku commented Nov 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

greg-rychlewski commented Nov 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

josevalim Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

josevalim Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

josevalim commented Nov 20, 2025

Uh oh!

Uh oh!

warmwaffles commented Nov 20, 2025

Uh oh!

ryanwinchester commented Nov 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

josevalim commented Nov 20, 2025

Uh oh!

ryanwinchester commented Nov 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

josevalim commented Nov 20, 2025

Uh oh!

dkuku commented Nov 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dkuku Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

josevalim Nov 21, 2025

Choose a reason for hiding this comment

Uh oh!

ryanwinchester Nov 21, 2025

Choose a reason for hiding this comment

Uh oh!

ryanwinchester Nov 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

josevalim left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

josevalim commented Nov 21, 2025

Uh oh!

warmwaffles commented Nov 21, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

dkuku commented Nov 19, 2025 •

edited

Loading

greg-rychlewski commented Nov 20, 2025 •

edited

Loading

ryanwinchester commented Nov 20, 2025 •

edited

Loading

ryanwinchester commented Nov 20, 2025 •

edited

Loading

dkuku commented Nov 20, 2025 •

edited

Loading

ryanwinchester Nov 21, 2025 •

edited

Loading