I recently discovered a foot gun when writing rails migrations.

Rails runs migrations inside a transaction by default, for those databases that support it (e.g. Postgres). It also provides a what to disable it if you so choose, by using disable_ddl_transaction!. That can be useful for example for creating a large index concurrently, which is not supported inside a transaction. It looks like this:

class FootGun < ActiveRecord::Migration[7.2]
  disable_ddl_transaction!

  def change
    create_table :foot_guns
  end
end

So far, so good. However, because of how disable_ddl_transaction! is implemented, there is also a disable_ddl_transaction method defined. That is an accessor that checks weather the migration should be run in a transaction or not. But it can be used by mistake:

class FootGun < ActiveRecord::Migration[7.2]
  disable_ddl_transaction # This doesn't do anything!!!

  def change
    create_table :foot_guns
  end
end

The migrations looks like it is disabling the transaction, but it’s actually not. It’ also a hard mistake to catch, because the output rails prints out when running the migration in both cases is the same:

$ rails db:migrate
== 20241116193728 FootGun: migrating ==========================================
-- create_table(:foot_guns)
   -> 0.0137s
== 20241116193728 FootGun: migrated (0.0158s) =================================

I’d love for disable_ddl_transaction not to exist at all, so that a NameError would be raised, and this mistake was impossible to make.

Waiting for PostgreSQL 18 – Add temporal PRIMARY KEY and UNIQUE constraints

In this article, and a follow up we learn about upcoming changes to Postgres 18 that will make temporal modeling much easier. A welcome change. Maybe soon after that we can get libraries to leverage it in popular web frameworks.

Rightward assiggment in Ruby

It’s now possible to use rightward (->) assignment in Ruby. The tweet talks about using it in “pipelines”:

rand(100)
  .then { _1 * 2 }
  .then { _1 -3 } => value

value # => 7

I am very fond of pipelines like that, but feel that the => is not very visible. What I want to write is:

rand(100)
  .then { _1 * 2 }
  .then { _1 -3 }
  => value

But that doesn’t work, because the parser balks. I can use a \, but that makes it awkward:

rand(100)
  .then { _1 * 2 }
  .then { _1 -3 } \
  => value

value # => 87

Goodhart’s Law Isn’t as Useful as You Might Think

when a measure becomes a target, it ceases to be a good measure

Long dive into concepts from operations research that go deeper than the pithy “law” and explain the mechanisms at play.

Often, I want to add a new column to a Postgres table with a default value for new records, but also want existing records to have a different value. Changing Postgres default value can make this a very fast operation.

Let’s see an example. Let’s assume we have a songs table, and we want to add a liked column. Existing records need to have the value be set to false, while new values have it set to true.

Table and initial data setup:

CREATE TABLE songs (
  name character varying NOT NULL
);
-- CREATE TABLE
-- Time: 16.084 ms

INSERT INTO songs(name) VALUES ('Stairway To Heaven');
-- INSERT 0 1
-- Time: 0.590 ms

SELECT * FROM songs;
--         name
-- --------------------
--  Stairway To Heaven
-- (1 row)
--
-- Time: 0.652 ms

Now, let’s add the new column with a default value of false. That is not our end-goal, but it will add that value to existing records¹:

ALTER TABLE songs
ADD COLUMN liked boolean DEFAULT false;
-- ALTER TABLE
-- Time: 3.745 ms

SELECT * FROM songs;
--         name        | liked
-- --------------------+-------
--  Stairway To Heaven | f
-- (1 row)
--
-- Time: 0.672 ms

ALTER TABLE songs
ALTER COLUMN liked SET NOT NULL;
-- ALTER TABLE
-- Time: 1.108 ms

Now, if we change the default value to true, and insert a new record:

ALTER TABLE songs ALTER COLUMN liked SET DEFAULT true;
-- ALTER TABLE
-- Time: 4.664 ms

INSERT INTO songs(name) VALUES ('Hotel California');
-- INSERT 0 1
-- Time: 1.447 ms

SELECT * FROM songs;
--
--         name        | liked
-- --------------------+-------
--  Stairway To Heaven | f
--  Hotel California   | t
-- (2 rows)
--
-- Time: 0.791 ms

As we can see, we have the schema in the shape that we want, and the correct data stored in it, without needing a “traditional” backfill to modify each existing row manually. The default value method is much faster, since Postgres doesn’t need to update each record, just check the default value when they were created. 👍🏻

OAuth from First Principles

The articles explains how the problem of sharing access between servers evolves into OAuth, as one starts trying to solve the security issues in the naive implementation.

The “email is authentication” pattern

For some people, logging into a website means using the “forgot your password” flow every time they want to log in. They do in lieu of other schemes like using the same password, using a password manager, using a password generation scheme, etc.

Are people well informed about the options? From the website’s perspective it doesn’t matter much: Essentially, having access to an email address grants you access to the website. As long as that is the case, we might as well use “magic links” for authentication and do away with passwords all together.

In fact, in many places, email is now also used as 2-factor authentication. If the website has a “forgot my password” flow via email, then 2-factor via email only adds the illusion of security.

Solid Queue 1.0 released

I’m happy about this development: Rails should definitely have a canonical queue implementation. I’m also interested in it’s performance because of the UPDATE FOR SKIP LOCKED usage. I plan on evaluating it in in the future vs GoodJob. I noticed a few things about the announcement:

37 Signals production setup uses claims 20M jobs per day with 800 workers. That seems like a lot of workers, but without the context of what those workers are doing.

They are using a separate database for the queue. While I get that it alleviates some of the performance concerns with the main database, you also loose transactionality between your jobs and the rest of your writes: To me, transactionality is one of the main selling points of using a db-based queueing system. I’ve chased many production issue where using a separate data store for the jobs ends up causing the queue workers to look for records that are not visible, either temporarily due to a race condition, or permanently because of a roll-back. Using 2 separate databases also means that each Rails process (web or worker) needs a connection to each database.

In the announcement there is a link to a Postgres-only issue recently fixed, that made me realize that Solid Queue has concurrency controls built-in, and uses INSERT ON CONFLICT DO NOTHING to enforce them. That is clever, and more efficient than checking for existence of the concurrency key before inserting.

I recently learned to leverage expect. According to the man pages it is:

expect - programmed dialogue with interactive programs

I often need to ssh into remote hosts that both are ephemeral and I don’t have much control over. Any environment customization that I do is one-time only, because odds are that next time I connect, it will be to a different host. Yet, I’d like to be in a familiar environment. That is where expect comes in.

expect provides a DSL of sorts to interactively use programs in the command line. I use it to front ssh and do a few common tasks every time I log in:

#!/usr/bin/env expect
# Front ssh and automate env settings once on remote machine

set environment [lindex $argv 0];

spawn ssh $environment
expect "\$ " {
  send ". entry.sh"

  send -- "alias db_replica='psql ..."
  send -- "\n"

  send -- "export EDITOR=nano"
  send -- "\n"

  expect "\$ " { interact }
}

That script:

1. Reads the first argument to it and sets to a local variable named “environment”.
2. Starts an instance (spawn) of ssh, passing the value of environment to it.
3. It then waits (expect) until it reads a $
4. It sends a few commands to source a file, create an alias, export a variable, etc.
5. It then expects another prompt and starts and switches to interactive mode (interact)

That last steps is what “drops” me into the terminal in the remote host, personalized with my taste.

This is just the proverbial tip of the iceberg. The expect manual has a lot more information.