Hanami 1.3 -> Hanami 2.0

Question
#1

I’ve started working on upgrading our projects from Hanami 1.3 → 2.0 and I’ll document my journey as I progress.

  1. This post will be updated as the time goes
  2. When (Not IF :sweat_smile: ) I’ll encounter some blockers, I’ll ask here for help in this thread.
  3. At the end I’ll compose a blog post (and guides chapter?) out of our findings here.

Any contribution to discussions appreciated :slight_smile:

Upgrade Steps

  1. Get Rid of Hanami Model → Switch to ROM 5.0
  2. Upgrade Ruby 2.7 → 3.1
  3. Upgrade all DRY Libraries
  4. Update Hanami → Hanami 2

Please let me know if you have more ideas, or need further help!

2 Likes
#2

My first trouble:

I have JSONB columns in my entities. Before, reading them by Hanami::Model, resulted in a hash. After switching to ROM, the column read from DB returns JSON.

# Before (Hanami::Model)
message.payload
# => { author: "Jsmith", content: "My message" }

# After (ROM 5.0)

# frozen_string_literal: true

module Entities
  class Message < ROM::Struct
  end
end

# frozen_string_literal: true

module Persistence
  module Relations
    class Messages < ROM::Relation[:sql]
      schema(:messages, infer: true)

      auto_struct true
    end
  end
end

class MessageRepository < Rom::Repository[:messages]
end

MessageRepository.new.messages.first.payload
# =>  "{\"author\":\"Jsmith\",\"content\":\"My message\"}"

Question: How I should configure a Mapper for all JSONB columns in my system at once?

2 Likes
#3

Enable the pg_json extension from Sequel.

I’m not using Hanami yet but my dry-system app has this:

rom_config = ROM::Configuration.new(
  :sql,
  target[:settings].database_url,
  extensions: %i[pg_array pg_json]
)

Nothing specific needs to be done within ROM itself, in my experience.

1 Like
#4

Thanks! Unfortunately did not work :(. Still when I read the payload, I get JSON instead Hash

#5

Here’s an example script that works for me

#!/usr/bin/env ruby

require 'bundler/inline'

gemfile(true) do
  source 'https://rubygems.org'

  gem 'pg'
  gem 'sequel_pg', require: 'sequel'
  gem 'rom', '~> 5'
  gem 'rom-sql'
end

require 'open3'

o, e, s = Open3.capture3("psql -c 'CREATE DATABASE rom_example'")
s.success? ? puts(o) : abort(e)

o, e, s = Open3.capture3('psql rom_example', stdin_data: <<~SQL)
  CREATE schema public;

  CREATE TABLE public.messages (
    id bigint NOT NULL,
    payload jsonb NOT NULL default '{}'
  );
SQL

s.success? ? puts(o) : abort(e)

require 'sequel'
require 'rom'
require 'rom/sql'

ROM::SQL.load_extensions :postgres

config = ROM::Configuration.new(:sql, "postgres://localhost:5432/rom_example", extensions: %i[pg_json])

config.relation(:messages) do
  schema(infer: true)
  auto_struct true
end

rom = ROM.container(config)
relation = rom.relations[:messages]

attrs = { id: 1, payload: { author: "John Smith", content: "My message" } }
relation.changeset(:create, attrs).commit

message = relation.where(id: 1).one
puts "MESSAGE PAYLOAD: #{message.payload.inspect}"

rom.gateways[:default].disconnect
system "psql -c 'drop database rom_example'"

$ ./example.rb
Fetching gem metadata from https://rubygems.org/.......
Resolving dependencies...
Using bundler 2.3.3
Using dry-initializer 3.1.1
Using method_source 1.0.0
Using pg 1.4.3
Using transproc 1.1.1
Using sequel 5.59.0
Using coderay 1.1.3
Using dry-inflector 0.3.0
Using ice_nine 0.11.2
Using concurrent-ruby 1.1.10
Using dry-equalizer 0.3.0
Using pry 0.14.1
Using sequel_pg 1.15.0
Using dry-core 0.8.1
Using dry-container 0.10.1
Using dry-logic 1.2.0
Using dry-types 1.5.1
Using dry-struct 1.4.0
Using rom-core 5.2.6
Using rom-changeset 5.2.3
Using rom-repository 5.2.2
Using rom 5.2.6
Using rom-sql 3.5.0
Expanded display is used automatically.
Null display is "¤".
CREATE DATABASE
Expanded display is used automatically.
Null display is "¤".
CREATE TABLE
MESSAGE PAYLOAD: {"author"=>"John Smith", "content"=>"My message"}
Expanded display is used automatically.
Null display is "¤".
DROP DATABASE
1 Like
#6

@alassek Thank you! It actually worked! It appears, I loaded extensions and plugins wrongly.

I experienced more problems, as the Sequel::Postgres::JSONHash returns stringified keys, while the rest of attributes symbolized and it can’t be stringified via: Hanami::Utils::Hash.deep_stymbolize(attrs)

I’ve handled that. However, I got stucked with other advanced query magic with Sequel and ROM mix :(.

OLD Code from Hanami 1 (ROM 3.X)

# foos_by_users
before = Date.today.to_time.strftime('%Y-%m-%dT%H:%M:%S.%LZ')

foos
  .select { `data->>'user_id'`.as(:user_id) }
  .select_append { `MAX(data->>'featured_at')`.as(:featured_at) }
  .where(Sequel.lit("data->>'featured_at' <= ?", before))
  .group(:user_id)
  .order(nil)
  .pluck(%i(user_id featured_at))
  .to_h

Question: How to translate this to ROM 5?

Running above returns the error:

KeyError:
 :user_id attribute doesn't exist in foos schema

In new ROM syntax, select blocks should contain array instead of single objects.

foos
  .select do
    [
      `data->>'user_id'`.as(:user_id)
      `MAX(data->>'featured_at')`.as(:featured_at)
    ]
  end
   # ...

But still, .group(:user_id) does not see the attribute.

I’ve tried also:

.group(`data->>'user_id'`)

but this returns the undefined method error

     NoMethodError:
       undefined method `qualified_projection' for [:user_id, :featured_at]:Array

This looks like the pluck happens before grouping, or whatever.

cc: @solnic

#7

I have a particular interest in SQL ASTs so this was an amusing diversion, although the software engineer in me wonders if this might be simpler to just implement as a database view.

This is complex so I’m going to take this a piece at a time.

SQL literals

There’s nothing wrong with using literals this way but there are Ruby interfaces for what you’re doing.

.select { data.pg_jsonb.get_text('user_id').as(:user_id) }
.select_append { function(:MAX, data.pg_jsonb.get_text('featured_at')).as(:featured_at) }
.where { data.pg_jsonb.get_text('featured_at') <= Date.today.to_time }

The pg_jsonb interface is less than ideal but it gets the job done.

GROUP BY

Method parameters are schema attributes, blocks are schema attributes and virtual columns

.group { user_id }

Pluck

Pluck is implemented as select + map. You’re already selecting so just try map here

.map(%i[user_id featured_at])
#8

@swilgosz Relation#pluck is supposed to be used to return values from specific columns, it doesn’t support projected columns. You can probably skip pluck and just do .to_hash(:user_id, :featured_at).