# Changesets Reference

## cast vs put_change vs force_change

| Function | Use When |
|----------|----------|
| `cast/4` | External data (user input, API) |
| `put_change/3` | Internal trusted data (timestamps, computed) |
| `change/2` | Internal data from existing struct |
| `force_change/3` | When you need to set even if value unchanged |

```elixir
# External data - use cast
def registration_changeset(user, attrs) do
  user
  |> cast(attrs, [:email, :password, :name])
  |> validate_required([:email, :password, :name])
  |> validate_email()
  |> hash_password()
end

# Internal data - use put_change
defp hash_password(changeset) do
  case changeset do
    %{valid?: true, changes: %{password: password}} ->
      put_change(changeset, :hashed_password, Bcrypt.hash_pwd_salt(password))
    _ ->
      changeset
  end
end
```

## Multiple Changesets per Schema

```elixir
# Different changesets for different operations
def registration_changeset(user, attrs) do
  user
  |> cast(attrs, [:email, :password, :name])
  |> validate_required([:email, :password, :name])
  |> validate_email()
  |> validate_length(:password, min: 12, max: 72)
  |> hash_password()
end

def profile_changeset(user, attrs) do
  user
  |> cast(attrs, [:name, :bio, :avatar])
  |> validate_length(:bio, max: 500)
end

def password_changeset(user, attrs) do
  user
  |> cast(attrs, [:password])
  |> validate_required([:password])
  |> validate_length(:password, min: 12, max: 72)
  |> hash_password()
end

def admin_changeset(user, attrs) do
  user
  |> cast(attrs, [:role, :permissions])
  |> validate_inclusion(:role, [:user, :moderator, :admin])
end
```

## Custom Validations

```elixir
def changeset(order, attrs) do
  order
  |> cast(attrs, [:quantity, :unit_price])
  |> validate_required([:quantity, :unit_price])
  |> validate_positive_total()
end

defp validate_positive_total(changeset) do
  validate_change(changeset, :quantity, fn :quantity, quantity ->
    unit_price = get_field(changeset, :unit_price) || 0

    if quantity * unit_price < 0 do
      [quantity: "total must be positive"]
    else
      []
    end
  end)
end
```

## prepare_changes for Transaction-Safe Operations

```elixir
def changeset(post, attrs) do
  post
  |> cast(attrs, [:title, :body, :published_at])
  |> prepare_changes(fn changeset ->
    # Runs inside the transaction
    if get_change(changeset, :published_at) do
      changeset.repo.update_all(
        from(p in Post, where: p.author_id == ^post.author_id),
        inc: [post_count: 1]
      )
    end
    changeset
  end)
end
```

## Embedded Schemas

**Use embedded_schema when:**

- Never query child independently
- Never share child across parents
- Always loaded with parent (single query)

```elixir
# Embedded schema (stored as JSONB)
defmodule MyApp.Accounts.Profile do
  use Ecto.Schema
  import Ecto.Changeset

  @primary_key false
  embedded_schema do
    field :dark_mode, :boolean, default: false
    field :timezone, :string
    field :language, :string, default: "en"
  end

  def changeset(profile, attrs) do
    profile
    |> cast(attrs, [:dark_mode, :timezone, :language])
  end
end

# Parent schema
defmodule MyApp.Accounts.User do
  schema "users" do
    field :email, :string
    embeds_one :profile, Profile, on_replace: :update
    embeds_many :addresses, Address, on_replace: :delete
  end

  def changeset(user, attrs) do
    user
    |> cast(attrs, [:email])
    |> cast_embed(:profile)
    |> cast_embed(:addresses)
  end
end
```

Migration:

```elixir
add :profile, :map
add :addresses, :map, default: "[]"
```

## Field Types

| Need | Ecto Type | PostgreSQL | Notes |
|------|-----------|------------|-------|
| Primary key | `:binary_id` | `uuid` | Prefer UUIDs |
| Text | `:string` | `varchar` | Default |
| Long text | `:text` | `text` | No limit |
| Integer | `:integer` | `integer` | |
| Money | `:integer` | `integer` | Store cents (never float!) |
| Decimal | `:decimal` | `numeric` | Precise calculations |
| Boolean | `:boolean` | `boolean` | |
| Date | `:date` | `date` | |
| DateTime | `:utc_datetime_usec` | `timestamptz` | With timezone + microseconds |
| JSON | `:map` | `jsonb` | Use for dynamic |
| Enum | `Ecto.Enum` | `varchar` | Type-safe |
| Array | `{:array, :string}` | `varchar[]` | |

# PostgreSQL Full-Text Search with Ecto

Native full-text search without external dependencies. Based on
[Search is Not Magic with PostgreSQL](https://www.codecon.sk/search-is-not-magic-with-postgresql).

## Strategy Decision Tree

| Need | Strategy | Extension |
|------|----------|-----------|
| Exact/weighted text search | Full-text search (tsvector) | Built-in |
| Typo tolerance / fuzzy | Trigram similarity (pg_trgm) | `pg_trgm` |
| Semantic / AI search | Vector search (pgvector) | `pgvector` |
| All of the above | Hybrid with RRF | Multiple |

## 1. Full-Text Search (tsvector/tsquery)

### Migration — Generated Column (Preferred, PostgreSQL 12+)

```elixir
def up do
  execute """
  ALTER TABLE articles
    ADD COLUMN searchable tsvector
    GENERATED ALWAYS AS (
      setweight(to_tsvector('english', coalesce(title, '')), 'A') ||
      setweight(to_tsvector('english', coalesce(body, '')), 'B')
    ) STORED;
  """

  create index(:articles, [:searchable], using: :gin)
end

def down do
  alter table(:articles) do
    remove :searchable
  end
end
```

Generated columns auto-update on INSERT/UPDATE — no triggers needed.

**When to use triggers instead**: When the tsvector depends on associated
records (e.g., tags from a join table). Generated columns can only reference
columns in the same row.

### Basic Search Query

```elixir
def search_articles(query_string) do
  from(a in Article,
    where: fragment(
      "searchable @@ websearch_to_tsquery('english', ?)",
      ^query_string
    ),
    order_by: [desc: fragment(
      "ts_rank_cd(searchable, websearch_to_tsquery('english', ?), 32)",
      ^query_string
    )]
  )
  |> Repo.all()
end
```

### Search with Highlights and Pagination

```elixir
def search_articles(query_string, opts \\ []) do
  page = Keyword.get(opts, :page, 1)
  per_page = Keyword.get(opts, :per_page, 20)

  from(a in Article,
    where: fragment("searchable @@ websearch_to_tsquery('english', ?)", ^query_string),
    select: %{
      id: a.id,
      title: a.title,
      headline: fragment(
        "ts_headline('english', ?, websearch_to_tsquery('english', ?), 'StartSel=<mark>, StopSel=</mark>')",
        a.body, ^query_string
      ),
      rank: fragment("ts_rank_cd(searchable, websearch_to_tsquery('english', ?), 32)", ^query_string)
    },
    order_by: [desc: fragment("ts_rank_cd(searchable, websearch_to_tsquery('english', ?), 32)", ^query_string)],
    offset: ^((page - 1) * per_page),
    limit: ^per_page
  )
  |> Repo.all()
end
```

### Multi-Language Support

```elixir
# Dynamic language via regconfig casting
from(a in Article,
  where: fragment(
    "to_tsvector(?::text::regconfig, ?) @@ to_tsquery(?::text::regconfig, ?)",
    ^language, a.body, ^language, ^query
  )
)
```

### websearch_to_tsquery Syntax (Google-style)

| Input | Matches |
|-------|---------|
| `elixir phoenix` | Both words |
| `"exact phrase"` | Exact phrase |
| `elixir OR phoenix` | Either word |
| `-deprecated` | Excludes word |

### Weight Meanings

| Weight | Use | Boost |
|--------|-----|-------|
| A | Title | Highest |
| B | Subtitles | High |
| C | Body | Medium |
| D | Metadata | Lower |

## 2. Trigram Similarity (pg_trgm) — Fuzzy/Typo Tolerance

```elixir
# Migration
execute "CREATE EXTENSION IF NOT EXISTS pg_trgm"
execute "CREATE INDEX products_name_trgm_idx ON products USING gin(name gin_trgm_ops)"

# Query
from(p in Product,
  where: fragment("similarity(?, ?) > ?", p.name, ^term, 0.3),
  order_by: [desc: fragment("similarity(?, ?)", p.name, ^term)]
)
```

Trigrams compare 3-character groups — handles typos, misspellings, partial matches.
Threshold 0.3 is a good default; tune based on your data.

## 3. Hybrid Search with RRF (Reciprocal Rank Fusion)

Combine multiple search strategies by normalizing ranks:

```elixir
# Each strategy returns %{id, rank} with: 1 / (60 + distance)
similarity_results = search_by_trigram(term)
fulltext_results = search_by_tsvector(term)

# Merge with deduplication
(similarity_results ++ fulltext_results)
|> Enum.group_by(& &1.id)
|> Enum.map(fn {id, ranks} -> {id, Enum.sum(Enum.map(ranks, & &1.rank))} end)
|> Enum.sort_by(&elem(&1, 1), :desc)
```

### Multi-Word Query Normalization

```elixir
defp normalize_query(text) do
  text
  |> String.trim()
  |> String.downcase()
  |> String.replace(~r/\s+/, " ")
  |> String.split(" ")
  |> Enum.join(" & ")
end
```

## Performance

```elixir
# ALWAYS use GIN index for tsvector
create index(:articles, [:searchable], using: :gin)

# Partial index for large tables
create index(:articles, [:searchable], using: :gin, where: "published_at IS NOT NULL")

# GIN indexes only used with LIMIT — always paginate
```

## Anti-patterns

```elixir
# WRONG: Computing tsvector at query time (slow, no index!)
from(a in Article,
  where: fragment("to_tsvector('english', title || ' ' || body) @@ to_tsquery(?)", ^q)
)

# WRONG: Using LIKE for search (no ranking, no stemming)
from(a in Article, where: ilike(a.title, ^"%#{query}%"))

# WRONG: Using triggers when generated columns suffice
# Generated columns are simpler and auto-maintained

# WRONG: Assuming PG can't do fuzzy search
# pg_trgm handles typo tolerance natively — no need for Elasticsearch just for fuzzy
```

## When to Use External Search

PostgreSQL handles most use cases (100K-10M docs). Consider Meilisearch/Elasticsearch for:

- Faceted search with complex filters across many dimensions
- Multi-language with mixed alphabets in same field
- Real-time indexing of 10M+ documents
- Advanced search analytics

**Further reading**: [Search is Not Magic with PostgreSQL](https://www.codecon.sk/search-is-not-magic-with-postgresql)
— covers trigrams, full-text, vector search, and hybrid patterns with Ecto examples.

# Migrations Reference

## Basic Migration

```elixir
defmodule MyApp.Repo.Migrations.CreateUsers do
  use Ecto.Migration

  def change do
    create table(:users, primary_key: false) do
      add :id, :binary_id, primary_key: true
      add :email, :string, null: false
      add :name, :string
      add :role, :string, default: "user"

      timestamps(type: :utc_datetime_usec)
    end

    create unique_index(:users, [:email])
  end
end
```

## Add Foreign Key (Safe - 2 steps)

```elixir
# Step 1: Add without validation (fast, no table lock)
def change do
  alter table(:posts) do
    add :user_id, references(:users, type: :binary_id, validate: false), null: false
  end

  create index(:posts, [:user_id])
end

# Step 2: Validate in separate migration (separate deploy)
def change do
  execute "ALTER TABLE posts VALIDATE CONSTRAINT posts_user_id_fkey", ""
end
```

## Add NOT NULL (Safe - 3 steps)

```elixir
# Step 1: Add check constraint without validation
def change do
  create constraint("products", :active_not_null,
    check: "active IS NOT NULL",
    validate: false
  )
end

# Step 2: Backfill data (separate deploy)
def change do
  execute "UPDATE products SET active = false WHERE active IS NULL", ""
end

# Step 3: Validate constraint, add NOT NULL, drop constraint
def change do
  execute "ALTER TABLE products VALIDATE CONSTRAINT active_not_null", ""
  execute "ALTER TABLE products ALTER COLUMN active SET NOT NULL", ""
  drop constraint("products", :active_not_null)
end
```

## Pre-Migration Safety: Check Data BEFORE Unique Index

A unique index on a table with existing duplicates fails at deploy time —
in the worst case mid-release. Check FIRST, and include soft-deleted rows:
they're invisible in the app but still block the index.

```sql
-- Find duplicates the index would reject (run before writing the migration)
SELECT email, COUNT(*) FROM users
GROUP BY email HAVING COUNT(*) > 1;

-- Soft-deleted rows count too — check whether they collide
SELECT email, COUNT(*) FILTER (WHERE deleted_at IS NOT NULL) AS deleted,
       COUNT(*) FILTER (WHERE deleted_at IS NULL) AS live
FROM users GROUP BY email HAVING COUNT(*) > 1;
```

Resolutions, in order of preference:

1. **Partial index** when soft-deleted rows may legitimately collide:
   `create unique_index(:users, [:email], where: "deleted_at IS NULL")`
2. **Data-fix migration first** (separate deploy), then the index
3. **Composite key** if the duplicates are actually valid scoping
   (e.g., unique per tenant: `unique_index(:users, [:org_id, :email])`)

Same discipline applies before adding NOT NULL (check for NULLs) and
before foreign keys (check for orphaned rows).

## Concurrent Index (Large Tables)

```elixir
@disable_ddl_transaction true
@disable_migration_lock true

def change do
  create index(:posts, [:slug], concurrently: true)
end
```

## Batched Data Migration

```elixir
def change do
  # For large tables, process in batches
  execute &migrate_data/0, &rollback_data/0
end

defp migrate_data do
  repo().transaction(fn ->
    from(u in "users", where: is_nil(u.status), select: u.id, limit: 1000)
    |> repo().all()
    |> Enum.each(&update_user_status/1)
  end)
end
```

## Mixed Primary Key Types (bigint + binary_id)

When integrating libraries that use UUID PKs (Sagents, Oban, etc.)
while your project uses bigint:

```elixir
# WRONG: Global @foreign_key_type affects ALL associations
@foreign_key_type :binary_id

# RIGHT: Explicit type ONLY on specific associations
schema "interviews" do
  belongs_to :user, MyApp.Accounts.User        # bigint (default)
  belongs_to :conversation, Agents.Conversation, type: :binary_id
end
```

In migrations, match the referenced table's PK type:

```elixir
alter table(:interviews) do
  add :user_id, references(:users, type: :bigint), null: false
  add :conversation_id, references(:conversations, type: :binary_id)
end
```

## Associations

```elixir
# One-to-many (always specify on_delete!)
has_many :posts, Post, on_delete: :delete_all
belongs_to :user, User

# Many-to-many
many_to_many :tags, Tag, join_through: "post_tags", on_replace: :delete

# Has one through
has_one :organization, through: [:user, :organization]

# Self-referential
belongs_to :parent, __MODULE__
has_many :children, __MODULE__, foreign_key: :parent_id
```

## Optimistic Locking

```elixir
schema "products" do
  field :name, :string
  field :lock_version, :integer, default: 1
end

def changeset(product, attrs) do
  product
  |> cast(attrs, [:name])
  |> optimistic_lock(:lock_version)
end

# Usage - raises Ecto.StaleEntryError if version changed
Repo.update!(changeset)
```

# Queries Reference

## Composable Query Functions

```elixir
defmodule MyApp.Posts.PostQuery do
  import Ecto.Query

  def base, do: from(p in Post, as: :post)

  def published(query \\ base()) do
    from p in query, where: not is_nil(p.published_at)
  end

  def by_author(query \\ base(), author_id) do
    from p in query, where: p.author_id == ^author_id
  end

  def recent(query \\ base(), days \\ 7) do
    cutoff = Date.add(Date.utc_today(), -days)
    from p in query, where: p.inserted_at >= ^cutoff
  end

  def ordered(query \\ base(), direction \\ :desc) do
    from p in query, order_by: [{^direction, p.inserted_at}]
  end
end

# Usage: Pipeline composition
PostQuery.base()
|> PostQuery.published()
|> PostQuery.by_author(author_id)
|> PostQuery.ordered()
|> Repo.all()
```

## Dynamic Queries

```elixir
def filter_where(params) do
  Enum.reduce(params, dynamic(true), fn
    {"author", value}, dynamic ->
      dynamic([p], ^dynamic and p.author == ^value)

    {"category", value}, dynamic ->
      dynamic([p], ^dynamic and p.category == ^value)

    {"title_contains", value}, dynamic ->
      dynamic([p], ^dynamic and ilike(p.title, ^"%#{value}%"))

    {_, _}, dynamic ->
      dynamic  # Ignore unknown params
  end)
end

# Usage
def list_posts(params) do
  Post
  |> where(^filter_where(params))
  |> Repo.all()
end
```

## Subqueries

```elixir
# Correlated subquery with parent_as
comment_count = from c in Comment,
  where: parent_as(:post).id == c.post_id,
  select: count()

from p in Post, as: :post,
  select: %{title: p.title, comment_count: subquery(comment_count)}
```

## Window Functions

```elixir
from p in Post,
  select: %{
    id: p.id,
    title: p.title,
    row_num: row_number() |> over(partition_by: p.category_id, order_by: p.inserted_at),
    rank: rank() |> over(partition_by: p.category_id, order_by: [desc: p.view_count])
  }
```

## JSONB Queries (Ecto 3.12+)

### json_extract_path/2

Extract values from JSONB columns without raw SQL:

```elixir
# Extract nested JSON value
from u in User,
  where: json_extract_path(u.settings, ["notifications", "email"]) == true,
  select: json_extract_path(u.metadata, ["theme"])

# Compare with older fragment approach (still works but verbose)
from u in User,
  where: fragment("?->>'email' = ?", u.settings["notifications"], "true")
```

### JSONB Pattern: Filtering on Nested Keys

```elixir
# Dynamic key extraction
def by_metadata_key(query, key, value) do
  from u in query,
    where: json_extract_path(u.metadata, ^[key]) == ^value
end

# Array access in JSONB
from p in Product,
  where: json_extract_path(p.attributes, ["tags", 0]) == "featured"
```

### JSONB Anti-patterns

```elixir
# WRONG: Loading all rows then filtering in Elixir
Repo.all(User)
|> Enum.filter(fn u -> u.settings["notifications"]["email"] == true end)

# RIGHT: Filter in database
from u in User,
  where: json_extract_path(u.settings, ["notifications", "email"]) == true

# TIP: For frequently queried JSONB paths, create expression index
# In migration:
# execute "CREATE INDEX users_settings_notifications_idx ON users ((settings->'notifications'))"
```

## Repo.reload Options (Ecto 3.12+)

```elixir
# Basic reload (unchanged)
user = Repo.reload!(user)

# With preloads (new in 3.12)
user = Repo.reload!(user, preload: [:posts, :comments])

# Force fresh query (skip query cache)
user = Repo.reload!(user, force: true)
```

## Full-Text Search

For PostgreSQL full-text search patterns, see
`fulltext-search.md`.

## Preload Strategies

```elixir
# Separate queries (default) - BEST for has_many
# Two queries: posts + comments
Repo.preload(post, :comments)

# Join (single query) - BEST for belongs_to/has_one
# One query with JOIN - watch for row multiplication with has_many!
from(p in Post, preload: [:author])

# Custom query for filtered/ordered preloads
Repo.preload(post, comments: from(c in Comment, order_by: c.inserted_at, limit: 10))
```

## Pagination

```elixir
def list_posts(params \\ %{}) do
  page = Map.get(params, :page, 1)
  per_page = Map.get(params, :per_page, 20)

  from(p in Post,
    order_by: [desc: p.inserted_at],
    offset: ^((page - 1) * per_page),
    limit: ^per_page
  )
  |> Repo.all()
end
```

## Anti-patterns

```elixir
# WRONG: N+1 queries
users = Repo.all(User)
Enum.map(users, fn u -> u.posts end)  # N queries!

# RIGHT: Preload
users = Repo.all(User) |> Repo.preload(:posts)

# WRONG: Getting all then filtering in Elixir
Repo.all(User) |> Enum.filter(& &1.active)

# RIGHT: Filter in query
from(u in User, where: u.active) |> Repo.all()

# WRONG: String interpolation (SQL injection!)
from(u in User, where: fragment("name = '#{name}'"))

# RIGHT: Parameterized queries
from(u in User, where: u.name == ^name)

# WRONG: Using Repo.get! with user input (may raise)
Repo.get!(User, user_provided_id)

# RIGHT: Handle not found
case Repo.get(User, id) do
  nil -> {:error, :not_found}
  user -> {:ok, user}
end
```

# Transactions Reference

## Repo.transact (Simpler for most cases)

```elixir
Repo.transact(fn ->
  with {:ok, user} <- create_user(params),
       {:ok, profile} <- create_profile(user) do
    {:ok, user}
  end
end)
```

## Repo.transaction/1: changeset error handling

Inside the callback form `Repo.transaction(fn -> ... end)`, a bare `{:ok, _} =`
match on a write is a footgun. When the write returns `{:error, changeset}`, the
match raises `MatchError`; the transaction rolls back and **re-raises** it, so the
caller crashes (a `500` in a request) and the changeset's validation errors are
lost. It does *not* return `{:error, %MatchError{}}` — nothing catches it for you.

```elixir
# BAD — invalid changeset raises MatchError, which propagates and crashes the caller
Repo.transaction(fn ->
  {:ok, updated} = Repo.update(changeset)
  updated
end)

# GOOD — explicit case + Repo.rollback/1 preserves the changeset
Repo.transaction(fn ->
  case Repo.update(changeset) do
    {:ok, updated} -> updated
    {:error, changeset} -> Repo.rollback(changeset)
  end
end)
# => {:error, %Ecto.Changeset{}} — caller gets actionable validation errors
```

The same applies to `Repo.insert/1` and `Repo.delete/1`. For multiple steps, use
`with ... else` and roll back on the error path:

```elixir
Repo.transaction(fn ->
  with {:ok, user}    <- Repo.insert(user_changeset),
       {:ok, profile} <- Repo.insert(profile_changeset(user)) do
    profile
  else
    {:error, changeset} -> Repo.rollback(changeset)
  end
end)
```

Prefer `Repo.transact/1` (above) or `Ecto.Multi` (below) when you can — both
surface `{:error, ...}` on failure without a manual `Repo.rollback/1`, steering
you away from the bare match. This footgun is specific to the classic
`Repo.transaction/1` callback form. (Ties to the plugin Iron Laws on checking
changeset errors and matching `{:error, %Ecto.Changeset{}}` explicitly.)

## Ecto.Multi (Complex operations, testing)

```elixir
alias Ecto.Multi

# Basic Multi
Multi.new()
|> Multi.insert(:user, user_changeset)
|> Multi.insert(:profile, fn %{user: user} ->
  Profile.changeset(%Profile{user_id: user.id}, %{})
end)
|> Multi.run(:welcome_email, fn _repo, %{user: user} ->
  MyApp.Mailer.deliver_welcome(user)
end)
|> Repo.transaction()

# Composition with merge
Multi.new()
|> Multi.insert(:order, order_changeset)
|> Multi.merge(fn %{order: order} ->
  Multi.new()
  |> Multi.insert_all(:items, OrderItem, build_items(order))
end)
|> Repo.transaction()

# Reusable Multi components (default to Multi.new())
def transfer_money(multi \\ Multi.new(), from_id, to_id, amount) do
  multi
  |> Multi.run(:validate, fn _, _ -> validate_accounts(from_id, to_id) end)
  |> Multi.update(:debit, fn _ -> debit_changeset(from_id, amount) end)
  |> Multi.update(:credit, fn _ -> credit_changeset(to_id, amount) end)
end

# Error handling
case Repo.transaction(multi) do
  {:ok, %{user: user, team: team}} ->
    {:ok, user}
  {:error, :user, changeset, _changes} ->
    {:error, :user_creation_failed, changeset}
  {:error, :team, changeset, _changes} ->
    {:error, :team_creation_failed, changeset}
end

# Testing Multi without DB
multi = PasswordManager.reset(account, params)
assert [{:account, {:update, changeset, []}}] = Ecto.Multi.to_list(multi)
```

## Upsert Patterns

```elixir
# Insert or update on conflict
Repo.insert(
  changeset,
  on_conflict: {:replace, [:name, :updated_at]},
  conflict_target: :external_id
)

# Insert or do nothing
Repo.insert(changeset, on_conflict: :nothing, conflict_target: :email)

# Insert all with upsert
Repo.insert_all(
  Post,
  posts,
  on_conflict: {:replace_all_except, [:id, :inserted_at]},
  conflict_target: :external_id
)
```

## Batch Operations

```elixir
# insert_all (fast bulk insert)
Repo.insert_all(Post, posts, returning: [:id])

# update_all (fast bulk update)
from(p in Post, where: p.status == :draft)
|> Repo.update_all(set: [status: :archived, updated_at: DateTime.utc_now()])

# delete_all (fast bulk delete)
from(p in Post, where: p.inserted_at < ^cutoff)
|> Repo.delete_all()
```

## Streaming

```elixir
# For processing large result sets without loading all into memory
Repo.transaction(fn ->
  from(p in Post)
  |> Repo.stream()
  |> Stream.each(&process_post/1)
  |> Stream.run()
end)
```

## Connection Pool Tuning

```elixir
# config/runtime.exs
config :my_app, MyApp.Repo,
  pool_size: String.to_integer(System.get_env("POOL_SIZE") || "10"),
  queue_target: 50,
  queue_interval: 1000
```

Rule of thumb: `pool_size = (CPU cores * 2) + disk spindles`