defmodule Ecto.Migrator do
@moduledoc """
Lower level API for managing migrations.
EctoSQL provides three mix tasks for running and managing migrations:
* `mix ecto.migrate` - migrates a repository
* `mix ecto.rollback` - rolls back a particular migration
* `mix ecto.migrations` - shows all migrations and their status
Those tasks are built on top of the functions in this module.
While the tasks above cover most use cases, it may be necessary
from time to time to jump into the lower level API. For example,
if you are assembling an Elixir release, Mix is not available,
so this module provides a nice complement to still migrate your
system.
To learn more about migrations in general, see `Ecto.Migration`.
## Example: Running an individual migration
Imagine you have this migration:
defmodule MyApp.MigrationExample do
use Ecto.Migration
def up do
execute "CREATE TABLE users(id serial PRIMARY_KEY, username text)"
end
def down do
execute "DROP TABLE users"
end
end
You can execute it manually with:
Ecto.Migrator.up(Repo, 20080906120000, MyApp.MigrationExample)
## Example: Running migrations in a release
Elixir v1.9 introduces `mix release`, which generates a self-contained
directory that consists of your application code, all of its dependencies,
plus the whole Erlang Virtual Machine (VM) and runtime.
When a release is assembled, Mix is no longer available inside a release
and therefore none of the Mix tasks. Users may still need a mechanism to
migrate their databases. This can be achieved with using the `Ecto.Migrator`
module:
defmodule MyApp.Release do
@app :my_app
def migrate do
for repo <- repos() do
{:ok, _, _} = Ecto.Migrator.with_repo(repo, &Ecto.Migrator.run(&1, :up, all: true))
end
end
def rollback(repo, version) do
{:ok, _, _} = Ecto.Migrator.with_repo(repo, &Ecto.Migrator.run(&1, :down, to: version))
end
defp repos do
Application.load(@app)
Application.fetch_env!(@app, :ecto_repos)
end
end
The example above uses `with_repo/3` to make sure the repository is
started and then runs all migrations up or a given migration down.
Note you will have to replace `MyApp` and `:my_app` on the first two
lines by your actual application name. Once the file above is added
to your application, you can assemble a new release and invoke the
commands above in the release root like this:
$ bin/my_app eval "MyApp.Release.migrate"
$ bin/my_app eval "MyApp.Release.rollback(MyApp.Repo, 20190417140000)"
## Example: Running migrations on application startup
Add the following to the top of your application children spec:
{Ecto.Migrator,
repos: Application.fetch_env!(:my_app, :ecto_repos),
skip: System.get_env("SKIP_MIGRATIONS") == "true"}
To skip migrations you can also pass `skip: true` or as in the example
set the environment variable `SKIP_MIGRATIONS` to a truthy value.
And all other options described in `up/4` are allowed,
for example if you want to log the SQL commands,
and run migrations in a prefix:
{Ecto.Migrator,
repos: Application.fetch_env!(:my_app, :ecto_repos),
log_migrator_sql: true,
prefix: "my_app"}
To roll back you'd do it normally:
$ mix ecto.rollback
"""
require Logger
require Ecto.Query
alias Ecto.Migration.Runner
alias Ecto.Migration.SchemaMigration
@doc """
Ensures the repo is started to perform migration operations.
All of the application required to run the repo will be started
before hand with chosen mode. If the repo has not yet been started,
it is manually started, with a `:pool_size` of 2, before the given
function is executed, and the repo is then terminated. If the repo
was already started, then the function is directly executed, without
terminating the repo afterwards.
Although this function was designed to start repositories for running
migrations, it can be used by any code, Mix task, or release tooling
that needs to briefly start a repository to perform a certain operation
and then terminate.
The repo may also configure a `:start_apps_before_migration` option
which is a list of applications to be started before the migration
runs.
It returns `{:ok, fun_return, apps}`, with all apps that have been
started, or `{:error, term}`.
## Options
* `:pool_size` - The pool size to start the repo for migrations.
Defaults to 2.
* `:mode` - The mode to start all applications.
Defaults to `:permanent`.
## Examples
{:ok, _, _} =
Ecto.Migrator.with_repo(repo, fn repo ->
Ecto.Migrator.run(repo, :up, all: true)
end)
"""
def with_repo(repo, fun, opts \\ []) do
config = repo.config()
mode = Keyword.get(opts, :mode, :permanent)
apps = [:ecto_sql | config[:start_apps_before_migration] || []]
extra_started =
Enum.flat_map(apps, fn app ->
{:ok, started} = Application.ensure_all_started(app, mode)
started
end)
{:ok, repo_started} = repo.__adapter__().ensure_all_started(config, mode)
started = extra_started ++ repo_started
pool_size = Keyword.get(opts, :pool_size, 2)
migration_repo = config[:migration_repo] || repo
case ensure_repo_started(repo, pool_size) do
{:ok, repo_after} ->
case ensure_migration_repo_started(migration_repo, repo) do
{:ok, migration_repo_after} ->
try do
{:ok, fun.(repo), started}
after
after_action(repo, repo_after)
after_action(migration_repo, migration_repo_after)
end
{:error, _} = error ->
after_action(repo, repo_after)
error
end
{:error, _} = error ->
error
end
end
@doc """
Gets the migrations path from a repository.
This function accepts an optional second parameter to customize the
migrations directory. This can be used to specify a custom migrations
path.
"""
@spec migrations_path(Ecto.Repo.t(), String.t()) :: String.t()
def migrations_path(repo, directory \\ "migrations") do
config = repo.config()
priv = config[:priv] || "priv/#{repo |> Module.split() |> List.last() |> Macro.underscore()}"
app = Keyword.fetch!(config, :otp_app)
Application.app_dir(app, Path.join(priv, directory))
end
@doc """
Gets all migrated versions.
This function ensures the migration table exists
if no table has been defined yet.
## Options
* `:prefix` - the prefix to run the migrations on
* `:dynamic_repo` - the name of the Repo supervisor process.
See `c:Ecto.Repo.put_dynamic_repo/1`.
* `:skip_table_creation` - skips any attempt to create the migration table
Useful for situations where user needs to check migrations but has
insufficient permissions to create the table. Note that migrations
commands may fail if this is set to true. Defaults to `false`. Accepts a
boolean.
"""
@spec migrated_versions(Ecto.Repo.t(), Keyword.t()) :: [integer]
def migrated_versions(repo, opts \\ []) do
lock_for_migrations(true, repo, opts, fn _config, versions -> versions end)
end
@doc """
Runs an up migration on the given repository.
## Options
* `:log` - the level to use for logging of migration instructions.
Defaults to `:info`. Can be any of `Logger.level/0` values or a boolean.
If `false`, it also avoids logging messages from the database.
* `:log_migrations_sql` - the level to use for logging of SQL commands
generated by migrations. Can be any of the `Logger.level/0` values
or a boolean. If `false`, logging is disabled. If `true`, uses the configured
Repo logger level. Defaults to `false`
* `:log_migrator_sql` - the level to use for logging of SQL commands emitted
by the migrator, such as transactions, locks, etc. Can be any of the `Logger.level/0`
values or a boolean. If `false`, logging is disabled. If `true`, uses the configured
Repo logger level. Defaults to `false`
* `:prefix` - the prefix to run the migrations on
* `:dynamic_repo` - the name of the Repo supervisor process.
See `c:Ecto.Repo.put_dynamic_repo/1`.
* `:strict_version_order` - abort when applying a migration with old timestamp
(otherwise it emits a warning)
"""
@spec up(Ecto.Repo.t(), integer, module, Keyword.t()) :: :ok | :already_up
def up(repo, version, module, opts \\ []) do
conditional_lock_for_migrations(module, version, repo, opts, fn config, versions ->
if version in versions do
:already_up
else
result = do_up(repo, config, version, module, opts)
if version != Enum.max([version | versions]) do
latest = Enum.max(versions)
message = """
You are running migration #{version} but an older \
migration with version #{latest} has already run.
This can be an issue if you have already ran #{latest} in production \
because a new deployment may migrate #{version} but a rollback command \
would revert #{latest} instead of #{version}.
If this can be an issue, we recommend to rollback #{version} and change \
it to a version later than #{latest}.
"""
if opts[:strict_version_order] do
raise Ecto.MigrationError, message
else
Logger.warning(message)
end
end
result
end
end)
end
defp do_up(repo, config, version, module, opts) do
async_migrate_maybe_in_transaction(repo, config, version, module, :up, opts, fn ->
attempt(repo, config, version, module, :forward, :up, :up, opts) ||
attempt(repo, config, version, module, :forward, :change, :up, opts) ||
{:error,
Ecto.MigrationError.exception(
"#{inspect(module)} does not implement a `up/0` or `change/0` function"
)}
end)
end
@doc """
Runs a down migration on the given repository.
## Options
* `:log` - the level to use for logging of migration commands. Defaults to `:info`.
Can be any of `Logger.level/0` values or a boolean.
* `:log_migrations_sql` - the level to use for logging of SQL commands
generated by migrations. Can be any of the `Logger.level/0` values
or a boolean. If `false`, logging is disabled. If `true`, uses the configured
Repo logger level. Defaults to `false`
* `:log_migrator_sql` - the level to use for logging of SQL commands emitted
by the migrator, such as transactions, locks, etc. Can be any of the `Logger.level/0`
values or a boolean. If `false`, logging is disabled. If `true`, uses the configured
Repo logger level. Defaults to `false`
* `:prefix` - the prefix to run the migrations on
* `:dynamic_repo` - the name of the Repo supervisor process.
See `c:Ecto.Repo.put_dynamic_repo/1`.
"""
@spec down(Ecto.Repo.t(), integer, module) :: :ok | :already_down
def down(repo, version, module, opts \\ []) do
conditional_lock_for_migrations(module, version, repo, opts, fn config, versions ->
if version in versions do
do_down(repo, config, version, module, opts)
else
:already_down
end
end)
end
defp do_down(repo, config, version, module, opts) do
async_migrate_maybe_in_transaction(repo, config, version, module, :down, opts, fn ->
attempt(repo, config, version, module, :forward, :down, :down, opts) ||
attempt(repo, config, version, module, :backward, :change, :down, opts) ||
{:error,
Ecto.MigrationError.exception(
"#{inspect(module)} does not implement a `down/0` or `change/0` function"
)}
end)
end
defp async_migrate_maybe_in_transaction(repo, config, version, module, direction, opts, fun) do
dynamic_repo = repo.get_dynamic_repo()
fun_with_status = fn ->
result = fun.()
apply(SchemaMigration, direction, [repo, config, version, opts])
result
end
fn -> run_maybe_in_transaction(repo, dynamic_repo, module, fun_with_status, opts) end
|> Task.async()
|> Task.await(:infinity)
end
defp run_maybe_in_transaction(repo, dynamic_repo, module, fun, opts) do
repo.put_dynamic_repo(dynamic_repo)
if module.__migration__()[:disable_ddl_transaction] ||
not repo.__adapter__().supports_ddl_transaction?() do
fun.()
else
{:ok, result} = repo.transaction(fun, log: migrator_log(opts), timeout: :infinity)
result
end
catch
kind, reason ->
{kind, reason, __STACKTRACE__}
end
defp attempt(repo, config, version, module, direction, operation, reference, opts) do
if Code.ensure_loaded?(module) and function_exported?(module, operation, 0) do
Runner.run(repo, config, version, module, direction, operation, reference, opts)
:ok
end
end
@doc """
Runs migrations for the given repository.
Equivalent to:
Ecto.Migrator.run(repo, [Ecto.Migrator.migrations_path(repo)], direction, opts)
See `run/4` for more information.
"""
@spec run(Ecto.Repo.t(), atom, Keyword.t()) :: [integer]
def run(repo, direction, opts) do
run(repo, [migrations_path(repo)], direction, opts)
end
@doc ~S"""
Apply migrations to a repository with a given strategy.
The second argument identifies where the migrations are sourced from.
A binary representing directory (or a list of binaries representing
directories) may be passed, in which case we will load all files
following the "#{VERSION}_#{NAME}.exs" schema. The `migration_source`
may also be a list of tuples that identify the version number and
migration modules to be run, for example:
Ecto.Migrator.run(Repo, [{0, MyApp.Migration1}, {1, MyApp.Migration2}, ...], :up, opts)
A strategy (which is one of `:all`, `:step`, `:to`, or `:to_exclusive`) must be given as
an option.
## Execution model
In order to run migrations, at least two database connections are
necessary. One is used to lock the "schema_migrations" table and
the other one to effectively run the migrations. This allows multiple
nodes to run migrations at the same time, but guarantee that only one
of them will effectively migrate the database.
A downside of this approach is that migrations cannot run dynamically
during test under the `Ecto.Adapters.SQL.Sandbox`, as the sandbox has
to share a single connection across processes to guarantee the changes
can be reverted.
## Options
* `:all` - runs all available if `true`
* `:step` - runs the specific number of migrations
* `:to` - runs all until the supplied version is reached
(including the version given in `:to`)
* `:to_exclusive` - runs all until the supplied version is reached
(excluding the version given in `:to_exclusive`)
Plus all other options described in `up/4`.
"""
@spec run(Ecto.Repo.t(), String.t() | [String.t()] | [{integer, module}], atom, Keyword.t()) ::
[integer]
def run(repo, migration_source, direction, opts) do
migration_source = List.wrap(migration_source)
pending =
lock_for_migrations(true, repo, opts, fn _config, versions ->
cond do
opts[:all] ->
pending_all(versions, migration_source, direction)
to = opts[:to] ->
pending_to(versions, migration_source, direction, to)
to_exclusive = opts[:to_exclusive] ->
pending_to_exclusive(versions, migration_source, direction, to_exclusive)
step = opts[:step] ->
pending_step(versions, migration_source, direction, step)
true ->
{:error,
ArgumentError.exception(
"expected one of :all, :to, :to_exclusive, or :step strategies"
)}
end
end)
# The lock above already created the table, so we can now skip it.
opts = Keyword.put(opts, :skip_table_creation, true)
ensure_no_duplication!(pending)
migrate(Enum.map(pending, &load_migration!/1), direction, repo, opts)
end
@doc """
Returns an array of tuples as the migration status of the given repo,
without actually running any migrations.
Equivalent to:
Ecto.Migrator.migrations(repo, [Ecto.Migrator.migrations_path(repo)])
"""
@spec migrations(Ecto.Repo.t()) :: [{:up | :down, id :: integer(), name :: String.t()}]
def migrations(repo) do
migrations(repo, [migrations_path(repo)])
end
@doc """
Returns an array of tuples as the migration status of the given repo,
without actually running any migrations.
"""
@spec migrations(Ecto.Repo.t(), String.t() | [String.t()], Keyword.t()) ::
[{:up | :down, id :: integer(), name :: String.t()}]
def migrations(repo, directories, opts \\ []) do
directories = List.wrap(directories)
repo
|> migrated_versions(opts)
|> collect_migrations(directories)
|> Enum.sort_by(fn {_, version, _} -> version end)
end
use GenServer
@doc """
Runs migrations as part of your supervision tree.
## Options
* `:repos` - Required option to tell the migrator which Repo's to
migrate. Example: `repos: [MyApp.Repo]`
* `:skip` - Option to skip migrations. Defaults to `false`.
Plus all other options described in `up/4`.
See "Example: Running migrations on application startup" for more info.
"""
def start_link(opts) do
GenServer.start_link(__MODULE__, opts, name: __MODULE__)
end
@impl true
def init(opts) do
{repos, opts} = Keyword.pop!(opts, :repos)
{skip?, opts} = Keyword.pop(opts, :skip, false)
{migrator, opts} = Keyword.pop(opts, :migrator, &Ecto.Migrator.run/3)
opts = Keyword.put(opts, :all, true)
unless skip? do
for repo <- repos do
{:ok, _, _} = with_repo(repo, &migrator.(&1, :up, opts))
end
end
:ignore
end
defp collect_migrations(versions, migration_source) do
ups_with_file =
versions
|> pending_in_direction(migration_source, :down)
|> Enum.map(fn {version, name, _} -> {:up, version, name} end)
ups_without_file =
versions
|> versions_without_file(migration_source)
|> Enum.map(fn version -> {:up, version, "** FILE NOT FOUND **"} end)
downs =
versions
|> pending_in_direction(migration_source, :up)
|> Enum.map(fn {version, name, _} -> {:down, version, name} end)
ups_with_file ++ ups_without_file ++ downs
end
defp versions_without_file(versions, migration_source) do
versions_with_file =
migration_source
|> migrations_for()
|> Enum.map(fn {version, _, _} -> version end)
versions -- versions_with_file
end
defp lock_for_migrations(lock_or_migration_number, repo, opts, fun) do
dynamic_repo = Keyword.get(opts, :dynamic_repo, repo.get_dynamic_repo())
skip_table_creation = Keyword.get(opts, :skip_table_creation, false)
previous_dynamic_repo = repo.put_dynamic_repo(dynamic_repo)
try do
config = repo.config()
unless skip_table_creation do
verbose_schema_migration(repo, "create schema migrations table", fn ->
SchemaMigration.ensure_schema_migrations_table!(repo, config, opts)
end)
end
{migration_repo, query, all_opts} = SchemaMigration.versions(repo, config, opts[:prefix])
migration_lock? =
Keyword.get(opts, :migration_lock, Keyword.get(config, :migration_lock, true))
opts =
opts
|> Keyword.put(:migration_source, config[:migration_source] || "schema_migrations")
|> Keyword.put(:log, migrator_log(opts))
result =
if lock_or_migration_number && migration_lock? do
# If there is a migration_repo, it wins over dynamic_repo,
# otherwise the dynamic_repo is the one locked in migrations.
meta_repo = if migration_repo != repo, do: migration_repo, else: dynamic_repo
meta = Ecto.Adapter.lookup_meta(meta_repo)
migration_repo.__adapter__().lock_for_migrations(meta, opts, fn ->
fun.(config, migration_repo.all(query, all_opts))
end)
else
fun.(config, migration_repo.all(query, all_opts))
end
case result do
{kind, reason, stacktrace} ->
:erlang.raise(kind, reason, stacktrace)
{:error, error} ->
raise error
result ->
result
end
after
repo.put_dynamic_repo(previous_dynamic_repo)
end
end
defp conditional_lock_for_migrations(module, version, repo, opts, fun) do
lock = if module.__migration__()[:disable_migration_lock], do: false, else: version
lock_for_migrations(lock, repo, opts, fun)
end
defp pending_to(versions, migration_source, direction, target) when is_integer(target) do
within_target_version? = fn
{version, _, _}, target, :up ->
version <= target
{version, _, _}, target, :down ->
version >= target
end
pending_in_direction(versions, migration_source, direction)
|> Enum.take_while(&within_target_version?.(&1, target, direction))
end
defp pending_to_exclusive(versions, migration_source, direction, target)
when is_integer(target) do
within_target_version? = fn
{version, _, _}, target, :up ->
version < target
{version, _, _}, target, :down ->
version > target
end
pending_in_direction(versions, migration_source, direction)
|> Enum.take_while(&within_target_version?.(&1, target, direction))
end
defp pending_step(versions, migration_source, direction, count) do
pending_in_direction(versions, migration_source, direction)
|> Enum.take(count)
end
defp pending_all(versions, migration_source, direction) do
pending_in_direction(versions, migration_source, direction)
end
defp pending_in_direction(versions, migration_source, :up) do
migration_source
|> migrations_for()
|> Enum.filter(fn {version, _name, _file} -> version not in versions end)
end
defp pending_in_direction(versions, migration_source, :down) do
migration_source
|> migrations_for()
|> Enum.filter(fn {version, _name, _file} -> version in versions end)
|> Enum.reverse()
end
defp migrations_for(migration_source) when is_list(migration_source) do
migration_source
|> Enum.flat_map(fn
directory when is_binary(directory) ->
Path.join([directory, "**", "*.{ex,exs}"])
|> Path.wildcard()
|> Enum.map(&extract_migration_info/1)
|> Enum.filter(& &1)
{version, module} ->
[{version, module, module}]
end)
|> Enum.sort()
end
defp extract_migration_info(file) do
base = Path.basename(file)
case Integer.parse(Path.rootname(base)) do
{integer, "_" <> name} ->
if Path.extname(base) == ".ex" do
# See: https://github.com/elixir-ecto/ecto_sql/issues/599
IO.warn(
"""
file looks like a migration but ends in .ex. \
Migration files should end in .exs. Use "mix ecto.gen.migration" to generate \
migration files with the correct extension.\
""",
stacktrace_info(file: file)
)
nil
else
{integer, name, file}
end
_ ->
nil
end
end
# TODO: Remove when we require Elixir 1.14
if Version.match?(System.version(), ">= 1.14.0") do
defp stacktrace_info(info), do: info
else
defp stacktrace_info(_info), do: []
end
defp ensure_no_duplication!([{version, name, _} | t]) do
cond do
List.keyfind(t, version, 0) ->
raise Ecto.MigrationError,
"migrations can't be executed, migration version #{version} is duplicated"
List.keyfind(t, name, 1) ->
raise Ecto.MigrationError,
"migrations can't be executed, migration name #{name} is duplicated"
true ->
ensure_no_duplication!(t)
end
end
defp ensure_no_duplication!([]), do: :ok
defp load_migration!({version, _, mod}) when is_atom(mod) do
if migration?(mod) do
{version, mod}
else
raise Ecto.MigrationError, "module #{inspect(mod)} is not an Ecto.Migration"
end
end
defp load_migration!({version, _, file}) when is_binary(file) do
loaded_modules = file |> Code.compile_file() |> Enum.map(&elem(&1, 0))
if mod = Enum.find(loaded_modules, &migration?/1) do
{version, mod}
else
raise Ecto.MigrationError,
"file #{Path.relative_to_cwd(file)} does not define an Ecto.Migration"
end
end
defp migration?(mod) do
Code.ensure_loaded?(mod) and function_exported?(mod, :__migration__, 0)
end
defp migrate([], direction, _repo, opts) do
level = Keyword.get(opts, :log, :info)
log(level, "Migrations already #{direction}")
[]
end
defp migrate(migrations, direction, repo, opts) do
for {version, mod} <- migrations,
do_direction(direction, repo, version, mod, opts),
do: version
end
defp do_direction(:up, repo, version, mod, opts) do
conditional_lock_for_migrations(mod, version, repo, opts, fn config, versions ->
unless version in versions do
do_up(repo, config, version, mod, opts)
end
end)
end
defp do_direction(:down, repo, version, mod, opts) do
conditional_lock_for_migrations(mod, version, repo, opts, fn config, versions ->
if version in versions do
do_down(repo, config, version, mod, opts)
end
end)
end
defp verbose_schema_migration(repo, reason, fun) do
try do
fun.()
rescue
error ->
Logger.error("""
Could not #{reason}. This error usually happens due to the following:
* The database does not exist
* The "schema_migrations" table, which Ecto uses for managing
migrations, was defined by another library
* There is a deadlock while migrating (such as using concurrent
indexes with a migration_lock)
To fix the first issue, run "mix ecto.create" for the desired MIX_ENV.
To address the second, you can run "mix ecto.drop" followed by
"mix ecto.create", both for the desired MIX_ENV. Alternatively you may
configure Ecto to use another table and/or repository for managing
migrations:
config #{inspect(repo.config()[:otp_app])}, #{inspect(repo)},
migration_source: "some_other_table_for_schema_migrations",
migration_repo: AnotherRepoForSchemaMigrations
The full error report is shown below.
""")
reraise error, __STACKTRACE__
end
end
defp log(false, _msg), do: :ok
defp log(true, msg), do: Logger.info(msg)
defp log(level, msg), do: Logger.log(level, msg)
defp migrator_log(opts) do
Keyword.get(opts, :log_migrator_sql, false)
end
defp ensure_repo_started(repo, pool_size) do
case repo.start_link(pool_size: pool_size) do
{:ok, _} ->
{:ok, :stop}
{:error, {:already_started, _pid}} ->
{:ok, :restart}
{:error, _} = error ->
error
end
end
defp ensure_migration_repo_started(repo, repo) do
{:ok, :noop}
end
defp ensure_migration_repo_started(migration_repo, _repo) do
case migration_repo.start_link() do
{:ok, _} ->
{:ok, :stop}
{:error, {:already_started, _pid}} ->
{:ok, :noop}
{:error, _} = error ->
error
end
end
defp after_action(repo, :restart) do
if Process.whereis(repo) do
%{pid: pid} = Ecto.Adapter.lookup_meta(repo)
Supervisor.restart_child(repo, pid)
end
end
defp after_action(repo, :stop) do
repo.stop()
end
defp after_action(_repo, :noop) do
:noop
end
end
