Database.Beam.Migrate.Actions

Data types and functions to discover sequences of DDL commands to go from one database state to another. Used for migration generation.

For our purposes, a database state is fully specified by the set of predicates that apply to that database.

Migration generation is approached as a graph search problem over the infinite graph of databases G. The nodes of G are database states, which (as said above) are simply sets of predicates (see DatabaseState for the realization of this concept in code). For two vertices S1 and S2 in G, there is an edge between the two if and only if there is a DDL command that can take a database at S1 to S2.

We generate migrations by exploring this graph, starting at the source state and ending at the destination state. By default we use an optimizing solver that weights each edge by the complexity of the particular command, and we attempt to find the shortest path using Dijkstra's algorithm, although a user may override this behavior and provide a custom edge selection mechanism (or even defer this choice to the user).

In order to conduct the breadth-first search, we must know which edges lead out of whichever vertex we're currently visiting. The solving algorithm thus takes a set of ActionProviders, which are means of discovering edges that are incident to the current database state.

Conceptually, an ActionProvider is a function of type ActionProviderFn, which takes the current database state and produces a list of edges in the form of PotentialAction objects. For optimization purposes, ActionProviders also take in the desired destination state, which it can use to select only edges that make sense. This does not affect the result, just the amount of time it may take to get there.

Note that because the graph of database states is infinite, a breadth-first search may easily end up continuing to explore when there is no chance of reaching our goal. This would result in non-termination and is highly undesirable. In order to prevent this, we limit ourselves to only exploring edges that take us closer to the destination state. Here, we measure distance between two states as the number of elements in the symmetric difference of two database states. Thus, every action we take must either remove a predicate that doesn't exist in the destination state, or add a predicate that does. If a potential action only adds predicates that do not exist in the final state or removes predicates that do not exist in the first, then we never explore that edge.

A note on speed

There are some issues with this approach. Namely, if there is no solution, we can end up exploring the entire action space, which may be quite a lot. While beam-migrate can solve all databases that can be made up of predicates in this module, other beam backends may not make such strict guarantees (although in practice, all do). Nevertheless, if you're hacking on this module and notice what seems like an infinite loop, you may have accidentally removed code that exposed the edge that leads to a solution to the migration.

Database state

data DatabaseStateSource

Used to indicate whether a particular predicate is from the initial database state, or due to a sequence of actions we've committed too. Used to prevent runaway action generation based off of derived states.

Constructors

DatabaseStateSourceOriginal

Predicate is from the original set given by the user

DatabaseStateSourceDerived

Predicate is from an action we've committed to in this action chain

Instances

Bounded DatabaseStateSource

Methods

minBound :: DatabaseStateSource

maxBound :: DatabaseStateSource

Enum DatabaseStateSource

Methods

succ :: DatabaseStateSource -> DatabaseStateSource

pred :: DatabaseStateSource -> DatabaseStateSource

toEnum :: Int -> DatabaseStateSource

fromEnum :: DatabaseStateSource -> Int

enumFrom :: DatabaseStateSource -> [DatabaseStateSource]

enumFromThen :: DatabaseStateSource -> DatabaseStateSource -> [DatabaseStateSource]

enumFromTo :: DatabaseStateSource -> DatabaseStateSource -> [DatabaseStateSource]

enumFromThenTo :: DatabaseStateSource -> DatabaseStateSource -> DatabaseStateSource -> [DatabaseStateSource]

Eq DatabaseStateSource

Methods

(==) :: DatabaseStateSource -> DatabaseStateSource -> Bool

(/=) :: DatabaseStateSource -> DatabaseStateSource -> Bool

Ord DatabaseStateSource

Methods

compare :: DatabaseStateSource -> DatabaseStateSource -> Ordering

(<) :: DatabaseStateSource -> DatabaseStateSource -> Bool

(<=) :: DatabaseStateSource -> DatabaseStateSource -> Bool

(>) :: DatabaseStateSource -> DatabaseStateSource -> Bool

(>=) :: DatabaseStateSource -> DatabaseStateSource -> Bool

max :: DatabaseStateSource -> DatabaseStateSource -> DatabaseStateSource

min :: DatabaseStateSource -> DatabaseStateSource -> DatabaseStateSource

Show DatabaseStateSource

Methods

showsPrec :: Int -> DatabaseStateSource -> ShowS

show :: DatabaseStateSource -> String

showList :: [DatabaseStateSource] -> ShowS

Generic DatabaseStateSource

Associated Types

type Rep DatabaseStateSource :: * -> *

Methods

from :: DatabaseStateSource -> Rep DatabaseStateSource x

to :: Rep DatabaseStateSource x -> DatabaseStateSource

NFData DatabaseStateSource

Methods

rnf :: DatabaseStateSource -> ()

type Rep DatabaseStateSource

type Rep DatabaseStateSource = D1 * (MetaData "DatabaseStateSource" "Database.Beam.Migrate.Actions" "beam-migrate-0.2.0.0-luEQn16wb5J06t50caeQc" False) ((:+:) * (C1 * (MetaCons "DatabaseStateSourceOriginal" PrefixI False) (U1 *)) (C1 * (MetaCons "DatabaseStateSourceDerived" PrefixI False) (U1 *)))

data DatabaseState cmd

Represents the state of a database as a migration is being generated

Constructors

DatabaseState

Fields

• dbStateCurrentState :: !(HashMap SomeDatabasePredicate DatabaseStateSource)

  The current set of predicates that apply to this database as well as their source (user or from previous actions)

• dbStateKey :: !(HashSet SomeDatabasePredicate)

  HS.fromMap of dbStateCurrentState, for maximal sharing

• dbStateCmdSequence :: !(Seq cmd)

  The current sequence of commands we've committed to in this state

Instances

Show cmd => Show (DatabaseState cmd)

Methods

showsPrec :: Int -> DatabaseState cmd -> ShowS

show :: DatabaseState cmd -> String

showList :: [DatabaseState cmd] -> ShowS

NFData (DatabaseState cmd)

Methods

rnf :: DatabaseState cmd -> ()

Action generation

data PotentialAction cmd

Represents an edge (or a path) in the database graph.

Given a particular starting point, the destination database is the database where each predicate in actionPreConditions has been removed and each predicate in actionPostConditions has been added.

Constructors

PotentialAction

Fields

• actionPreConditions :: !(HashSet SomeDatabasePredicate)

  Preconditions that will no longer apply

• actionPostConditions :: !(HashSet SomeDatabasePredicate)

  Conditions that will apply after we're done

• actionCommands :: !(Seq cmd)

  The sequence of commands that accomplish this movement in the database graph. For an edge, actionCommands contains one command; for a path, it will contain more.

• actionEnglish :: !Text

  An english description of the movement

• actionScore :: !Int

  A heuristic notion of complexity or weight; used to find the "easiest" path through the graph.

Instances

Monoid (PotentialAction cmd)

PotentialActions can represent edges or paths. Monadically combining two PotentialActions results in the path between the source of the first and the destination of the second. mempty here returns the action that does nothing (i.e., the edge going back to the same database state)

Methods

mempty :: PotentialAction cmd

mappend :: PotentialAction cmd -> PotentialAction cmd -> PotentialAction cmd

mconcat :: [PotentialAction cmd] -> PotentialAction cmd

newtype ActionProvider cmd

Edge discovery mechanism. A newtype wrapper over ActionProviderFn.

An ActionProviderFn takes two arguments. The first is the set of predicates that exist in the current database.

The function should a set of edges from the database specified in the first argument to possible destination databases. For optimization purposes, the second argument is the set of predicates that ought to exist in the destination database. This can be used to eliminate edges that will not lead to a solution.

This second argument is just an optimization and doesn't change the final result, although it can significantly impact the time it takes to get there.

Both the current database set and the destination database set are given as polymorphic lists of predicates. When you instantiate the type, the current database predicate set is queried for predicates of that type.

For example, dropTableActionProvider provides a DROP TABLE action edge whenever it encounters a table that exists. In order to do this, it attempts to find all TableExistsPredicate that do not exist in the destination database. Its ActionProviderFn may be implemented like such:

dropTableActionProvider preConditions postConditions = do
    TableExistsPredicate srcTblNm <- preConditions
    ensuringNot_ $ $
      do TableExistsPredicate destTblNm <- postConditions
         guard (srcTblNm == destTblNm)

ensuringNot_ is a function that causes the action provider to return no results if there are any elements in the provided list. In this case, it's used to stop DROP TABLE action generation for tables which must be present in the final database.

Constructors

ActionProvider

Fields

• getPotentialActions :: ActionProviderFn cmd

Instances

Monoid (ActionProvider cmd)

Methods

mempty :: ActionProvider cmd

mappend :: ActionProvider cmd -> ActionProvider cmd -> ActionProvider cmd

mconcat :: [ActionProvider cmd] -> ActionProvider cmd

type ActionProviderFn cmd = (forall preCondition. Typeable preCondition => [preCondition]) -> (forall postCondition. Typeable postCondition => [postCondition]) -> [PotentialAction cmd]

See ActionProvider

ensuringNot_ :: Alternative m => [a] -> m ()

Proceeds only if no predicate matches the given pattern. See the implementation of dropTableActionProvider for an example of usage.

justOne_ :: [a] -> [a]

Used to ensure that only one predicate matches the given pattern. See the implementation of createTableActionProvider for an example of usage.

createTableActionProvider :: forall cmd. (Sql92SaneDdlCommandSyntax cmd, Sql92SerializableDataTypeSyntax (Sql92DdlCommandDataTypeSyntax cmd)) => ActionProvider cmd

Action provider for SQL92 CREATE TABLE actions.

dropTableActionProvider :: forall cmd. (Sql92SaneDdlCommandSyntax cmd, Sql92SerializableDataTypeSyntax (Sql92DdlCommandDataTypeSyntax cmd)) => ActionProvider cmd

Action provider for SQL92 DROP TABLE actions

addColumnProvider :: forall cmd. (Sql92SaneDdlCommandSyntax cmd, Sql92SerializableDataTypeSyntax (Sql92DdlCommandDataTypeSyntax cmd)) => ActionProvider cmd

Action provider for SQL92 ALTER TABLE ... ADD COLUMN ... actions

addColumnNullProvider :: forall cmd. Sql92SaneDdlCommandSyntax cmd => ActionProvider cmd

Action provider for SQL92 ALTER TABLE ... ALTER COLUMN ... SET NULL

dropColumnNullProvider :: forall cmd. Sql92SaneDdlCommandSyntax cmd => ActionProvider cmd

Action provider for SQL92 ALTER TABLE ... ALTER COLUMN ... SET NOT NULL

defaultActionProvider :: (Sql92SaneDdlCommandSyntax cmd, Sql92SerializableDataTypeSyntax (Sql92DdlCommandDataTypeSyntax cmd)) => ActionProvider cmd

Default action providers for any SQL92 compliant syntax.

In particular, this provides edges consisting of the following statements:

• CREATE TABLE
• DROP TABLE
• ALTER TABLE ... ADD COLUMN ...
• ALTER TABLE ... DROP COLUMN ...
• ALTER TABLE ... ALTER COLUMN ... SET [NOT] NULL

Solver

data Solver cmd where

Represents current state of a database graph search.

If ProvideSolution, the destination database has been reached, and the given list of commands provides the path from the source database to the destination.

If SearchFailed, the search has failed. The provided DatabaseStates represent the closest we could make it to the destination database. By default, only the best 10 are kept around (to avoid unbounded memory growth).

If ChooseActions, we are still searching. The caller is provided with the current state as well as a list of actions, provided as an opaque type f. The getPotentialActionChoice function can be used to get the PotentialAction corresponding to any given f. The caller is free to cull the set of potential actions according however they'd like (for example, by prompting the user). The selected actions to explore should be passed to the continueSearch function.

Use of the f existential type may seem obtuse, but it prevents the caller from injecting arbitrary actions. Instead the caller is limited to choosing only valid actions as provided by the suppled ActionProvider.

Constructors

ProvideSolution :: [cmd] -> Solver cmd

SearchFailed :: [DatabaseState cmd] -> Solver cmd

ChooseActions :: {..} -> Solver cmd

Fields

• choosingActionsAtState :: !(DatabaseState cmd)

  The current node we're searching at

• getPotentialActionChoice :: f -> PotentialAction cmd

  Convert the opaque f type to a PotentialAction. This can be used to present the actions to the user or to inspect the action to make a more informed choice on exploration strategies.

• potentialActionChoices :: [f]

  The possible actions that we can take, presented as an opaque list. Use the getPotentialActionChoice function to get the corresponding PotentialAction.

• continueSearch :: [f] -> Solver cmd

  Continue the search and get the next Solver

data FinalSolution cmd

Represents the final results of a search

Constructors

Solved [cmd]

The search found a path from the source to the destination database, and has provided a set of commands that would work

Candidates [DatabaseState cmd]

The search failed, but provided a set of DatbaseStates it encountered that were the closest to the destination database. By default, only 10 candidates are provided.

Instances

Show cmd => Show (FinalSolution cmd)

Methods

showsPrec :: Int -> FinalSolution cmd -> ShowS

show :: FinalSolution cmd -> String

showList :: [FinalSolution cmd] -> ShowS

finalSolution :: Solver cmd -> FinalSolution cmd

An exhaustive solving strategy that simply continues the search, while exploring every possible action. If there is a solution, this will find it.

heuristicSolver

Arguments

:: ActionProvider cmd

Edge discovery function

-> [SomeDatabasePredicate]

Source database state

-> [SomeDatabasePredicate]

Destination database state

-> Solver cmd

Conduct a breadth-first search of the database graph to find a path from the source database to the destination database, using the given ActionProvider to discovere "edges" (i.e., DDL commands) between the databases.

See the documentation on Solver for more information on how to consume the result.