Data.HashTable.Internal

You can find benchmarks and more information about the internals of this package here: https://lowerbound.io/blog/2019-10-24_concurrent_hash_table_performance.html

Documentation

data MigrationStatus

Constructors

NotStarted

Ongoing

Finished

Instances

Eq MigrationStatus

Defined in Data.HashTable.Internal

Methods

(==) :: MigrationStatus -> MigrationStatus -> Bool

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

Show MigrationStatus

Defined in Data.HashTable.Internal

Methods

showsPrec :: Int -> MigrationStatus -> ShowS

show :: MigrationStatus -> String

showList :: [MigrationStatus] -> ShowS

data Chain k v

Used for chain-hashing.

Constructors

Chain

Fields

• _itemsTV :: TVar [(k, v)]

  stores the items for this index

• _migrationStatusTV :: TVar MigrationStatus

  used internally for resizing

Instances

Eq (Chain k v)

Defined in Data.HashTable.Internal

Methods

(==) :: Chain k v -> Chain k v -> Bool

(/=) :: Chain k v -> Chain k v -> Bool

newChainIO :: IO (Chain k v)

data HashTable k v

A thread-safe hash table that supports dynamic resizing.

Constructors

HashTable

Fields

• _chainsVecTV :: TVar (Vector (Chain k v))

  vector of linked lists

• _totalLoad :: IORef Int
• _config :: Config k

data Config k

Configuration options that may affect the performance of the hash table

Constructors

Config

Fields

• _scaleFactor :: Float

  scale factor for resizing

• _threshold :: Float

  load threshold for initiating resizing.

• _numResizeWorkers :: Int

  maximum number of worker threads used during resizing

• _hashFunc :: k -> Int

Instances

Show (Config k)

Defined in Data.HashTable.Internal

Methods

showsPrec :: Int -> Config k -> ShowS

show :: Config k -> String

showList :: [Config k] -> ShowS

mkDefaultConfig :: Hashable k => IO (Config k)

Default configuration: scale factor = 2.0; resizing threshold = 0.75; number of worker threads for resizing = getNumCapabilities; hash function = use hashWithSalt with a random salt.

new

Arguments

:: Eq k

=> Int

Initial size of the hash table

-> Config k

-> IO (HashTable k v)

Creates a new hash table with an initial size. See newWithDefaults for more details. You probably either want to use newWithDefaults instead or something like this: > mkDefaultConfig { _field = myValue } >>= new 10

newWithDefaults

Arguments

:: (Eq k, Hashable k)

=> Int

Initial size of the hash table

-> IO (HashTable k v)

Creates a new hash table with the given initial vector size, scale factor 2.0, a resizing load threshold of 0.75, and we use as many threads for resizing as we have cores available. This will use a hash function with a (single) random salt, so if you need security, you MUST supply your own hash function. To be replaced by universal hashing in future versions.

readSizeIO :: HashTable k v -> IO Int

Returns the size of the vector representing the hash table. (Atomic)

readSize :: HashTable k v -> STM Int

Returns the size of the vector representing the hash table.

resize :: Eq k => HashTable k v -> IO ()

Resizes the hash table by scaling it according to the _scaleFactor in the configuration.

lookup :: Eq k => HashTable k v -> k -> IO (Maybe v)

Lookup the value for the key in the hash table if it exists.

type STMAction k v a = TVar [(k, v)] -> STM (Maybe a)

An action to be executed atomically for the chain (list) stored at a specific table idnex. Used by genericModify.

genericModify

Arguments

:: Eq k

=> HashTable k v

-> k

key

-> STMAction k v a

-> IO a

Used by insert, insertIfNotExists, delete, and update.

insert

Arguments

:: Eq k

=> HashTable k v

-> k

key

-> v

value

-> IO Bool

Inserts the key-value pair k v into the hash table. Uses chain hashing to resolve collisions. If you want to update the entry only if it already exists, use update. If you want to update the entry only if it does *not* exist, use insertIfNotExists.

insertIfNotExists

Arguments

:: Eq k

=> HashTable k v

-> k

key

-> v

value

-> IO Bool

Inserts a key and value pair into the hash table only if the key does not exist yet. Returns True if the insertion was successful, i.e., the hash table did not contain this key before. To get the same behaviour as insert, use insert. If you want to update an entry only if it exists, use update.

update

Arguments

:: Eq k

=> HashTable k v

-> k

key

-> v

value

-> IO Bool

Updates the value for key k. If k is not in the hash table, it skips the update and returns False.

modify

Arguments

:: Eq k

=> HashTable k v

-> k

key

-> (v -> v)

update-function

-> IO Bool

Applies an update-function to the value for key k. If k is not in the hash table, it just returns False.

delete

Arguments

:: Eq k

=> HashTable k v

-> k

key of entry that will be deleted

-> IO Bool

Deletes the entry for the given key from the hash table. Returns True if and only if an entry was deleted from the table.

atomicallyChangeLoad

Arguments

:: Eq k

=> HashTable k v

-> Int

increment/decrement value

-> IO ()

Atomically increment/decrement the table load value by adding the provided integer value to the current value.

readLoad :: HashTable k v -> IO Int

The load (i.e. number of stored items) in the table. Note that this is not synchronized for performance reasons and hence might be somewhat out of date if a lot of contention is happening.

readAssocs :: Eq k => HashTable k v -> STM [(k, v)]

Atomically retrieves list of key-value pairs. If there is a lot of contention going on, this may be very inefficient.

deleteFirstKey :: Eq a => a -> [(a, b)] -> [(a, b)]

Takes a key k and an assocation list ys, and deletes the first entry with key k in ys. Used internally.

readChainForKeyIO :: HashTable k v -> k -> IO (Chain k v)

Atomically read the chain for the given key.

readChainForIndexIO :: HashTable k v -> Int -> IO (Chain k v)

Atomically read the chain for the given index. (Warning: bounds are not checked.)

readChainForIndex :: HashTable k v -> Int -> STM (Chain k v)

Atomically read the chain for the given index. (Warning: bounds are not checked.)

debug :: Show a => a -> IO ()