Struct garage_table::crdt::LwwMap

pub struct LwwMap<K, V> { /* fields omitted */ }

Last Write Win Map

This types defines a CRDT for a map from keys to values. The values have an associated timestamp, such that the last written value takes precedence over previous ones. As for the simpler LWW type, the value type V is also required to implement the CRDT trait. We do not encourage mutating the values associated with a given key without updating the timestamp, in fact at the moment we do not provide a .get_mut() method that would allow that.

Internally, the map is stored as a vector of keys and values, sorted by ascending key order. This is why the key type K must implement Ord (and also to ensure a unique serialization, such that two values can be compared for equality based on their hashes). As a consequence, insertions take O(n) time. This means that LWWMap should be used for reasonably small maps. However, note that even if we were using a more efficient data structure such as a BTreeMap, the serialization cost O(n) would still have to be paid at each modification, so we are actually not losing anything here.

Implementations

impl<K, V> LwwMap<K, V> where
    K: Ord,
    V: Crdt, 

pub fn new() -> LwwMap<K, V>

Create a new empty map CRDT

pub fn migrate_from_raw_item(k: K, ts: u64, v: V) -> LwwMap<K, V>

Used to migrate from a map defined in an incompatible format. This produces a map that contains a single item with the specified timestamp (copied from the incompatible format). Do this as many times as you have items to migrate, and put them all together using the CRDT merge operator.

pub fn update_mutator(&self, k: K, new_v: V) -> LwwMap<K, V>

Returns a map that contains a single mapping from the specified key to the specified value. This map is a mutator, or a delta-CRDT, such that when it is merged with the original map, the previous value will be replaced with the one specified here. The timestamp in the provided mutator is set to the maximum of the current system’s clock and 1 + the previous value’s timestamp (if there is one), so that the new value will always take precedence (LWW rule).

Typically, to update the value associated to a key in the map, you would do the following:

let my_update = my_crdt.update_mutator(key_to_modify, new_value);
my_crdt.merge(&my_update);

However extracting the mutator on its own and only sending that on the network is very interesting as it is much smaller than the whole map.

pub fn take_and_clear(&mut self) -> LwwMap<K, V>

Takes all of the values of the map and returns them. The current map is reset to the empty map. This is very usefull to produce in-place a new map that contains only a delta that modifies a certain value:

let mut a = get_my_crdt_value();
let old_a = a.take_and_clear();
a.merge(&old_a.update_mutator(key_to_modify, new_value));
put_my_crdt_value(a);

Of course in this simple example we could have written simply pyt_my_crdt_value(a.update_mutator(key_to_modify, new_value)), but in the case where the map is a field in a struct for instance (as is always the case), this becomes very handy:

let mut a = get_my_crdt_value();
let old_a_map = a.map_field.take_and_clear();
a.map_field.merge(&old_a_map.update_mutator(key_to_modify, new_value));
put_my_crdt_value(a);

pub fn clear(&mut self)

Removes all values from the map

pub fn get(&self, k: &K) -> Option<&V>

Get a reference to the value assigned to a key

pub fn items(&self) -> &[(K, u64, V)]

Gets a reference to all of the items, as a slice. Usefull to iterate on all map values. In most case you will want to ignore the timestamp (second item of the tuple).

pub fn len(&self) -> usize

Returns the number of items in the map

pub fn is_empty(&self) -> bool

Returns true if the map is empty

Trait Implementations

impl<K, V> Clone for LwwMap<K, V> where
    K: Clone,
    V: Clone, 

pub fn clone(&self) -> LwwMap<K, V>

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<K, V> Crdt for LwwMap<K, V> where
    K: Clone + Ord,
    V: Clone + Crdt, 

pub fn merge(&mut self, other: &LwwMap<K, V>)

Merge the two datastructures according to the CRDT rules. self is modified to contain the merged CRDT value. other is not modified.

impl<K, V> Debug for LwwMap<K, V> where
    K: Debug,
    V: Debug, 

pub fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl<K, V> Default for LwwMap<K, V> where
    K: Ord,
    V: Crdt, 

pub fn default() -> LwwMap<K, V>

Returns the “default value” for a type.

impl<'de, K, V> Deserialize<'de> for LwwMap<K, V> where
    K: Deserialize<'de>,
    V: Deserialize<'de>, 

pub fn deserialize<__D>(
    __deserializer: __D
) -> Result<LwwMap<K, V>, <__D as Deserializer<'de>>::Error> where
    __D: Deserializer<'de>, 

Deserialize this value from the given Serde deserializer.

impl<K, V> PartialEq<LwwMap<K, V>> for LwwMap<K, V> where
    K: PartialEq<K>,
    V: PartialEq<V>, 

pub fn eq(&self, other: &LwwMap<K, V>) -> bool

This method tests for self and other values to be equal, and is used by ==.

pub fn ne(&self, other: &LwwMap<K, V>) -> bool

This method tests for !=.

impl<K, V> Serialize for LwwMap<K, V> where
    K: Serialize,
    V: Serialize, 

pub fn serialize<__S>(
    &self,
    __serializer: __S
) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error> where
    __S: Serializer, 

Serialize this value into the given Serde serializer.

impl<K, V> StructuralPartialEq for LwwMap<K, V>