liquid_ml/kv/
mod.rs

1//! A module for distributed [`Key`], [`Value`] stores that hold their data in
2//! memory and are generic for (deserialized) [`Value`]s of type `T`.
3//!
4//! A [`KVStore`] is essentially a very simple in-memory distributed database
5//! (with no persistence) that stores data as a collection of key-value pairs
6//! where a [`Key`] is a unique identifier to a [`Value`]. The [`KVStore`]
7//! utilizes the [`network`] module to communicate between nodes using
8//! [`KVMessage`]s.
9//!
10//! Internally [`KVStore`]s store their data in memory as serialized blobs
11//! (a [`Value`] aka a `Vec<u8>`). The [`KVStore`] caches deserialized
12//! [`Value`]s into their type `T` on a least-recently used basis. A hard limit
13//! for the cache size is set to be `1/3` the amount of total memory on the
14//! machine, though this will be changed to be configurable.
15//!
16//! # Provided [`KVStore`] Functionality
17//! - [`get`]: Retrieve data stored locally on a [`KVStore`] or in its cache
18//! - [`wait_and_get`]: Retrieve data either locally or over the network
19//! - [`put`]: Store a [`Key`], [`Value`] pair either locally on a [`KVStore`]
20//!    or send it over the network to store it on another [`KVStore`]
21//! - [`send_blob`]: a lower level interface to facilitate sending any
22//!    serialized data. In `liquid_ml`, this is used for sending
23//!    [`Rower`](../dataframe/trait.Rower.html)s
24//!    (and other use cases) in a
25//!    [`DistributedDataFrame`](../dataframe/struct.DistributedDataFrame.html)
26//!
27//!
28//!
29//! [`Key`]: struct.Key.html
30//! [`Value`]: type.Key.html
31//! [`KVStore`]: struct.KVStore.html
32//! [`get`]: struct.KVStore.html#method.get
33//! [`wait_and_get`]: struct.KVStore.html#method.wait_and_get
34//! [`put`]: struct.KVStore.html#method.put
35//! [`send_blob`]: struct.KVStore.html#method.send_blob
36//! [`KVMessage`]: enum.KVMessage.html
37//! [`Data`]: enum.KVMessage.html#variant.Data
38//! [`Put`]: enum.KVMessage.html#variant.Put
39//! [`Blob`]: enum.KVMessage.html#variant.Blob
40use rand::{self, Rng};
41use serde::{Deserialize, Serialize};
42
43mod kv_store;
44pub use crate::kv::kv_store::{KVMessage, KVStore};
45
46/// A `Key` defines where in a [`KVStore`] a [`Value`] is stored, as well as
47/// which node (and thus which [`KVStore`]) 'owns' the [`Value`]
48///
49/// [`KVStore`]: struct.KVStore.html
50/// [`Value`]: type.Value.html
51#[derive(PartialEq, Eq, Hash, Serialize, Deserialize, Debug, Clone)]
52pub struct Key {
53    /// Defines where in a [`KVStore`] a value is stored
54    ///
55    /// [`KVStore`]: struct.KVStore.html
56    pub name: String,
57    /// Defines which node 'owns' the associated [`Value`]
58    ///
59    /// [`Value`]: type.Key.html
60    pub home: usize,
61}
62
63/// A serialized blob of data. Is associated with a [`Key`] which defines where
64/// this `Value` is stored in a [`KVStore`], as well as its 'owner'
65///
66/// [`Key`]: struct.Key.html
67/// [`KVStore`]: struct.KVStore.html
68pub type Value = Vec<u8>;
69
70impl Key {
71    /// Creates a new [`Key`] that is owned by the [`KVStore`] running on the
72    /// node with id equal to `home`. The given `name` defines where in the
73    /// [`KVStore`] the value is stored.
74    ///
75    /// [`Key`]: struct.Key.html
76    /// [`KVStore`]: struct.KVStore.html
77    pub fn new(name: &str, home: usize) -> Self {
78        Key {
79            name: String::from(name),
80            home,
81        }
82    }
83
84    /// Make a key with an automatically generated name
85    pub(crate) fn generate(name: &str, home: usize) -> Self {
86        let mut rng = rand::thread_rng();
87        Key {
88            name: format!("{}-{}-{}", name, home, rng.gen::<i16>()),
89            home,
90        }
91    }
92}
liquid_ml/kv/mod.rs

liquid_ml/kv/
mod.rs
