Skip to content

Latest commit

 

History

History

ledger

Flow Ledger Package

Ledger is a stateful fork-aware key/value storage. Any update (value change for a key) to the ledger generates a new ledger state. Updates can be applied to any recent state. These changes don't have to be sequential and ledger supports a tree of states. Ledger provides value lookup by key at a particular state (historic lookups) and can prove the existence/non-existence of a key-value pair at the given state. Ledger assumes the initial state includes all keys with an empty bytes slice as value.

This package provides two ledger implementations:

  • Complete Ledger implements a fast, memory-efficient and reliable ledger. It holds a limited number of recently used states in memory (for speed) and uses write-ahead logs and checkpointing to provide reliability. Under the hood complete ledger uses a collection of MTries(forest). MTrie is a customized in-memory binary Patricia Merkle trie storing payloads at specific storage paths. The payload includes both key-value pair and storage paths are determined by the PathFinder. Forest utilizes unchanged sub-trie sharing between tries to save memory.

  • Partial Ledger implements the ledger functionality for a limited subset of keys. Partial ledgers are designed to be constructed and verified by a collection of proofs from a complete ledger. The partial ledger uses a partial binary Merkle trie which holds intermediate hash value for the pruned branched and prevents updates to keys that were not part of proofs.

Definitions

In this section we provide an overview of some of the concepts. Hence it is highly recommended to checkout this doc for the formal and technical definitions in more details.

binary Merkle tree

binary Merkle tree image

In this context a binary Merkle tree is defined as perfect binary tree with a specific height including three type of nodes:

  • leaf nodes: holds a payload (data), a path (where the node is located), and a hash value (hash of path and payload content)

  • interim nodes: holds a hash value which is defined as hash of hash value of left and right children.

node types image

A path is a unique address of a node storing a payload. Paths are derived from the content of payloads (see common/pathfinder). A path is explicitly a hash of 256 bits.

paths image

Operations

Get: Fetching a payload from the binary Merkle tree is by traversing the tree based on path bits. (0: left branch, 1: right branch)

Update: Updates to the tree starts with traversing the tree to the leaf node, updating payload, hash value of that node and hash value of all the ancestor nodes (nodes on higher level connected to this node).

update image

Prove: A binary Merkle tree can provide an inclusion proof for any given payload. A Merkle proof in this context includes all the information needed to walk through a tree branch from an specific leaf node (key) up to the root of the tree (yellow node hash values are needed for inclusion proof for the green node).

proof image

Memory-trie (Mtrie)

An Mtrie in this context is defined as a compact version of binary Merkle tree, providing the exact same functionality but doesn't explicitly store empty nodes. Formally, a node is empty:

  • the node is an empty leaf node: it doesn't hold any data and only stores a path. Its hash value is defined as a default hash based on the height of tree.
  • an interim node is defined to be empty, if its two children are empty.

binary partial trie image

forest

Formally, a forest is any acyclic graph. Any set of disjoint trees forms a forest. In the context of Flow, we take an existing state, represented by a Merkle tree. Updating the payload of some of the leafs creates a new Merkle tree, which we add to the forest. In other words, the forest holds a set of state snapshots

compact forest

A compact forest constructs a new trie after each update (copy on change) and reuses unchanged sub-tries from the parent.

compact forest image

path finder

Path finder deterministically computes a path for a given payload. Path finder is responsible for making sure the trie grows in balance.

partial binary Merkle trie

A partial Merkle trie is similar to a Merkle trie but only keeping a subset of nodes and having intermediate nodes without the full sub-trie. It can be constructed from batch of inclusion and non-inclusion proofs. It provides functionality to verify outcome of updates to a trie without the need to have the full trie.

partial trie image