Crate safe_vault [] [src]

Implementation of the "Vault" node for the SAFE Network.

General

Vaults form an overlay network on top of a Kademlia-like routing network to provide decentralised, autonomous, cloud storage.

Clients can connect to this network and make requests to store, retrieve, mutate or delete data. Currently the network handles two types of data chunks: ImmutableData and StructuredData.

The underlying routing network provides the ability to send and receive messages from peers or groups of "close" peers. These singletons or groups are referred to as authorities. See the general routing docs and the Authority docs for further details.

Vaults have two main sets of responsibilities, each set being associated with a "persona" of the Vault; managing Client accounts and managing data stored on the network. Among other things, it involves passing responsibility to new Vaults when there is local churn (i.e. a Vault joins or leaves nearby).

Client Manager Persona

The Client Manager persona (a.k.a. MaidManager) holds an account for each Client which is close to it in the network address space. Each such account is managed by a group of GROUP_SIZE Vaults.

A Client's account contains details of how many chunks of data have been put to the network by that Client and how many new chunks can still be put. If a Client's account indicates that no more chunks can be put to the network, the Client Managers for that Client disallow any further Put requests, responding with a LowBalance error.

Clients can retrieve their account balances by sending a specific request to their Managers, namely a GetAccountInfo request.

Churn

When a Vault joins the network nearby, the Client Manager will remove all accounts for which it is no longer responsible and will send the remainder to the close group for each account. These are termed "refresh" messages; they allow all members of each group to synchronise their account information.

When a close Vault leaves the network, the Client Manager simply sends a set of refreshes. It will thus be given any accounts which it was not previously managing, but for which it is now responsible.

Data Manager Persona

The Data Manager persona holds a chunk store where data chunks are held and is responsible for chunks whose names are close to it in the network address space. Not every Data Manager in a given group close to a chunk will necessarily hold that chunk, but every one should be aware of which peers do hold it.

Chunk Cache

The Data Manager also holds a cache of chunks for which it is not directly responsible. These are chunks for which the Vault was recently responsible but has churned out of the close group, or chunks which have recently been requested by a Client.

The purpose for caching the former is that there is a reasonable chance that the Vault will soon become responsible for that chunk again due to churn. This saves the chunk from having to be retrieved via new Get requests.

The rationale for the latter is that popular chunks will receive many Get requests by many different Clients and the work of satisfying these requests needs to be spread across more Vaults than normal. By opportunistically caching chunks which have been retrieved by Clients, popular chunks will end up being cached by many more Vaults than the close group, hence the close group will be protected from excessive numbers of Get requests.

Churn

In a similar way to the Client Manager, the Data Manager will clear out records and send refresh messages. In addition, it will also try to retrieve any chunks for which it is responsible from any close peer which is not currently busy, i.e. a peer which is not currently being asked for a chunk by this Vault. Chunk replication to this Vault continues repeatedly until it holds all chunks for which it is responsible.

Message Flows

As mentioned above, two chunk types are handled by Vaults: ImmutableData and StructuredData. Both types can be stored to the network by using Put messages and retrieved from the network by using Get messages. StructuredData can also be mutated by using Post messages or removed by using Delete messages.

The following sections give an overview of the end-to-end process involved in each of these operations.

Put ImmutableData

  1. Client sends Put to its MaidManager group
  2. MaidManager responds with failure (ending the message flow) if:
    • Client doesn't have an account
    • Client's account has insufficient balance to be allowed to store a new chunk
  3. MaidManager sends Put to DataManager group for the chunk
  4. if DataManager already has a copy of the chunk, it responds with success to MaidManager group. Otherwise, it:
    • responds with failure (NetworkFull error) to MaidManager group if its chunkstore is full
    • tries to store the chunk and responds with appropriate success or failure to MaidManager group
    • if the store attempt was successful, sends Refresh to its fellow DataManagers
  5. MaidManager then:
    • refunds the Client's account if the DataManager group reports failure
    • sends Refresh to its fellow MaidManagers
    • responds with appropriate success or failure to Client

Put StructuredData

  1. Client sends Put to its MaidManager group
  2. if the type tag of the chunk is 0, (representing an account creation request) MaidManager responds with failure (ending the message flow) if this account already exists
  3. if the type tag of the chunk is not 0, MaidManager responds with failure (ending the message flow) if:
    • Client doesn't have an account
    • Client's account has insufficient balance to be allowed to store a new chunk
  4. MaidManager sends Put to DataManager group for the chunk
  5. if DataManager already has a copy of the chunk, it responds with failure to MaidManager group. Otherwise, it:
    • responds with failure ("network full" error) to MaidManager group if its chunkstore is full
    • tries to store the chunk and responds with appropriate success or failure to MaidManager group
    • if the store attempt was successful, sends Refresh to its fellow DataManagers
  6. MaidManager then:
    • refunds the Client's account if the DataManager group reports failure
    • sends Refresh to its fellow MaidManagers
    • responds with appropriate success or failure to Client

Get ImmutableData or Get StructuredData

  1. Client sends Get to the DataManager group for the chunk
  2. DataManager responds with appropriate success or failure to Client depending on whether or not it can retrieve the requested chunk from its chunkstore

Post StructuredData

  1. Client sends Post to the DataManager group for the chunk
  2. DataManager responds with failure (ending the message flow) to the Client if:
    • there isn't an existing StructuredData chunk with the same name
    • the new chunk isn't a valid successor of the existing one
    • the chunkstore can't store the new chunk
  3. if storing the chunk succeeds, DataManager:
    • responds with success to the Client
    • sends Refresh to its fellow DataManagers

Delete StructuredData

Note that deleting involves sending a new version of the chunk to be deleted in order to allow validation of the request.

  1. Client sends Delete to the DataManager group for the chunk
  2. DataManager responds with failure (ending the message flow) to the Client if:
    • there isn't an existing StructuredData chunk with the same name
    • the new chunk isn't a valid successor of the existing one
    • the chunkstore can't delete the existing chunk
  3. if deleting the chunk succeeds, DataManager responds with success to the Client

Structs

Config

Lets a vault configure a wallet address and storage limit.

Vault

Main struct to hold all personas and Routing instance

Enums

DataId

Identifier for a data (immutable or mutable)

Constants

DEFAULT_MAX_OPS_COUNT

Default available number of operations per account.

TYPE_TAG_INVITE

The type tag of invitations to create an account.