Skip to main content

7. Light Clients

7.1. Requirements for Light Clients

We list the requirements of a Light Client categorized along the three dimensions of Functionality, Efficiency, and Security.

  • Functional Requirements:

    1. Synchronize with full nodes to obtain the latest finalized Block Header Definition 10, and in turn, the state trie root.

    2. (Optional) Verify validity of runtime transitions (Section 2.6.).

    3. Make queries for data at the latest block height or across a range of blocks.

    4. Append extrinsics (Section 2.3.) to the blockchain via full nodes.

  • Efficiency Requirements:

    1. Efficient bootstrapping and syncing: initializations and update functions of the state have tractable computation and communication complexity and grows at most linearly with the chain size. Generally, the complexity is proportional to the GRANDPA validator set change.

    2. Querying operations happen by requesting the key-value pair from a full node.

    3. Further, verifying the validity of responses by the full node is logarithmic in the size of the state.

  • Security Requirements:

    1. Secure bootstrapping and Synchronizing: The probability that an adversarial full node convinces a light client of a forged blockchain state is negligible.

    2. Secure querying: The probability that an adversary convinces a light client to accept a forged account state is negligible.

    3. Assure that the submitted extrinsics are appended in a successor block or inform the user in case of failure.

  • Polkadot Specific Requirements:

    1. The client MUST be able to connect to a relay chain using chain state.

    2. The client MUST be able to retrieve the checkpoint state from a trusted source to speed up initialization.

    3. The client MUST be able to subscribe/unsubscribe to/from any polkadot-spec-conformant relay chain (Polkadot, Westend, Kusama)

    4. The client MUST be able to subscribe/unsubscribe to/from parachains that do not use custom protocols or cryptography methods other than those that Polkadot, Westend and Kusama use.

    5. The client MUST support the following RPC methods: rpc_methods, chainHead_unstable_follow, chainHead_unstable_unfollow, chainHead_unstable_unpin, chainHead_unstable_storage, chainHead_unstable_call chainHead_unstable_stopCall. transaction_unstable_submitAndWatch, and transaction_unstable_unwatch

    6. The client MUST support the @substrate/connect connection extension protocol: ToApplicationError, ToApplicationChainReady, ToApplicationRpc, ToExtensionAddChain, ToExtensionAddWellKnownChain, ToExtensionRpc, ToExtensionRemoveChain.

7.2. Warp Sync for Light Clients

Warp sync (Section 4.8.5.) only downloads the block headers where authority set changes occurred, so-called fragments (Definition 46), and by verifying the GRANDPA justifications (Definition 83). This protocol allows nodes to arrive at the desired state much faster than fast sync. Warp sync is primarily designed for Light Clients. Although, warp sync could be used by full nodes, the sync process may lack information to cater to complete functionality set of full nodes.

For light clients, it is too expensive to download the state (approx. 550MB) to respond to queries. Rather, the queries are submitted to the Full node, and only the response of the full node is validated using the hash of the state root. Requests for warp sync are performed using the /dot/sync/warp Request-Response substream, the corresponding network messages are detailed in Section 4.7..

Light clients base their trust in provided snapshots and the ability to slash grandpa votes for equivocation for the period they are syncing via warp sync. Full nodes and above, in contrast, verify each block individually.

In theory, the warp sync process takes the Genesis Block as input and outputs the hash of the state trie root at the latest finalized block. This root hash acts as proof to further validate the responses to queries by the full node. The warp sync works by starting from a trusted specified block (e.g., from a snapshot) and verifying the block headers only at the authority set changes.

Eventually, the light client verifies the finality of the block returned by a full node to ensure that the block is indeed the latest finalized block. This entails two things:

  1. Check the authenticity of GRANDPA Justifications messages from Genesis to the last finalized block.

  2. Check the timestamp of the last finalized block to ensure that no other blocks might have been finalized at a later timestamp.


Long-Range Attack Vulnerabilities: Warp syncing is particularly vulnerable to what is called long-range attacks. The authorities allowed to finalize blocks can generate multiple proofs of finality for multiple different blocks of the same height. Hence, they can finalize more than one chain at a time. It is possible for two-thirds of the validators that were active at a certain past block N to collude and decide to finalize a different block N', even when N has been finalized for the first time several weeks or months in the past. When a client then warp syncs, it can be tricked to consider this alternative block N' as the finalized one. However, in practice, to mitigate Long-Range Attacks, the starting point of the warp syncing is not too far in the past. How far exactly depends on the logic of the runtime of the chain. For example, in Polkadot, the starting block for the sync should be at max 28 days old to be within the purview of the slashing period for misbehaving nodes. Hence, even though, in theory, warp sync can start from Genesis Block, it is not advised to implement the same in practice.

We outline the warp sync process, abstracting out details of verifying the finality and how the full node to sync with is selected.

Algorithm 23. Warp Sync Light Clients
    \INPUT BlockHeader startblock, the initial block to start the sync. May not be the Genesis Block. 
    \OUTPUT CommitmentRootHash $root$, State Tries Root hash of the latest finalized Block. 
    \STATE \textsc{fullnode} $\leftarrow$ SelectFullNode  
    \STATE \textsc{latestBlockHeader, grandpaJustifications} $\leftarrow$ SyncWithNode(\textsc{fullnode})
    \STATE \textsc{isVerified} $\leftarrow$ verifyAuthoritySetChange(\textsc{grandpaJustifications}) $\land$ verifyFinality(\textsc{latestBlockHeader})
        \STATE $return$ $SOME$ getCommitmentRootHash(\textsc{latestBlockHeader})
    \STATE {$throw$ $ERROR$}

Abstraction of Warp Sync and verification of the latest block’s finality.

SelectFullNode{SelectFullNode}: Determines the full node that the light client syncs with.

SyncSithNode{SyncSithNode}: Returns the header of the latest finalized block and a list of Grandpa Justifications by the full node.

verifyAuthoritySetChange{verifyAuthoritySetChange}: Verification algorithm which checks the authenticity of the header only at the end of an era where the authority set changes iteratively until reaching the latest era.

verifyFinalty{verifyFinalty}: Verifies the finality of the latest block using the Grandpa Justifications messages.

The warp syncing process is closely coupled with the state querying procedure used by the light client. We outline the process of querying the state by a light client and validating the response.

Algorithm 24. Querying State Light Clients
    \INPUT Query q, BlockHeight h, CommitmentRootHash $root$
    \OUTPUT Maybe Result $res$
    \STATE ($res$, $\pi$) $\leftarrow QueryFullNode (q, h)$
    \IF{$validityCheck_{root}(res, \pi)$}
        \STATE $return$ $SOME$ $res$
    \STATE {$throw$ $ERROR$}

Querying State Algorithm.

QueryFullNode{QueryFullNode}: Returns the response to the query requested from the Full Node for the query q{q} at block height h{h}.

validityCheckroot{validityCheck}_{root}: Predicate that checks the validity of response res{res} and associated merkle proof π\pi by matching it against the Commit Root Hash root{root} obtained as a result of warp sync.

7.3. Runtime Environment for Light Clients

Technically, though a runtime execution environment is not necessary to build a light client, most clients require interacting with the Runtime and the state of the blockchain for integrity checks at the minimum. One can imagine an application scenario like an on-chain light client which only listens to the latest state without ever adding extrinsics. Current implementations of Light Nodes (for e.g., Smoldot) use the wasmtime as its runtime environment to drastically simplify the code. The performance of wasmtime is satisfying enough not to require a native runtime. The details of the runtime API that the environment needs to support can be found in (Appendix C).

7.4. Light Client Messages

Light clients are applications that fetch the required data that they need from a Polkadot node with an associated proof to validate the data. This makes it possible to interact with the Polkadot network without requiring to run a full node or having to trust the remote peers. The light client messages make this functionality possible.

All light client messages are protobuf encoded and are sent over the /dot/light/2 substream.

7.4.1. Request

A message with all possible request messages. All messages are sent as part of this message.

oneof (request)The request type

Where the request can be one of the following fields:

RemoteCallRequest1A remote call request (Definition 106)
RemoteReadRequest2A remote read request (Definition 108)
RemoteReadChildRequest4A remote read child request (Definition 110)

7.4.2. Response

A message with all possible response messages. All messages are sent as part of this message.

oneof (response)The response type

Where the response can be one of the following fields:

RemoteCallResponse1A remote call response (Definition 107)
RemoteReadResponse2A remote read response (Definition 109)

7.4.3. Remote Call Messages

Execute a call to a contract at the given block.

Definition 106. Remote Call Request

Remote call request.

bytes2Block at which to perform call
string3Method name
bytes4Call data
Definition 107. Remote Call Response

Remote call response.

bytes2An Option type (Definition 200) containing the call proof or None if proof generation failed.

7.4.4. Remote Read Messages

Read a storage value at the given block.

Definition 108. Remote Read Request

Remote read request.

bytes2Block at which to perform call
repeated bytes3Storage keys
Definition 109. Remote Read Response

Remote read response.

bytes2An Option type (Definition 200) containing the read proof or None if proof generation failed.

7.4.5. Remote Read Child Messages

Read a child storage value at the given block.

Definition 110. Remote Read Child Request

Remote read child request.

bytes2Block at which to perform call
bytes3Child storage key, this is relative to the child type storage location
bytes6Storage keys

The response is the same as for the Remote Read Request message, respectively Definition 109.

7.5. Storage for Light Clients

The light client requires a persistent storage for saving the state of the blockchain. In addition, it requires efficient Serialization/De-serialization methods to transform SCALE (Section A.2.2.) encoded network traffic for storing and reading from the persistent storage.