specs/standards/core/store.md

22 KiB
Raw Blame History

title name editor contributors
WAKU2-STORE Waku Store Query Hanno Cornelius <hanno@status.im>
Dean Eigenmann <dean@status.im>
Oskar Thorén <oskarth@titanproxy.com>
Aaryamann Challani <aaryamann@status.im>
Sanaz Taheri <sanaz@status.im>

Note: This version of WAKU2-STORE is earmarked to replace RFC 13/WAKU2-STORE once it reaches draft status


Abstract

This specification explains the WAKU2-STORE protocol which enables querying of messages received through the relay protocol and stored by other nodes. It also supports pagination for more efficient querying of historical messages.

Protocol identifier*: /vac/waku/store-query/3.0.0

Terminology

The term PII, Personally Identifiable Information, refers to any piece of data that can be used to uniquely identify a user. For example, the signature verification key, and the hash of one's static IP address are unique for each user and hence count as PII.

Design Requirements

The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in RFC2119.

Nodes willing to provide the storage service using WAKU2-STORE protocol, SHOULD provide a complete and full view of message history. As such, they are required to be highly available and specifically have a high uptime to consistently receive and store network messages. The high uptime requirement makes sure that no message is missed out hence a complete and intact view of the message history is delivered to the querying nodes. Nevertheless, in case storage provider nodes cannot afford high availability, the querying nodes may retrieve the historical messages from multiple sources to achieve a full and intact view of the past.

The concept of ephemeral messages introduced in WAKU2-MESSAGE affects WAKU2-STORE as well. Nodes running WAKU2-STORE SHOULD support ephemeral messages as specified in 14/WAKU2-MESSAGE. Nodes running WAKU2-STORE SHOULD NOT store messages with the ephemeral flag set to true.

Adversarial Model

Any peer running the WAKU2-STORE protocol, i.e. both the querying node and the queried node, are considered as an adversary. Furthermore, we currently consider the adversary as a passive entity that attempts to collect information from other peers to conduct an attack but it does so without violating protocol definitions and instructions. As we evolve the protocol, further adversarial models will be considered. For example, under the passive adversarial model, no malicious node hides or lies about the history of messages as it is against the description of the WAKU2-STORE protocol.

The following are not considered as part of the adversarial model:

  • An adversary with a global view of all the peers and their connections.
  • An adversary that can eavesdrop on communication links between arbitrary pairs of peers (unless the adversary is one end of the communication). In specific, the communication channels are assumed to be secure.

Wire Specification

Payloads

syntax = "proto3";

// Protocol identifier: /vac/waku/store-query/3.0.0
package waku.store.v3;

import "waku/message/v1/message.proto";

message WakuMessageKeyValue {
  optional bytes message_hash = 1; // Globally unique key for a Waku Message

  // Full message content and associated pubsub_topic as value
  optional waku.message.v1.WakuMessage message = 2;
  optional string pubsub_topic = 3;
}

message StoreQueryRequest {
  string request_id = 1;
  bool include_data = 2; // Response should include full message content
  
  // Filter criteria for content-filtered queries
  optional string pubsub_topic = 10;
  repeated string content_topics = 11;
  optional sint64 time_start = 12;
  optional sint64 time_end = 13;

  // List of key criteria for lookup queries
  repeated bytes message_hashes = 20; // Message hashes (keys) to lookup
  
  // Pagination info. 50 Reserved
  optional bytes pagination_cursor = 51; // Message hash (key) from where to start query (exclusive)
  bool pagination_forward = 52;
  optional uint64 pagination_limit = 53;
}

message StoreQueryResponse {
  string request_id = 1;

  optional uint32 status_code = 10;
  optional string status_desc = 11;

  repeated WakuMessageKeyValue messages = 20;

  optional bytes pagination_cursor = 51;
}

General store query concepts

Waku message key-value pairs

The store query protocol operates as a query protocol for a key-value store of historical Waku messages, with each entry having a 14/WAKU2-MESSAGE and associated pubsub topic as value, and deterministic message hash as key. The store can be queried to return either a set of keys or a set of key-value pairs. Within the store query protocol, Waku message keys and values MUST be represented in a WakuMessageKeyValue message. This message MUST contain the deterministic message_hash as key. It MAY contain the full WakuMessage and associated pubsub topic as value in the message and pubsub_topic fields, depending on the use case as set out below. If the message contains a value entry in addition to the key, both the message and pubsub_topic fields MUST be populated. The message MUST NOT have either message or pubsub_topic populated with the other unset. Both fields MUST either be set or unset.

Waku message store eligibility

In order for a Waku message to be eligible for storage:

  • it MUST be a valid 14/WAKU2-MESSAGE.
  • the timestamp field MUST be populated with the Unix epoch time at which the message was generated in nanoseconds. If at the time of storage the timestamp deviates by more than 20 seconds either into the past or the future when compared to the store nodes internal clock, the store node MAY reject the message.
  • the ephemeral field MUST be set to false.

Waku message sorting

The key-value entries in the store MUST be time-sorted by the WakuMessage timestamp attribute. Where two or more key-value entries have identical timestamps, the entries MUST be further sorted by the natural order of their message hash keys. Within the context of traversing over key-value entries in the store, "forward" indicates traversing the entries in ascending order, whereas "backward" indicates traversing the entries in descending order.

Pagination

If a large number of entries in the store service node match the query criteria provided in a StoreQueryRequest, the client MAY make use of pagination in a chain of store query request and response transactions to retrieve the full response in smaller batches termed "pages". Pagination can be performed either in a forward or backward direction.

A store query client MAY indicate the maximum number of matching entries it wants in the StoreQueryResponse, by setting the page size limit in the pagination_limit field. Note that a store service node MAY enforce its own limit if the pagination_limit is unset or larger than the service node's internal page size limit.

A StoreQueryResponse with a populated pagination_cursor indicates that more stored entries match the query than included in the response.

A StoreQueryResponse without a populated pagination_cursor indicates that there are no more matching entries in the store.

The client MAY request the next page of entries from the store service node by populating a subsequent StoreQueryRequest with the pagination_cursor received in the StoreQueryResponse. All other fields and query criteria MUST be the same as in the preceding StoreQueryRequest.

A StoreQueryRequest without a populated pagination_cursor indicates that the client wants to retrieve the "first page" of the stored entries matching the query.

Store Query Request

A client node MUST send all historical message queries within a StoreQueryRequest message. This request MUST contain a request_id. The request_id MUST be a uniquely generated string.

If the store query client requires the store service node to include Waku message values in the query response, it MUST set include_data to true. If the store query client requires the store service node to return only message hash keys in the query response, it SHOULD set include_data to false. By default, therefore, the store service node assumes include_data to be false.

A store query client MAY include query filter criteria in the StoreQueryRequest. There are two types of filter use cases:

  1. Content filtered queries and
  2. Message hash lookup queries

Content filtered queries

A store query client MAY request the store service node to filter historical entries by a content filter. Such a client MAY create a filter on content topic, on time range or on both.

To filter on content topic, the client MUST populate both the pubsub_topic and content_topics field. The client MUST NOT populate either pubsub_topic or content_topics and leave the other unset. Both fields MUST either be set or unset. A mixed content topic filter with just one of either pubsub_topic or content_topics set, SHOULD be regarded as an invalid request.

To filter on time range, the client MUST set time_start, time_end or both. Each time_ field should contain a Unix epoch timestamp in nanoseconds. An unset time_start SHOULD be interpreted as "from the oldest stored entry". An unset time_end SHOULD be interpreted as "up to the youngest stored entry".

If any of the content filter fields are set, namely pubsub_topic, content_topic, time_start, or time_end, the client MUST NOT set the message_hashes field.

Message hash lookup queries

A store query client MAY request the store service node to filter historical entries by one or more matching message hash keys. This type of query acts as a "lookup" against a message hash key or set of keys already known to the client.

In order to perform a lookup query, the store query client MUST populate the message_hashes field with the list of message hash keys it wants to lookup in the store service node.

If the message_hashes field is set, the client MUST NOT set any of the content filter fields, namely pubsub_topic, content_topic, time_start, or time_end.

Presence queries

A presence query is a special type of lookup query that allows a client to check for the presence of one or more messages in the store service node, without retrieving the full contents (values) of the messages. This can, for example, be used as part of a reliability mechanism, whereby store query clients verify that previously published messages have been successfully stored.

In order to perform a presence query, the store query client MUST populate the message_hashes field in the StoreQueryRequest with the list of message hashes for which it wants to verify presence in the store service node. The include_data property MUST be set to false. The client SHOULD interpret every message_hash returned in the messages field of the StoreQueryResponse as present in the store. The client SHOULD assume that all other message hashes included in the original StoreQueryRequest but not in the StoreQueryResponse is not present in the store.

Pagination info

The store query client MAY include a message hash as pagination_cursor, to indicate at which key-value entry a store service node SHOULD start the query. The pagination_cursor is treated as exclusive and the corresponding entry will not be included in subsequent store query responses.

For forward queries, only messages following (see sorting) the one indexed at pagination_cursor will be returned. For backward queries, only messages preceding (see sorting) the one indexed at pagination_cursor will be returned.

If the store query client requires the store service node to perform a forward query, it MUST set pagination_forward to true. If the store query client requires the store service node to perform a backward query, it SHOULD set pagination_forward to false. By default, therefore, the store service node assumes pagination to be backward.

A store query client MAY indicate the maximum number of matching entries it wants in the StoreQueryResponse, by setting the page size limit in the pagination_limit field. Note that a store service node MAY enforce its own limit if the pagination_limit is unset or larger than the service node's internal page size limit.

See pagination for more on how the pagination info is used in store transactions.

Store Query Response

In response to any StoreQueryRequest, a store service node SHOULD respond with a StoreQueryResponse with a requestId matching that of the request. This response MUST contain a status_code indicating if the request was successful or not. Successful status codes are in the 2xx range. Client nodes SHOULD consider all other status codes as error codes and assume that the requested operation had failed. In addition, the store service node MAY choose to provide a more detailed status description in the status_desc field.

Filter matching

For content filtered queries, an entry in the store service node matches the filter criteria in a StoreQueryRequest if each of the following conditions are met:

  • its content_topic is in the request content_topics set and it was published on a matching pubsub_topic OR the request content_topics and pubsub_topic fields are unset
  • its timestamp is larger or equal than the request start_time OR the request start_time is unset
  • its timestamp is smaller than the request end_time OR the request end_time is unset

Note that for content filtered queries, start_time is treated as inclusive and end_time is treated as exclusive.

For message hash lookup queries, an entry in the store service node matches the filter criteria if its message_hash is in the request message_hashes set.

The store service node SHOULD respond with an error code and discard the request if the store query request contains both content filter criteria and message hashes.

Populating response messages

The store service node SHOULD populate the messages field in the response only with entries matching the filter criteria provided in the corresponding request. Regardless of whether the response is to a forward or backward query, the messagesfield in the response MUST be ordered in a forward direction according to the message sorting rules.

If the corresponding StoreQueryRequest has include_data set to true, the service node SHOULD populate both the message_hash and message for each entry in the response. In all other cases, the store service node SHOULD populate only the message_hash field for each entry in the response.

Paginating the response

The response SHOULD NOT contain more messages than the pagination_limit provided in the corresponding StoreQueryRequest. It is RECOMMENDED that the store node defines its own maximum page size internally. If the pagination_limit in the request is unset, or exceeds this internal maximum page size, the store service node SHOULD ignore the pagination_limit field and apply its own internal maximum page size.

In response to a forward StoreQueryRequest:

  • if the pagination_cursor is set, the store service node SHOULD populate the messages field with matching entries following the pagination_cursor (exclusive).
  • if the pagination_cursor is unset, the store service node SHOULD populate the messages field with matching entries from the first entry in the store.
  • if there are still more matching entries in the store after the maximum page size is reached while populating the response, the store service node SHOULD populate the pagination_cursor in the StoreQueryResponse with the message hash key of the last entry included in the response.

In response to a backward StoreQueryRequest:

  • if the pagination_cursor is set, the store service node SHOULD populate the messages field with matching entries preceding the pagination_cursor (exclusive).
  • if the pagination_cursor is unset, the store service node SHOULD populate the messages field with matching entries from the last entry in the store.
  • if there are still more matching entries in the store after the maximum page size is reached while populating the response, the store service node SHOULD populate the pagination_cursor in the StoreQueryResponse with the message hash key of the first entry included in the response.

Security Consideration

The main security consideration to take into account while using this protocol is that a querying node have to reveal their content filters of interest to the queried node, hence potentially compromising their privacy.

Future Work

  • Anonymous query: This feature guarantees that nodes can anonymously query historical messages from other nodes i.e., without disclosing the exact topics of 14/WAKU2-MESSAGE they are interested in.
    As such, no adversary in the WAKU2-STORE protocol would be able to learn which peer is interested in which content filters i.e., content topics of 14/WAKU2-MESSAGE. The current version of the WAKU2-STORE protocol does not provide anonymity for historical queries, as the querying node needs to directly connect to another node in the WAKU2-STORE protocol and explicitly disclose the content filters of its interest to retrieve the corresponding messages. However, one can consider preserving anonymity through one of the following ways:
    • By hiding the source of the request i.e., anonymous communication. That is the querying node shall hide all its PII in its history request e.g., its IP address. This can happen by the utilization of a proxy server or by using Tor. Note that the current structure of historical requests does not embody any piece of PII, otherwise, such data fields must be treated carefully to achieve query anonymity.
    • By deploying secure 2-party computations in which the querying node obtains the historical messages of a certain topic, the queried node learns nothing about the query. Examples of such 2PC protocols are secure one-way Private Set Intersections (PSI).
  • Robust and verifiable timestamps: Messages timestamp is a way to show that the message existed prior to some point in time. However, the lack of timestamp verifiability can create room for a range of attacks, including injecting messages with invalid timestamps pointing to the far future.
    To better understand the attack, consider a store node whose current clock shows 2021-01-01 00:00:30 (and assume all the other nodes have a synchronized clocks +-20seconds). The store node already has a list of messages, (m1,2021-01-01 00:00:00), (m2,2021-01-01 00:00:01), ..., (m10:2021-01-01 00:00:20), that are sorted based on their timestamp.
    An attacker sends a message with an arbitrary large timestamp e.g., 10 hours ahead of the correct clock (m',2021-01-01 10:00:30). The store node places m' at the end of the list, (m1,2021-01-01 00:00:00), (m2,2021-01-01 00:00:01), ..., (m10:2021-01-01 00:00:20), (m',2021-01-01 10:00:30). Now another message arrives with a valid timestamp e.g., (m11, 2021-01-01 00:00:45). However, since its timestamp precedes the malicious message m', it gets placed before m' in the list i.e., (m1,2021-01-01 00:00:00), (m2,2021-01-01 00:00:01), ..., (m10:2021-01-01 00:00:20), (m11, 2021-01-01 00:00:45), (m',2021-01-01 10:00:30). In fact, for the next 10 hours, m' will always be considered as the most recent message and served as the last message to the querying nodes irrespective of how many other messages arrive afterward.

A robust and verifiable timestamp allows the receiver of a message to verify that a message has been generated prior to the claimed timestamp. One solution is the use of open timestamps e.g., block height in Blockchain-based timestamps. That is, messages contain the most recent block height perceived by their senders at the time of message generation. This proves accuracy within a range of minutes (e.g., in Bitcoin blockchain) or seconds (e.g., in Ethereum 2.0) from the time of origination.

Copyright

Copyright and related rights waived via CC0.

References

  1. 14/WAKU2-MESSAGE
  2. protocol buffers v3
  3. 11/WAKU2-RELAY
  4. Open timestamps
  5. 13/WAKU2-STORE v2 previous version