* feat!: do not send sync messages with empty history
A sync message without any history as no value. If there are no messages in the channel, then a sync messages does not help.
If there are messages in the channel, but this participant is not aware of them, then it can confuse other participants to assume that the channel is empty.
* fix test by adding a message to channel history
* make `pushOutgoingSyncMessage` return true even if no callback passed
* fix!: avoid SDS lamport timestamp overflow
The SDS timestamp is initialized to the current time in milliseconds, which is a 13 digits value (e.g. 1,759,223,090,052).
The maximum value for int32 is 2,147,483,647 (10 digits), which is clearly less than the timestamp.
Maximum value for uint32 is 4,294,967,295 (10 digits), which does not help with ms timestamp.
uint64 is BigInt in JavaScript, so best to be avoided unless strictly necessary as it creates complexity.
max uint64 is 18,446,744,073,709,551,615 (20 digits).
Using seconds instead of milliseconds would enable usage of uint32 valid until the year 2106.
The lamport timestamp is only initialized to current time for a new channel. The only scenario is when a user comes in a channel, and thinks it's new (did not get previous messages), and then starts sending messages. Meaning that there may be an initial timestamp conflict until the logs are consolidated, which is already handled by the protocol.
* change lamportTimestamp to uint64 in protobuf
* lamport timestamp remains close to current time
* feat: query on connect stops on predicate
* test: query on connect stops at predicate
* feat: reliable channels search up to 30 days to find message
Queries stop once a valid sync or content message is found in the channel.
* fix: protect against decoding exceptions
* stop range queries on messages with a causal history
* SDS: pushOutgoingMessage is actually sync
* SDS: ensure that `ContentMessage` class is stored in local history with `valueOf` method
* feat: introduce reliable channels
Easy to use Scalable Data Sync (SDS, e2e reliability) wrapper, that includes:
- store queries upon connection to store nodes
- store queries to retrieve missing messages
* remove `channel` prefix
* attempt to improve performance when processing a lot of incoming messages
* test: split test file
* use index.ts for re-export only.
* improve if condition
* use getter for isStarted
* waku node already auto-start
* rename send
* fix lightPush.send type post rebase
* test: remove extra console.log
* SDS: emit messages as missing as soon as they are received
* make configurable elapse time for task process
* typo
* use string instead of enum for event types
* ReliableChannel.send returns the message id
* SDS: export `MessageId`
* SDS: attach retrieval hints to incoming messages
* sds: ensure items are ordered by timestamp
* test: sds: avoid using "as any" as it bypasses type checks
* test: filter: avoid using "as any" as it bypasses type checks
* test: fix tests without introducing proxy
* feat: query on connect
Perform store time-range queries upon connecting to a store node.
Some heuristics are applied to ensure the store queries are not too frequent.
* make `maybeQuery` private
* query-on-connect: use index.ts only for re-export
* query-on-connect: update doc
* store connect evt: use enum instead of free strings for Waku event types
* store connect evt: more accurate enum name
* store connect evt: add store connect event on peer manager
* store connect evt: simplify logic statements
* store connect evt: test store connect
* store connect evt: export event types
* test: use enum
* Shorter name for waku events
* update local peer discovery, make it configurable for cache
* move to separate file
* up tests, remove local storage from tests
* pass local peer cache options
* add e2e tests
* add aditional e2e tests for local cache
* rename local-peer-cache into peer-cache
* update tests, ci
* prevent filterign ws addresses
* rename, mark stuff to remove
* rename type
* update exports from discovery package, improve PX implementation
* re-structure px discovery, comment tests
* implement UT, E2E Tests, clean code a bit, implement recurring PX discovery, move DNS E2E Tests to separate folder, remove not needed E2E tests
* fix discovery dialer e2e test
* mark as started
* fix dns tests
* attempt to fix e2e
* skip test
* update tests
* fix typo
* add catch all in stream manager, update tests
* update mock
* update test
* introduce `MessageId` type
# Conflicts:
# packages/sds/src/message_channel/message_channel.ts
* fix: own messages are not used for ack
* fix: own messages are not used for ack
* doc: long term solution is SDS protocol change
* SDS: renaming to match message function
* SDS: introduce `Message` class for easier encoding/decoding
# Conflicts:
# packages/sds/src/message_channel/events.ts
# packages/sds/src/message_channel/message_channel.ts
* SDS Message is a class now
* SDS: it's "possibly" not "partially" acknowledged.
* SDS: TODO
* SDS: fix tests
* SDS: make logs start with `waku`
* SDS: add bloom filter test
# Conflicts:
# packages/sds/src/message_channel/events.spec.ts
* SDS: improve naming
* SDS: improve naming
Messages are not "sent" or received, but pushed for processing in local queues.
* SDS: sync message should not be delivered
* SDS: renaming from earlier
* SDS: remove useless variable
* SDS: Fix comment
* SDS: sync messages do not get "delivered"
* SDS: acks
* SDS: simplify delivered event
* SDS: improve event naming
* SDS: fix comment
* SDS: make task error an official event
* SDS: Mark messages that are irretrievably lost
* SDS: remove default for irretrievable and simplify config
* SDS: typo on sync event
* SDS: add and user sender id
* SDS: resent message never get ack'd
* SDS: fix cylic dependencies
* SDS: helpful logs
* SDS: avoid duplicate history entries
* SDS: export options
* introduce `MessageId` type
* fix deprecated import
* test: own messages are not used for acks
* fix: own messages are not used for ack
* fix: own messages are not used for ack
* test: do not self-possible-ack
* doc: long term solution is SDS protocol change
Nodes returned by DNS Discovery are not guaranteed to be reachable.
Hence, setting an upfront limit based on sought capability does not provide
any guarantees that such capability should be reached.
The discovery's mechanism role is to find Waku 2 nodes. As many as possible.
It is then the role of the peer manager to decide:
- whether to attempt connecting to the node based on advertised capabilities
- retain connection to the node based on actual mounted protocols.
We still want to prevent infinite loops, hence the `maxGet` parameter.
Also, there was a dichotomy between code tested, and code actually used by libp2p peer discovery, now resolved.
# Conflicts:
# packages/discovery/src/dns/constants.ts
