GitLab

fix: remove local provider check

fix: use the routing table filter

It doesn't actually work. Blocks are stored under /blocks and have been since
2015.

Furthermore, this is just wrong. We shouldn't go digging through the datastore
to see if we happen to have the block. If we want something like this, we should
include a blockstore in the provider manager.

Unfortunately, constructing a router is _not_ cheap. Cache it so we don't have
to re-compute it every time we get a query.

fixes #575

fix: handle empty keys

Handle empty keys, both when sent in RPC requests and in the local API.

Set record handlers for the default protocol prefix

Users can override the pk and ipns default namespaced validators with other validators, however they are of an invalid type then they will fail if using the default /ipfs protocol prefix

DHT defaults to adding a namespaced pk and ipns validator unless the Validator has been changed, in which case neither of them are added. Simply adding additional NamespacedValidators does not remove the pk or ipns validators

DHT defaults to adding namespaced pk and ipns validators unless the validator defaults are modified in which case neither of them are added

fix incorrect error handling during provider record lookups

Proposed DHTv2 Changes

test: fix TestFindPeerQuery and TestFindPeerQueryMinimal to not try and refresh routing tables for orphaned nodes

We no longer need this now that we explicitly support v1 compatibility mode.

When this option is passed, the DHT node will listen on and query _only_ using
the old DHT protocol. Importantly, the node won't even pretend to be a new DHT
node because it's routing table includes V1 peers.

The querying node _is_ allowed to reset and go away.

fixes https://github.com/libp2p/go-libp2p-kad-dht/issues/431

* Remove deprecated events.
* Fix warning -> warn.
* Use structured logging (some places).

Except deprecated warn calls

* record valuable peers algorithm only takes into account latency of seed peers

* fix Low Peers and Cleanup

* new RT management policy

Previously, we'd only return the target peer if we were connected to them.
However, this makes it difficult to impossible to find peers that are mostly
disconnected from the network.

This change also changes `p` to `from` in several places as `p` is _very_
confusing. We should probably switch away from using `p` everywhere.

* Add Kademlia async lookup implementation (replaces prior implementation)
* Add lookup events that allow full lookup reconstruction from logs
* Introduce lookup stopping condition, based on parameter "beta"
Co-authored-by: Adin Schmahmann <adin.schmahmann@gmail.com>