Repository navigation
Implement Social Discovery to add found DIs to the Gestalt Graph #316
Description
Activity
From this comment (https://gitlab.com/MatrixAI/Engineering/Polykey/js-polykey/-/merge_requests/144#note_501235581) it looks like the
identitiesInfoGetConnectedGRPC handler (which isn't currently used by any CLI commands) might be intended to be used for social discovery, and thus growing the gestalt graph. Right now all it does is return data about connected digital identities, however, if these identities were added into the gestalt graph we would have a way of growing it. However, we would also need to scrape these found identities for node claims, since a gestalt consisting of just an identity is pretty useless to us (since we need nodes in order to set permissions for a gestalt).Once you have figured out what needs to be changed, it's better to break down this issue from a general "review issue" to focused sub issues on things that need to be changed. We can do a meeting to review the details.
Looks like my hunch about
identitiesInfoGetConnectedwas right - https://gitlab.com/MatrixAI/Engineering/Polykey/js-polykey/-/merge_requests/144#note_501361799So the process of growing the gestalt graph is as follows:
- When first bootstrapping, your own node should be added to the gestalt graph
- After bootstrapping, you HAVE to augment one of your digital identities. This is the only way to be able to eventually discover other gestalts. This augmented DI should also be added to the gestalt graph
- Anytime you claim another node, this node should be added to the gestalt graph
- Provided you've done step 2 already, you can now call
identitiesInfoGetConnectedto search the provider of your DI for other identities. These found identities should be set into the gestalt graph andDiscovery.discoverGestaltByIdentity()should be called on them in order to discover their gestalts - The discovery functions can also be called at any time to flesh out the links between nodes and identities in the gestalt graph
Possibility for the gestalt graph to be edited manually: https://gitlab.com/MatrixAI/Engineering/Polykey/js-polykey/-/merge_requests/195#note_600203842 - not sure if this functionality is still intended to exist
I was actually incorrect in my original understanding of the discoverGestaltBy* functions - I had originally assumed that discovery did NOT set nodes/identities into the gestalt graph and instead only created links between them in the graph, however this is incorrect. Discovery does in fact set discovered nodes and identities into the gestalt graph and so this is a way of growing the gestalt graph.
The caveat to this is that the setting of nodes and identities only happens when a link is found, so you cannot set a new node/identity with no links into the gestalt graph through discovery.
Discovery is meant to be for discovering other gestalts, not your own. In this case, we can assume that the
Discoveryfunctions won't be used for updating our own gestalt in the gestalt graph. In this case, whenever we make a node or identity claim, the claimed node/identity needs to be put into our gestalt graph manually (along with the relevant links). This can be done usinglinkNodeAndNode()andlinkNodeAndIdentity(), since these methods also set the nodes/identities into the gestalt graph. This could be added to the grpc handlers foridentitiesClaimandnodesClaim.Right now, all
identitiesInfoGetConnected()(the grpc handler) does is return connected identity datas (the definition of this depends on the provider and doesn't matter sinceProvideris an abstract class). From my understanding after reading through the inital Gestalt Graph discussion,identitiesInfoGetConnected()should also be callingGestaltGraph.setIdentity()andDiscovery.discoverGestaltByIdentity()for each of the found connected identities. The gestalt graph is a small domain so it seems like it shouldn't be too much trouble to inject that intoidentitiesInfoGetConnected(), however discovery is much larger so it's probably best not to. Either we can assume that the user will call discovery themselves on the found identities, or we could potentially use the new event emitter to call discovery every time a new connected identity is found?We also need our node to be added to the gestalt graph when bootstrapping, which is easy enough to add to
bootstrapState. However, wouldn't this node need to be updated whenever we change the root keypair? I think this relates to #317.Not only would the node itself need to be updated in the gestalt graph, but all of the other nodes/identities in its gestalt, since they would now be linked to a node with a different ID. https://gitlab.com/MatrixAI/Engineering/Polykey/js-polykey/-/merge_requests/195#note_630611063 - this comment, in particular the line "This is by design, node id changes induces a need to recreate claims. Which is why key changes imply a change of identity", makes me wonder if key pair changes should trigger links in the gestalt graph to be removed? This could probably be a separate issue though.
Discovery is meant to be for discovering other gestalts, not your own. In this case, we can assume that the
Discoveryfunctions won't be used for updating our own gestalt in the gestalt graph. In this case, whenever we make a node or identity claim, the claimed node/identity needs to be put into our gestalt graph manually (along with the relevant links). This can be done usinglinkNodeAndNode()andlinkNodeAndIdentity(), since these methods also set the nodes/identities into the gestalt graph. This could be added to the grpc handlers foridentitiesClaimandnodesClaim.I believe discovery was meant to be a social discovery (not DHT discovery which is supposed to occur automatically in kademlia/nodegraph). And your own gestalt doesn't need to be discovered simply because it's already linked in your sigchain.
However the node graph may need to be populated when you gestalt updates.
This is where the gossip protocol is meant to come in #190, but I haven't fully specced out how that's supposed to work.
I believe discovery was meant to be a social discovery (not DHT discovery which is supposed to occur automatically in kademlia/nodegraph). And your own gestalt doesn't need to be discovered simply because it's already linked in your sigchain.
However the node graph may need to be populated when you gestalt updates.
This is where the gossip protocol is meant to come in #190, but I haven't fully specced out how that's supposed to work.
So does this mean that your own gestalt doesn't need to be represented in your gestalt graph, only other people's gestalts? Since this information is stored on your sigchain? This would simplify things a lot considering the gestalt graph is currently not updated/modified by any other domains (besides discovery). The gestalt graph also currently doesn't update or communicate with any other domains besides the ACL. The gestalt graph at the moment is completely cut off from nodes/identities/sigchain and there's no direct communication between them.
From the original gestalt graph discussion it seems that your own gestalt was expected to be represented in your gestalt graph, but I don't think there's a reason that it needs to be?
No your own gestalt would still in the gestalt graph.
Identity related manipulation should mutate the gestalt graph.
The gestalt graph would be consulted when identity decisions and trust decisions, it would be shown to the end user. This is hard to visualise until you see the GUI. The CLI is quite barebones on its visualisation.
So the final checklist of what needs to be done to get the relevant tests working as expected is:
- Create a CLI command for
identitiesInfoGetConnected - Inject
GestaltGraphintoidentitiesInfoGetConnectedand callsetIdentity()on each connected identity - Either inject
DiscoveryintoidentitiesInfoGetConnectedand calldiscoverGestaltByIdentity()on each connected identity OR set up an event for discovering an identity that triggersdiscoverGestaltByIdentity()wheneversetIdentity()is called OR expect the user to manually run theidentities discovercommand on the identities that are returned byidentitiesInfoGetConnected
Additionally (as a separate issue most likely), in order to have your own gestalt represented in your gestalt graph we need to:
- Add newly bootstrapped nodes into the gestalt graph inside
bootstrapState() - Modify node claiming to add the claimed node to the gestalt graph for both nodes (could potentially just inject GestaltGraph into
nodesClaimto do this for the node that callsnodes claim ...by callinglinkNodeAndNode(), but it's likely more complicated for the other node) - Modify identity claiming to add the claimed identity to the gestalt graph (can just inject
GestaltGraphintoidentitiesClaimand calllinkNodeAndIdentity()) - We also need to consider expected behaviour when the root keypair is changed and when claims otherwise expire/become invalid and what we expect to happen on the gestalt graph. This could be something we use the event bus for.
So there are really two sub issues here:
- Fixing and integrating the
identitiesInfoGetConnectedrpc handler into the CLI - Adding the ability for the gestalt graph to contain your own gestalt
The first issue is the only one blocking #311, the second issue is more complex and can also likely be done alongside #317 (mostly point 4).
- Create a CLI command for
I think this issue can probably be renamed and cover the
identitiesInfoGetConnectedissues since those relate more closely to test splitting (which is what this issue was created from). I can then create a separate issue for the "gestalt graph containing own gestalt" issues, which can be addressed later. If I were to create a new branch for fixingidentitiesInfoGetConnectedwould that branch off from master or from the test splitting branch @CMCDragonkai?If you're just making changes to the src, branch from master first. You can always rebase later.
The overall purpose of the gestalt graph is create and maintain a graph of trust. So you know that you're sharing a vault with a gestalt and not a single node. So all the other pieces come together to enable this. The GG is also eventually consistent and also completely decentralised. Every node in a gestalt should share the gestalt graph DB via the gossip protocol but other nodes may have a completely different gestalt graph simply because their field of view is different.
I'll circle back to this once nodes and vaults are working and we can review in detail.
This is a major part of the decentralised trust arm of PK.
Btw alot of gestalt graph theory was written into the polykey-design issues in gitlab. And was meant to have fully realised in our gestalts wiki article. You might want to refer back to the polykey-design repo too.
- changed the title
[-]Review Gestalt Graph CLI Integration[/-][+]Extending and exposing the `identitiesInfoGetConnected` RPC Handler[/+]on Jan 21, 2022 - changed the title
[-]Extending and exposing the `identitiesInfoGetConnected` RPC Handler[/-][+]Implement Social Discovery to add found DIs to the Gestalt Graph[/+]on Jan 23, 2022 I want to point out a few related issues to this:
- Decentralized Indexing of Gestalt Discovery #223 - this is about sharing gestalt graph indexing information across the PK network, either within a gestalt, or between gestalts
- Gestalt Synchronisation with Gossip Protocol #190 - this is about sharing acl, permission information and other metadata between nodes of a single gestalt
- Implement ACL synchronisation across keynodes in gestalt #185 - a subissue of Gestalt Synchronisation with Gossip Protocol #190
- Indexing Domains and DB Levels for Faster & Flexible Lookup #188 - this will improve performance of discovery and sigchain operations
During our discussion about discovery, and how it should be a background asynchronous spidering process. I was thinking of making use of some async abstractions to implement this:
- https://github.com/toolbuilder/await-for-it
- https://github.com/toolbuilder/await-for-it/blob/master/docs/queue.md
Unlike the
EventBus, things can trigger discovery by queueing up discovery tasks, but this is all done asynchronously asynchronous, so they don't wait for the discovery to finish. The discovery system allows eventual consistency of the gestalt graph.Everything from this issue has been resolved in #320, except for adding nodes/identities to the Gestalt Graph during vault sharing (since this has to wait until the vaults work being done in #266 is completed).
Now we have the following features in order to implement social discovery:
- To search for connected identities, run the command
identities search- under the hood this calls eitheridentitiesInfoGetfor finding information about specific identities oridentitiesInfoConnectedGetto get information about all connected identities across one or more providers- The result of this command will be a list of identities with all of the available information on each identity found
- Once you have found an identity that you want to trust, you can set the
notifypermission for it by running the commandidentities trust [identity](this also works for nodes) - under the hood this callsgestaltsGestaltTrustByIdentity(orgestaltsGestaltTrustByNode), which sets the identity/node into the gestalt graph and sets the permission for it. This command also queues the identity/node for discovery in order to find the rest of the linked gestalt. - Discovery is done automatically and unassisted in the background at all times (while there are identities/nodes queued for discovery), however, you can also manually choose to discover a gestalt by running the command
identities discover [identity/node], which will queue the identity/node for discovery and eventually add it to the gestalt graph
With the combination of these features, the gestalt graph will be maintained with the current state of all of the gestalts you trust.
- To search for connected identities, run the command
Specification
Currently, most of the Gestalts CLI commands do not work in isolation as they expect nodes and identities to have already been set in the Gestalt Graph. The commands will fail with an
ErrorGestaltsGraphNodeIdMissingorErrorGestaltsGraphIdentityIdMissingerror, or will return no data.This is because(outdated)identitiesInfoGetConnected, which is supposed to add found connected identities to the gestalt graph, is currently incomplete and is inaccessible to the CLI.When choosing to trust a node/identity (i.e. to send notifications/vault share), this should add the node/identity to our gestalt graph. This can be done inside the handlers for these commands. Once done, this should trigger discovery on the added node/identity.
We don't want to have to await this discovery, so the
Discoverydomain needs to be converted fromCreateDestroytoCreateDestroyStartStopso that it can have underlying state that it can crawl in the background.Additional context
Tasks
identitiesInfoGetConnectedto the CLI