# PQs and Caches for ACA-Py and Mediators ## Mediators ### Design 1: No Relay/Deliverer, No Controller A "no controller deployment" is viable by either setting the `--auto` flags, or by extending the mediation negotiation plugin to include business logic, so that there is no need for a controller. ```plantuml cloud Internet as ags { node agent1 as a1 node agent2 as a2 node agent3 as a3 } cloud Wallets { node wallet1 as w1 node wallet2 as w2 node wallet3 as w3 } component "Scalable Mediator" { component "Mediator" { () "External\nLoad Balancer" as lb node M1 as m1 node M2 as m2 node M3 as m3 } component "Redis" as red database "Askar\nDatabase" as ask } a1 --> lb a2 --> lb a3 --> lb w1 --> lb w2 --> lb w3 --> lb lb --> m1 lb --> m2 lb --> m3 m1 <--> ask m2 <--> ask m3 <--> ask m1 <--> red m2 <--> red m3 <--> red ``` See [Overview](#Mediator-Overview) below for more details about both designs. ### Design 2: Combined Relay/Deliverer ```plantuml cloud Internet as ags { node agent1 as a1 node agent2 as a2 node agent3 as a3 } cloud Wallets { node wallet1 as w1 node wallet2 as w2 node wallet3 as w3 } component "Scalable Mediator" { component "Mediator" { node M1 as m1 node M2 as m2 node M3 as m3 } component "Redis" as red database "Askar\nDatabase" as ask component "Relay/Deliverer" as rd { () "External\nLoad Balancer" as lb node RD1 as rd1 node RD2 as rd2 node RD3 as rd3 } } a1 --> lb a2 --> lb a3 --> lb w1 --> lb w2 --> lb w3 --> lb w1 -right-> ags w2 -right-> ags w3 -right-> ags lb --> rd1 lb --> rd2 lb --> rd3 rd1 <--> red rd2 <--> red rd3 <--> red m1 <--> ask m2 <--> ask m3 <--> ask m1 <--> red m2 <--> red m3 <--> red ``` ### Overview * The "agents" only send messages destined for wallets. * They are received and queued on Redis with a topic based on the recipient wallet. * Each wallet connects with WS to a mediator instance and when connected, has a session with that mediator endpoint. * In establishing a web socket session, the agent sends a DIDComm message to pickup messages. * The wallet that establishes the web socket and sends the DIComm message is not known until the DIDComm message from the wallet is processed. * If the endpoint is managed from a different component (e.g., a Relay/Deliverer instance) than is processing the DIDComm message (e.g., an ACA-Py/Mediator instance), the two components must coordinate about which recipient is processing the messages for what wallet. * Proposal is as follows: * Two identifiers are needed, a tag for the web socket session, and a Wallet identifier The wallet identifier is the same identifier the ACA-Py/Mediator (AM) instances will use in queing messages destined to a wallet to Redis. * When a Relay/Deliverer (RD) instance receives a message to establish a new web socket session, the RD tags the message with a locally understood session identifier, puts it on the Redis queue, and adds the tag to a list of things on the Redis queue it is tracking. * An AM instance retrieves the "new session" message from the Redis queue, noting the tag from the RD, processes the DIDComm message and extracts the wallet ID. * The AM queues a message on Redis with the topic being the RD-generated tag and the content being the Wallet ID. * When the RD instance sees the message with the tag, it retrieves the message, and adds the Wallet Identifier to the list of things on the Redis queue it is tracking. * While the web socket is active, the RD monitors Redis for messages for the wallet identifier and sends them to the wallet using the web socket session. * When the web socket is lost, the RD instance stops looking on Redis for the wallet ID messages. * If the RD dies, the wallet will establish a new web socket connection, and that RD instance will take over sending messages to that wallet. * Wallets only retrieve messages from the mediator. * Exception: When a wallet is negotiating with the mediator, messages are sent from the wallet to the mediator, and from the mediator to the wallet. * In such cases, it is likely the business logic is involved to make decisions on the mediation setup -- accept the wallet or not, at minimum. * Messages from the Wallet should be sent to the mediator, as if it was any other agent. * Messages from the Mediator would be sent to the wallet as if sourced from an external agent. ### Logic * On receipt of a message destined for a wallet, the mediator instance checks if it has a WS session with the wallet. If not, it puts the message on the Redis queue. * If the mediator is the source of the message, it would process the outbound message to a wallet as if it had already been retrieved from an external agent. * When connected with wallet(s), an agent polls the Redis queue for messages for wallet(s), and when found, retrieves message from the queue and send the message to the wallet. ### Other Issues * There is no confirmation of a message being received by the wallet, which could lead to messages being lost because of an undetected lost connection. * The mediator should handle mobile notifications to ## ACA-Py ### Design 1: Separate Relay and Deliverer Components ```plantuml cloud Internet { node agent1 as a1 node agent2 as a2 node agent3 as a3 } component "Scalable Aries Agent" { component "ACA-Py" as ap { node "AP1" as ac1 node "AP2" as ac2 node "AP3" as ac3 } component "Deliverer" as del { node "D1" as d1 node "D2" as d2 node "D3" as d3 } component "Relay" { () "External Agent\nLoad Balancer" as lb () "Admin API\nLoad Balancer" as alb node R1 as r1 node R2 as r2 node R3 as r3 } component "Redis" as red database "Askar\nDatabase" as ask component Controller as ct { () "Admin WebHook\nLoad Balancer" as awhlb node "C1" as ct1 node "C2" as ct2 node "C3" as ct3 } database "Controller\nDB" as ctdb } a1 -down-> lb a2 --> lb a3 --> lb lb --> r1 lb --> r2 lb --> r3 r1 --> red r2 --> red r3 --> red red --> d1 red --> d2 red --> d3 del -up-> a1 del --> a2 del --> a3 del --> awhlb ac1 <--> red ac2 <--> red ac3 <--> red ac1 <-right-> ask ac2 <--> ask ac3 <--> ask awhlb -->ct1 awhlb -->ct2 awhlb -->ct3 ct1 --> alb ct2 --> alb ct3 --> alb alb --> r1 alb --> r2 alb --> r3 ct1 <-right-> ctdb ct2 <-right-> ctdb ct3 <-right-> ctdb ``` ### Design 2: Combined Relay and Deliverer Component ```plantuml cloud Internet { node agent1 as a1 node agent2 as a2 node agent3 as a3 } component "Scalable Aries Agent" { component "ACA-Py" as ap { node "AP1" as ac1 node "AP2" as ac2 node "AP3" as ac3 } component "Relay/Deliverer" as rd { () "External Agent\nLoad Balancer" as lb () "Admin API\nLoad Balancer" as alb node RD1 as rd1 node RD2 as rd2 node RD3 as rd3 } component "Redis" as red database "Askar\nDatabase" as ask component Controller as ct { () "Admin WebHook\nLoad Balancer" as awhlb node "C1" as ct1 node "C2" as ct2 node "C3" as ct3 } database "Controller\nDB" as ctdb } a1 --> lb a2 --> lb a3 --> lb lb --> rd1 lb --> rd2 lb --> rd3 rd1 <--> red rd2 <--> red rd3 <--> red rd -up-> a1 rd -up-> a2 rd -up-> a3 rd -down-> awhlb ac1 <--> red ac2 <--> red ac3 <--> red ac1 <--> ask ac2 <--> ask ac3 <--> ask awhlb -->ct1 awhlb -->ct2 awhlb -->ct3 ct1 --> alb ct2 --> alb ct3 --> alb alb --> rd1 alb --> rd2 alb --> rd3 ct1 <--> ctdb ct2 <--> ctdb ct3 <--> ctdb ``` ### Overview ### Optional Deployment: Controller Shares Relay/Deliverer To make a cloud native controller scalable, have it use Redis and the Relay / Deliverer. Webhooks arrive into the Relay, are put into Redis and retrieved from Redis by the Controller instances, who process the requests. Results of the controller processing that result in requests to the ACA-Py Admin API are put into Redis, where the deliverer picks them up and sends them to ACA-Py. An optimization would be to have the controller put them in Redis such that ACA-Py picks them up directly, without the deliverer involved. Likewise, the webhooks would go into Redis from ACA-Py where they are picked up by the Controller without the deliverer or relay involvement. ```plantuml cloud Internet { node agent1 as a1 node agent2 as a2 node agent3 as a3 } component "Scalable Aries Agent" { component "ACA-Py" as ap { node "AP1" as ac1 node "AP2" as ac2 node "AP3" as ac3 } component "Relay/Deliverer" as rd { () "External Agent\nLoad Balancer" as lb () "Admin API\nLoad Balancer" as alb () "Admin WebHook\nLoad Balancer" as awhlb node RD1 as rd1 node RD2 as rd2 node RD3 as rd3 } component "Redis" as red database "Askar\nDatabase" as ask component Controller as ct { node "C1" as ct1 node "C2" as ct2 node "C3" as ct3 } database "Controller\nDB" as ctdb } a1 --> lb a2 --> lb a3 --> lb lb --> rd1 lb --> rd2 lb --> rd3 rd1 <--> red rd2 <--> red rd3 <--> red rd --> a1 rd --> a2 rd --> a3 rd --> awhlb rd --> alb ac1 <--> red ac2 <--> red ac3 <--> red ac1 <--> ask ac2 <--> ask ac3 <--> ask awhlb -->rd1 awhlb -->rd2 awhlb -->rd3 ct1 <--> red ct2 <--> red ct3 <--> red alb --> rd1 alb --> rd2 alb --> rd3 ct1 <--> ctdb ct2 <--> ctdb ct3 <--> ctdb ```