Orchestrating Other Teams' Agents: Routing, Fault Tolerance, and Push Notifications
Move the centralized orchestration from Day 2 onto A2A and call the agents of four teams. Route by skill, and fail loudly instead of picking someone at random when nothing matches. Retry on timeout, but reuse the same message ID so the other side never does the work twice. When one team goes down, deliver an honest result that says what is missing. When the user cancels, carry the cancellation to every downstream task. Finally, replace long-held connections with push notifications and block the SSRF hiding in the push URL.
Today's Goals
- Can route to remote agents by skill, and use message ID deduplication so that retrying after a timeout never duplicates work
- Can deliver a result that states what is missing and why when some downstream agents fail, and carry a cancellation to every in-flight remote task
- Can implement push notification registration and delivery, and explain why the push URL is an SSRF entry point and at which two places the server has to block it
For five days you talked to one remote agent at a time. Today all four teams join in. The orchestration logic was finished on Day 2; everything new today comes from one fact: the other side is not in your process. When you finish, come back to the top and check off the three goals.
Plain-Language Walkthrough
Hiring contractors: pick by license, chase without double orders, deliver even when one falls through
The annual company offsite has been handed to four contractors: an events company for the venue, a caterer for the food, a coach company for transport, and an outside accountant to confirm prices. You coordinate. Four problems show up at once.
First, whom to ask. You do not give the catering order to Wang because he did a good job last time; Wang works for the events company. You go by the business license: whoever is licensed for food service gets the catering. If nobody in your contacts is, you tell your boss that nobody can take this on. You do not quietly hand it to the events company.
Second, chasing. You sent the caterer an order and heard nothing for hours. You cannot tell whether they never got it, or got it, booked it, and simply never called back. Sending it again is right, but quote the same order number, or they will treat it as a new order and book twice.
Third, one of them falls through. The coach company stops answering. Your report to the boss cannot say everything is arranged, and you cannot make something up either, such as "just use the venue shuttle." You are not the coach company and do not know whether a shuttle is free that day. What you deliver is: venue, catering, and venue fee are settled; transport is open, because the coach company cannot be reached.
Fourth, the boss cancels the offsite. Hanging up your own phone is not enough. You call every contractor who has already started: stop booking, stop scheduling.
Cross-team orchestration is these four problems written as code. The orchestration itself, choose the venue, send round two with a brief, then summarize, was written on Day 2 and does not change by a single line today. Each delegation simply stops being a function call and becomes an A2A call across the network.
Sign in, then link your WeChat mini program account on the account page. Linked accounts share one free article per day. Unlock additional articles with rewarded videos in the mini program.
Scan in WeChat to open Dayward AI. Open the same course and chapter, choose the rewarded video, then refresh this page after completion.
Interview questions
When a remote call times out and you retry, can the other side end up doing the same work twice? How do you design against it?远端调用超时重试，会不会让对方把同一件事做两遍？怎么设计？
Common in ChinaCommon overseasIntermediate#a2a#idempotency#retry🔒 Unlock this chapter to see the answer keys.
One of four downstream agents is down. What result should the orchestrator deliver?四个下游里挂了一个，编排方该交出什么样的结果？
Common in ChinaCommon overseasDeep dive#multi-agent#partial-failure#reliability🔒 Unlock this chapter to see the answer keys.
What are the security risks of push notifications, and how should the server defend against them?推送通知有哪些安全风险？服务端该怎么防？
Common in ChinaCommon overseasDeep dive#a2a#push-notifications#ssrf🔒 Unlock this chapter to see the answer keys.
The user cancels. How do you propagate the cancellation to every downstream agent?用户取消了，怎么把取消传到每一个下游？
Common in ChinaCommon overseasIntermediate#a2a#cancellation#multi-agent🔒 Unlock this chapter to see the answer keys.