Replies: 1 comment
-
If you use long running tools (https://google.github.io/adk-docs/tools/function-tools/#2-long-running-function-tool), it will end the event loop and give you the info about the tool. After a few minutes, hours or days, you can resume the conversation by sending the tool response to the agent. This is working sort like a pub/sub system. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
While ADK's event-loop design is powerful and intuitive, it poses challenges for long-running tools. Having a generator function waiting to yield an event makes it difficult to scale ADK instances up and down reliably, since the execution time can exceed a reasonable grace period.
I'm curious to figure out how others are handling this.
For example, could long-running tools publish a message to a Pub/Sub topic once complete, and have that message pushed into the event loop?
Maybe I'm missing something but I think in order to scale robustly we will need to figure out a way to make the event-driven design support events that aren't just in memory?
(same question of course for AgentTool)
I think it is similar to:
#165 (comment)
WDYT?
Beta Was this translation helpful? Give feedback.
All reactions