Replies: 5 comments 11 replies
-
|
@mengw15 Can you rotate the image to make it easy to read? :-) |
Beta Was this translation helpful? Give feedback.
-
|
Thanks. @SarahAsad23 You may also want to check this discussion and the diagram as they are related to your work of Python virtual environments. |
Beta Was this translation helpful? Give feedback.
-
|
For the design, we would want UDFs to be placed at an isolated environment so that it does not touch driver environment. The current design puts all UDFs with driver, and I think it is not aligned with the goal? |
Beta Was this translation helpful? Give feedback.
-
|
I think the container on the right is used to run all UDFs. @mengw15 Can you clarify the purpose of the "Amber Worker (Driver)"? |
Beta Was this translation helpful? Give feedback.
-
|
A minor comment: we recently changed Akka to Apache Pekko. @mengw15 Please revise the writing accordingly. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Summary
Currently, a computing unit is a single Pod with one Container running the Amber Coordinator (master). We propose adding a separate Container in the same Pod that runs an Amber worker. Operator scheduling will be updated so all UDF operators execute on the Worker container.
Motivation
This design leverages Pekko’s distribution features
Design (high level)
Separate Worker Container in the same Pod
Extend the Pod spec to start one new container for an Amber Worker.
Scheduling change
Update the scheduler so that all UDF operators are placed on the Worker container.
Non-UDF operators keep current placement (initially unchanged).
Beta Was this translation helpful? Give feedback.
All reactions