You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I had searched in the DSIP and found no similar DSIP.
Motivation
Right now, in master DS will use one single thread to dispatch the tasks. The problem is all tasks are stored at one waiting queue, if a task dispatch failed then it will be moved to the end of waiting queue.
There exist a case, if a worker is overload at one worker group, and A, B ,C are using the worker, the priority is A > B > C, if the A dispatched failed, then begin to dispatch B, but the worker changed from overload to normal, then B will be dispatched successfully, this will caused problem, since priority of A is higher than B.
Design Detail
Different worker group use different waiting queue.
We need to make one waiting queue should only be handled by one dispatch thread. Once a task dispatched failed it should be put back to its worker group queue.
SbloodyS
changed the title
[DSIP-47][Master] Separate the waiting dispatched task into different queue by worker group
[DSIP-55][Master] Separate the waiting dispatched task into different queue by worker group
Jul 3, 2024
Search before asking
Motivation
Right now, in master DS will use one single thread to dispatch the tasks. The problem is all tasks are stored at one waiting queue, if a task dispatch failed then it will be moved to the end of waiting queue.
There exist a case, if a worker is overload at one worker group, and A, B ,C are using the worker, the priority is A > B > C, if the A dispatched failed, then begin to dispatch B, but the worker changed from overload to normal, then B will be dispatched successfully, this will caused problem, since priority of A is higher than B.
Design Detail
Different worker group use different waiting queue.
We need to make one waiting queue should only be handled by one dispatch thread. Once a task dispatched failed it should be put back to its worker group queue.
Compatibility, Deprecation, and Migration Plan
No response
Test Plan
No response
Code of Conduct
The text was updated successfully, but these errors were encountered: