-
Notifications
You must be signed in to change notification settings - Fork 482
DPL: earlier forwarding #14910
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: dev
Are you sure you want to change the base?
DPL: earlier forwarding #14910
Conversation
Use a single helper function to improve readability.
If one (header, payload, ...) tuple in a MessageSet was to be copied, all the subsequent ones would have been copied. If one (header, payload, ...) tuple got redirected to more than one destination, all the subsequent ones would have been redirected there.
|
REQUEST FOR PRODUCTION RELEASES: This will add The following labels are available |
|
@shahor02 this works in my synthetic tests (stage/bin/o2-testworkflows-early-forwarding -s --severity detail --early-forward-policy=always) . In the end I refactored the code to find the earliest spot where messages are guaranteed to be seen only once and I moved the early forward there. @davidrohr @shahor02 I have noticed that the early forwarding is disabled by default. Is this expected? |
|
@jgrosseo @nicolaspoffley I expect this to improve parallelism on hyperloop as well. |
|
@ktf for me it is not expected that the EF is disabled, when I was debugging the slow turnover of Polaris jobs, I thought the forwarding is done at the beginning of run method. Was not this the supposed behaviour of the EF? |
|
@shahor02 I need to have a better look. Maybe it's just my small reproducer to be wrong. I also see there is some issues with some of the tests. I will debug better tomorrow morning. |
This is most likely faster, and it will allow us to move the early forwarding at an earlier stage where the data is not yet in a MessageSet.
Add splitPayloadIndex / splitPayloadParts to the default printout
This anticipates the forwarding to the earliest possible moment, i.e. when we are about to insert the messages in a slot. This is the earliest moment we can guarantee messages will be seen only once.
|
Ok, fixed the off by one issue with multiparts. |
|
Error while checking build/O2/fullCI_slc9 for f6dfcce at 2025-12-23 21:42: Full log here. |
For online and offline reco we enable it here: https://github.com/davidrohr/O2DPG/blob/a5af1be2a96bbe3b2eeb2cf13d41c4afd1b81e4a/DATA/common/getCommonArgs.sh#L12 |
|
@ktf this seems to be genuine crash: |
|
@shahor02 indeed. I am investigating. |
|
I suspect it's an issue with the back pressure. I will try to replicate. |
This anticipates the forwarding to the earliest possible moment, i.e. when
we are about to insert the messages in a slot. This is the earliest moment
we can guarantee messages will be seen only once.
Stack created with Sapling. Best reviewed with ReviewStack.