Attention is currently required from: laforge, pespin.
1 comment:
Patchset:
I tend to agree with pespin here. […]
Your responses on the blocking behavior still do not have empirical substance.
We do also block the main thread for
You cannot request to put everything in a different thread just because there is some wait time involved.
It is and remains a tradeoff, based on the actual amount of time it takes,
and also based on how often it occurs.
You are basing your opinions on the impression that asking nft is always very very very slow, which is not a fact. My impression is otherwise. I would appreciate some trust in the experience that I've gathered by actually working with nft myself.
I have explained this aspect at least three times now and would like to see this acknowledged in any way, thanks.
I still think the entire idea to add a separate thread was premature optimisation, overly complicating and not worth the effort. This is being proliferated by current CR, while the jury is still out on whether there even is a problem in the first place.
Be aware that a customer has tested and AFAIK actually is still running the fully non-threaded patch that ALWAYS blocks the main thread for these KPI, and this issue was not being raised. All of this is just a hunch by a person that hasn't tried it out in person and happens to be my boss.
Let's rather get this multi-thread patch out to the customer, and improve on it later in case that it still needs to. Instead you guys are blocking progress by being inappropriately pedantic here and missing the point entirely.
I explained many times now that it makes sense to add an it_q to this design later, to get rid of the last tiny bit of blocking. This patch is a basis for that -- it is enabling future progress, not excluding it.
To view, visit change 36540. To unsubscribe, or for help writing mail filters, visit settings.