Yi Cheng
Yi Cheng
> > The rest lgtm. I notice the serve.start is not in the code. Could you confirm that it's working? (I see your comment that it's not working). > >...
lint failure is not related.
I think it hasn't fixed the issue ``` 255 ray 35 15 16.3g 2.3g 67012 S 67.0 7.3 17:32.35 ray::ServeRepli 70 ray 35 15 11.3g 490600 65576 S 11.0 1.5...
Btw, do you know why it keeps increasing? It should stay constant I think?
> Python hasn't gc'ed in a while I think ray core will force GC once a while. This pod got runs for half an hour, and it's consuming 2.3g memory....
Should we remove the back-pressure in client side given that it's not working and adding too much complexity? @rkooo567 @scv119 ?
some cpp test is not updated. i'll work on that.
@scv119 I've put it to a unique ptr and reset it once used. I think it should be good enough since whatever goes wrong, it'll crash.
There is still one bug in ray syncer which cause the reconstruction stress test run slower. I'll follow that in the next PR.
@jjyao, thanks for the reviewing! I'm going to pick up this work next week.