I've heard 'em say "given enough eyeballs all bugs are shallow" - but they don't tell you it still takes perseverance. The bazaar is messy way to write software, and on projects of some age and importance it seems to me it can be kinda SLOW... the only reason I can imagine we put up with it (and all the best software we love and use ends up being open source) is because somehow the results justify the cost?!
> Do you know if it will be difficult to develop a working fix for this ?
I don't think it'll be "difficult" but there's a lot of ramp up to contributing initially; Matt is amazing and has been working on Swift a long time - it still took him a month to get a fix for a similar bug landed:
post_as_copy was a questionable trade off since the beginning; it took Alistair two years to get "fast-post" working the way we wanted so we could start to deprecate post_as_copy:
I understand for your workflows and configuration it's a regression, but from a maintainers perspective I can't imagine a realistic path back to post_as_copy. You'd have a much better shot advocating/contributing to getting this bug fixed.
But honestly, if you're operating a growing cluster - you'll be happier running with write-affinity turned off. Thanks for confirming the issue with write-affinity + POST; I'll mark the other bug as a duplicate of this one.
I've heard 'em say "given enough eyeballs all bugs are shallow" - but they don't tell you it still takes perseverance. The bazaar is messy way to write software, and on projects of some age and importance it seems to me it can be kinda SLOW... the only reason I can imagine we put up with it (and all the best software we love and use ends up being open source) is because somehow the results justify the cost?!
> Do you know if it will be difficult to develop a working fix for this ?
I don't think it'll be "difficult" but there's a lot of ramp up to contributing initially; Matt is amazing and has been working on Swift a long time - it still took him a month to get a fix for a similar bug landed:
https:/ /review. opendev. org/#/c/ 114120/
> Or to reimplement it by any chance ?
post_as_copy was a questionable trade off since the beginning; it took Alistair two years to get "fast-post" working the way we wanted so we could start to deprecate post_as_copy:
https:/ /review. opendev. org/#/c/ 135380/
I understand for your workflows and configuration it's a regression, but from a maintainers perspective I can't imagine a realistic path back to post_as_copy. You'd have a much better shot advocating/ contributing to getting this bug fixed.
But honestly, if you're operating a growing cluster - you'll be happier running with write-affinity turned off. Thanks for confirming the issue with write-affinity + POST; I'll mark the other bug as a duplicate of this one.