Comment by acqq

> this would only mask bugs, not fix them

If I get a corruption once in 100 outages instead of on every one, I'm satisfied. That it "masks" anything is not an argument at all.

The writes happen in bursts. The behavior of bursts won't change if one more write is done after the burst is over (an example) only a second later instead of waiting 30.

The "delayed allocation" is a red herring: in the optimal case, the software can instruct the filesystem to preallocate the whole file size without having to actually fill the content. If it's not common by some specific applications on Linux, that's the place to fix it.