Sitsofe Wheeler

Hmm... Technically on https://ci.appveyor.com/project/vincentkfu/fio/builds/33008676 it was the 64 bit build that failed (and the 32 bit one succeeded) but we're talking about WIndows which is a LLP64 platform. At any...

(Although looking back in time I see 298d751e5f7dcde630d90a38f9ad5ab6670483a1 so maybe truncating doesn't work...)

> The right-hand side really should evaluate to 43978 I keep coming back to this one. I think the underlying case is python provides 64 bit floating point arithmetic by...

@Vladyyy this comes down to what happens when you do rw=(rw|randrw) job. ``` verify=crc32c [...] rw=randrw [...] time_based=1 [...] runtime=${RUNTIME} ``` The "problem" arises when you create a verify job...

@Vladyyy now I'm wondering whether my previous analysis is correct now I've re-read what you wrote: > When looking at the debug log it seems there there is a race...

@Vladyyy OK thanks for the update. For now (as you stated in your first comment) people will have to `serialize_overlap` to avoid triggering this...

@bardavid sorry for the slow reply. > i'm not sure why fio optimizes the randommap case where it uses a linked list instead of having a unified data structure to...

@bardavid :-) I vote for you to start - you've got a good track record of working on necessary things...

@bardavid arguably would 2. be taken care of by `serialize_overlap=1`?

@brettmilford If you can afford the time give it a go! It might take some back and for but most things can be worked out...