Rob Norris

I did some semi-scientific test runs this morning. Production (non-debug) build at the given commit. Each run is on a freshly-created pool of 14 disks. 100 threads, either 73 writing...

I am likely to withdraw this. It destroys read throughput on mirrors, because it ends up waiting for reads spread across multiple mirror vdevs to return to the single extra...

Yep, pretty sure I don't need this now. Lets keep making `zio_t` smaller though, that's gonna help everything :)

FYI, I managed to try this on a RPi4b this morning: ``` ASSERT at cmd/zdb/zdb.c:693:usage() "ohno" == 0 (0x5555d71550 == 0) PID: 62086 COMM: zdb TID: 62086 NAME: zdb Registers:...

Alright, sorted: was holding it wrong (libunwind is kinda weird, but also quite powerful). Turns out register names are just a static lookup table by enum. Register _existence_ is more...

> I have subtle feeling that you are overdoing something with the `zl_lwb_crash_list`. ZIL already has a mechanism to chain errors via `lwb_error`, and error on one LWB automatically mean...

I've spent most of the day studying how page writeback errors wire up to this (wrt https://github.com/openzfs/zfs/pull/17398#discussion_r2119848333 above) and it seems it all needs work. As best I can tell,...

Alright, last push has a new commit, sorting out the error reporting and page dirtying dance between `zpl_fsync()`, `zfs_putpage()` and the itx callbacks. I believe its (close to) right, but...

I've got some more changes coming on this too, though I am fast running out of week for it: - incorporate (rebase atop) #17420, and make sure the `zil_commit()` call...

Last push rebased on #17420 and #17445, and includes matching changes to both, most importantly that the FreeBSD putpage callbacks now receive and handle errors too. I'm still chewing on...