[illumos-gate] 17734 ZFS fsync can trigger ZIL transaction reordering and data corruption -- Andy Fiddaman <illumos⊙fn>

nice!

People using my openjdk patches may be interested in the xattrs.patch I've just added for JDK21 and later, which restores extended attribute support

Thanks, ptribble!

Definitely interested ptribble, thanks, it'll be incorporated into next week's OmniOS JDK updates.

ptribble: neat!

Thank you very much, @ptribble.

i don't think there is, but in case i missed it.. is there a function that will atomically update a value as long as it's larger than the current value?

i.e. I want to track a max value amongst threads

(if not, atomic_cas in a loop until the target value is larger than the one we want to replace, or we succeed)

is how I was going to do it..

There's no primitive that I know of.

ok.. figured it couldn't hurt to ask

jbk: I would keep a per-thread max value and do the max() aggregation on read?

this is in the kernel (sd)... was looking to collect stats on if/when it's throttling requests because the HBA (or fabric for FC) is reporting full...

mostly to see if the queue depth could be too low (since annoyingly SCSI doesn't actually have a way to tell you what the depth is -- at least that I've been able to find)

for things like SSDs

you could do it per-CPU then instead of per-thread

yeah.. i suppose I could just do a mutex as well -- my expectation is that even if we hit this, it shouldn't be in a crticial path...(if the device's queue is full, the overhead of taking a lock -- even a contested one -- is going to be noise compared to the delay in waiting for the queue to drain)

at this point, it's more about gathering data since it's not something easily visible right now...

and it would not surprise me at all if sd.c's values might not be optimal for modern SSDs

just all the other sd stats use atomic updates.. though that probably doesn't matter too much here...

[illumos-gate] 17340 Want IPv6 Fast Path -- Ryan Goodfellow <ryan.goodfellow⊙oc>

took a village to get that one across the line

very valuable though!

Thank you for working on it!