> Do you guys have any more ideas to properly 'feel this disk at its
> teeth' ?
While an 'end-to-end' test using PG is fine, I think it would be easier
to determine if the drive is behaving correctly by using a simple test
program that emulates the storage semantics the WAL expects. Have it
write a constant stream of records, fsync'ing after each write. Record
the highest record number flushed so far in some place that won't be
lost with the drive under test (e.g. send it over the network to another
machine).
Kill the power, bring the system back up again and examine what's at the
tail end of that file. I think this will give you the worst case test
with the easiest result discrimination.
If you want to you could add concurrent random writes to another file
for extra realism.
Someone here may already have a suitable test program. I know I've
written several over the years in order to test I/O performance, prove
the existence of kernel bugs, and so on.
I doubt it matters much how long the power is turned of. A second should
be plenty time to flush pending writes if the drive is going to do so.