>
>> Their workaround had been to run a daily autovacuum at the lowest load
>> time of day, to cause the least disruption.
>
> What is a "daily autovacuum"? It sounds like some tables just need
> vacuuming more often. If they find that the system is not responsive
> during that, it tells us that they need more disk bandwidth or that
> they need to integrate vacuuming some tables with their program.
>
Sorry, I meant a daily manual vacuum.
On paper, their hardware is plenty fast for their workload. Out of
hundreds of sites, all running the same software putting load on the
database, this is only the second time where we have seen this odd
behaviour of very slow vacuums.
I guess I was hoping that someone would be able to chime in and say -
yes, in so and so version, we fixed an obscure bug that sometimes
caused huge slowdowns, perhaps when combined with certain linux
kernels. It was a nice dream anyway :)
iozone looks useful. I'll see if I can get on their system and do
some proper benchmarks.
Thanks,
Dan