"Mark Cave-Ayland" <m.cave-ayland@webbased.co.uk> writes:
> In fact, my colleague has just done a test with SELECT..INTO on our
> dev
> version and it compacted 600Mb -> 400Mb in just 40s(!). We then did
> a
> vacuum full on the same original 600Mb table which is still going
> after
> 20mins.
>>
>> Are there indexes on the original table? If so, this isn't a fair
>> comparison.
> Fair point actually, I should have made it a better comparison. The
> source table has 5 btree indexes, each on a bigint field. However, it
> has taken just under a minute to recreate the first! The vacuum full on
> the original 600Mb table has finished after 100mins, so it looks as if I
> used the SELECT..INTO method could be up and done in 10mins! I can
> continue recreating the other indexes to get a proper final time
> comparison if you are interested?
Yeah. Also, I don't suppose you made that a VACUUM VERBOSE and kept the
output? It'd be interesting to see which stages took the most time.
regards, tom lane