On 16/01/15 16:28, Josh Berkus wrote:
> On 01/16/2015 04:17 PM, Mark Kirkwood wrote:
>> On 16/01/15 16:06, Mark Kirkwood wrote:
>>
>>> A bit more poking about shows that the major factor (which this fake
>>> dataset anyway) is the default for effective_cache_size (changes from
>>> 128MB to 4GB in 9.4). Increasing this makes 9.2 start using the
>>> files_in_flight index in a plain index scan too.
>>>
>>
>> Arrg - misread the planner output....in 9.2 what changes is a plan that
>> uses an index scan on the *file_state* index (not
>> files_in_flight)...which appears much faster than the bitmap scan on
>> file_state. Apologies for the confusion.
>>
>> I'm thinking that I'm seeing the effect Tom has just mentioned.
>
> It's not using a bitmapscan in either case; it's a straight indexscan.
>
>
Right, I suspect that bloating is possibly the significant factor then -
can you REINDEX?
Cheers
Mark