Re: Accounting for between table correlation

On 1/15/21 11:54 AM, Adrian Klaver wrote:
> On 1/15/21 10:49 AM, Alexander Stoddard wrote:
> 
> Please reply to list also.
> Ccing list.
> 
>>
>>     So to be clear, the process imports the data, then you run a query 
>> and
>>     it completes in x time, you then ANALYZE the same data and it runs 
>> in y
>>     time. Is that correct?
>>
>> The process imports data, ANALYZE is run and then queries run in x time.
>> A subsequent ANALYZE, may or may not, change the time to y.
>> x may be greater or less than y for any given pair of runs, and the 
>> difference is vast. Two very different performance domains, due to the 
>> plan, I believe. If I am correctly reading the EXPLAIN plans the row 
>> estimates are always way off (and low), regardless of if a high or low 
>> performing plan is actually chosen.
> 
> Well I'm going to say this is not going to get a useful answer without 
> some concrete numbers. Too many variables involved to just start 
> guessing at solutions.

Not sure if it would work for the vendor or not but:

https://explain.depesz.com/

offers an option to obfuscate EXPLAIN/ANALYZE output.

> 
>>
>> Thank you,
>> Alex
> 
> 


-- 
Adrian Klaver
adrian.klaver@aklaver.com