From: | Laurenz Albe <laurenz(dot)albe(at)cybertec(dot)at> |
---|---|
To: | Stephen Froehlich <s(dot)froehlich(at)cablelabs(dot)com>, "pgsql-novice(at)postgresql(dot)org" <pgsql-novice(at)postgresql(dot)org> |
Subject: | Re: Bitmap Heap Scan taking ~60x as long for table when queried as partition |
Date: | 2018-02-15 20:58:37 |
Message-ID: | 1518728317.2562.4.camel@cybertec.at |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-novice |
Stephen Froehlich wrote:
> Some of the partition scans are quick (those with nothing to return),
> but my point is that the scan PER TABLE is significantly slower than if
> I call the hypertable than if I call the table directly.
> THIS SHOULDN'T BE THE CASE. Most tables are scanned quickly and return nothing.
>
> There IS a combined index on client_ip_md5, start_time ... its my primary key for all of these tables.
>
> Also, the scans are typically parallelized (go back in the thread to the original excerpt),
> the only reason why not this time is that the server was busy with a backup.
> Its still much slower when calling the hypertable than the table directly.
> The parallelization is usually my first clue that an index scan is not being
> used but instead a heap scan.
In your complete plan, scanning "raptor_global_bitrate_20171101_cmts1" took
only 382.247 microseconds as opposed to 24760.668 in your first e-mail.
Also the strange "loops=6" is not present.
So it is hard to say what was going on there in the first place...
Often caching causes big differences in execution time.
Yours,
Laurenz Albe
From | Date | Subject | |
---|---|---|---|
Next Message | Stephen Froehlich | 2018-02-15 21:02:41 | RE: Bitmap Heap Scan taking ~60x as long for table when queried as partition |
Previous Message | Stephen Froehlich | 2018-02-15 15:55:02 | RE: Bitmap Heap Scan taking ~60x as long for table when queried as partition |