Postgres 12 -> 13 any need to re-index?

On 2/9/21 12:00 PM, Wells Oliver wrote:
> Hey guys, I just upgraded a DB from 12 to 13. In testing this seemed to go
> well, but it seems like most of my queries are not using the indexes after
> the upgrade, so everything is quite so.
>
> Is there a general need to REINDEX after upgrading to 13, or might there
> be another issue? The configuration parameters are the same as they were
> persisted via pg_upgradecluster.

Did you also change or upgrade the distro, or modify the collation? (That's
an automatic need to rebuild indices.)


--
Angular momentum makes the world go 'round.

--

Wells Oliver schrieb am 09.02.2021 um 19:00:
> Hey guys, I just upgraded a DB from 12 to 13. In testing this seemed
> to go well, but it seems like most of my queries are not using the
> indexes after the upgrade, so everything is quite so.
>
> Is there a general need to REINDEX after upgrading to 13, or might
> there be another issue? The configuration parameters are the same as
> they were persisted via pg_upgradecluster.


If you want to profit from the new de-duplication in indexes, you will need to reindex.

On 2/9/21 12:13 PM, Wells Oliver wrote:

Do you mean Ubuntu distro? I did not. My database collation is also still en_US.UTF-8, but everything has ground to a halt after the upgrade.

Did you run an ANALYZE after the conversion?

On Tue, Feb 9, 2021 at 10:07 AM Ron <[hidden email]> wrote:

On 2/9/21 12:00 PM, Wells Oliver wrote:
> Hey guys, I just upgraded a DB from 12 to 13. In testing this seemed to go
> well, but it seems like most of my queries are not using the indexes after
> the upgrade, so everything is quite so.
>
> Is there a general need to REINDEX after upgrading to 13, or might there
> be another issue? The configuration parameters are the same as they were
> persisted via pg_upgradecluster.

Did you also change or upgrade the distro, or modify the collation? (That's
an automatic need to rebuild indices.)

--

On 2/9/21 12:18 PM, Wells Oliver wrote:

No, I didn't. I had done that in testing

Tsk tsk.

and wanted to get the thing back up and running. Would i just run 'analyze dbname' or something more verbose?

Essentially.  I'd do something like:

vacuumdb --analyze-only --jobs=$(nproc) -d ${YOUR_DB}

On Tue, Feb 9, 2021 at 10:17 AM Ron <[hidden email]> wrote:

On 2/9/21 12:13 PM, Wells Oliver wrote:

Do you mean Ubuntu distro? I did not. My database collation is also still en_US.UTF-8, but everything has ground to a halt after the upgrade.

Did you run an ANALYZE after the conversion?

On Tue, Feb 9, 2021 at 10:07 AM Ron <[hidden email]> wrote:

On 2/9/21 12:00 PM, Wells Oliver wrote:
> Hey guys, I just upgraded a DB from 12 to 13. In testing this seemed to go
> well, but it seems like most of my queries are not using the indexes after
> the upgrade, so everything is quite so.
>
> Is there a general need to REINDEX after upgrading to 13, or might there
> be another issue? The configuration parameters are the same as they were
> persisted via pg_upgradecluster.

Did you also change or upgrade the distro, or modify the collation? (That's
an automatic need to rebuild indices.)

--

--

On Tuesday, February 9, 2021, Wells Oliver <[hidden email]> wrote:

OK, that seems to have made a huge difference. and ran very quickly, ~15 mins on a ~1TB database. 

Is there a way to better understand what it did?

There's no way it rebuilt indexes in that amount of time. Also, as I said, I never did that going from 9->10->11->12, so it's interesting to me that it was so critical this time around.

This command is well documented:

vacuumdb --analyze-only --jobs=$(nproc) -d ${YOUR_DB}

OK, that seems to have made a huge difference. and ran very quickly, ~15 mins on a ~1TB database. Is there a way to better understand what it did? There's no way it rebuilt indexes in that amount of time. Also, as I said, I never did that going from 9->10->11->12, so it's interesting to me that it was so critical this time around.

On Tue, Feb 9, 2021 at 10:27 AM Wells Oliver <[hidden email]> wrote:

Yeah. Running that now. Honestly, I went from 9 to 10 to 11 to 12 without doing that in prod, and it was marvelous. 13 hit a wall, and I'm not sure why.

It's running now, though...

On Tue, Feb 9, 2021 at 10:25 AM Ron <[hidden email]> wrote:

On 2/9/21 12:18 PM, Wells Oliver wrote:

No, I didn't. I had done that in testing

Tsk tsk.

and wanted to get the thing back up and running. Would i just run 'analyze dbname' or something more verbose?

Essentially.  I'd do something like:

vacuumdb --analyze-only --jobs=$(nproc) -d ${YOUR_DB}

On Tue, Feb 9, 2021 at 10:17 AM Ron <[hidden email]> wrote:

On 2/9/21 12:13 PM, Wells Oliver wrote:

Do you mean Ubuntu distro? I did not. My database collation is also still en_US.UTF-8, but everything has ground to a halt after the upgrade.

Did you run an ANALYZE after the conversion?

On Tue, Feb 9, 2021 at 10:07 AM Ron <[hidden email]> wrote:

On 2/9/21 12:00 PM, Wells Oliver wrote:
> Hey guys, I just upgraded a DB from 12 to 13. In testing this seemed to go
> well, but it seems like most of my queries are not using the indexes after
> the upgrade, so everything is quite so.
>
> Is there a general need to REINDEX after upgrading to 13, or might there
> be another issue? The configuration parameters are the same as they were
> persisted via pg_upgradecluster.

Did you also change or upgrade the distro, or modify the collation? (That's
an automatic need to rebuild indices.)

--
Angular momentum makes the world go 'round.

--

Wells Oliver
[hidden email]

--

Wells Oliver
[hidden email]