Disable stats and cost calculators by default

[arhimondr]

Stats calculator fails on assertions for complex queries, thus it is not production ready yet.

The #11511 changes the place where the stats are computed to be
displayed in the final plan. Before this patch, the stats were computed in QueryMonitor, when printing a final plan.
Before if the stats couldn't be computed for whatever reason, only the text plan generation would fail.

Currently, when the stats calculator is invoked for every query during the initial planning, queries may
be failing, even when the CBO is not used.

This change disables stats calculator by default. It can be enabled back with a session property on per query basis.

[arhimondr]

Failures: https://gist.github.com/arhimondr/75e88a1f4bed502e28fca776ccdbb464, https://gist.github.com/arhimondr/bbae933cc8915241ba403bc5444aaa53

[arhimondr]

CC: @rschlussel, @mbasmanova, @findepi, @sopel39, @kokosing

[electrum]

Looks good to me, but someone else should take a look

[electrum]

This change seems unrelated to making it disabled by default and should probably be a separate commit?

[rschlussel]

is this basically a shortcut so you don't keep trying to calculate cost and coming up empty because all stats are unknown?

[arhimondr]

is this basically a shortcut so you don't keep trying to calculate cost and coming up empty because all stats are unknown?

I'm really being paranoid here. Just want to make sure that this part of functionality is really killed for good.

[arhimondr]

I extracted it into a separate commit

[rschlussel]

is this basically a shortcut so you don't keep trying to calculate cost and coming up empty because all stats are unknown?

[kokosing]

As I understand this was found during the release verification and it was a release blocker, am I right?

[arhimondr]

@kokosing Yeah =\ I'm actively working on fixing those bugs. Hopefully to the next release we will be able to turn it on back by default.

[findepi]

Stats calculator fails on assertions for complex queries

Would you be able to share a query or example stacktrace? In particular, is it an old problem?

(I am aware previously those bugs would be hidden as implicit EXPLAIN was async, but our release ships with CBO on by default, and I don't recall any planning failures due to that)

[arhimondr]

I attached gists with the stack traces in the very first comment. Unfortunately i cannot share the queries. I will try to reproduce it with simpler queries that use the tpch data set. We've seen 2 issues so far. One is related to column length. I checked 0.208 and the issue is still there. In was introduced when we started to propagate varchar column length in that release. Another one is related to nulls fraction. In some cases it goes above 1. I checked the previous release(0.212), and it is still there. I didn't check the older releases though. What is the latest release you had?

…

On Wed, Oct 24, 2018 at 3:20 AM Piotr Findeisen ***@***.***> wrote: Stats calculator fails on assertions for complex queries Would you be able to share a query or example stacktrace? In particular, is it an old problem? (I am aware previously those bugs would be hidden as implicit EXPLAIN was async, but our release ships with CBO on by default, and I don't recall any planning failures due to that) — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#11765 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AFUBrJyIbQkfUDjw_k05ynWY8Z0UAmwIks5uoBS9gaJpZM4X16AU> .