Exclude sample content and Metabase Analytics from usage stats

[calherries]

Now that we're adding sample content to new instances in #40753, this PR excludes the content from usage stats. The relevant section of the product doc is here:

When sending usage stats, we should not include this sample content (or the audit v2 content) in any question/dashboards aggregates

It also excludes Metabase Analytics content (behaviour approved here by Arakaki)

Both changes the same approach:

1. We now exclude any cards and dashboards that were created by the internal user from the stats. We also exclude any dashcards that belong to a dashboard that was created by the internal user.
2. We exclude any collections that have is_sample=true from the stats. This required creating a new is_sample column on the collection table.

[replay-io]

Status	Complete ↗︎
Commit	068f605
Results	⚠️ 7 Flaky ✅ 2402 Passed

[crisptrutski]

At first I was appalled at doing table scans to filter such potentially-large tables, but then I realized we're already pulling them all into memory in any case - ouch! 😆

[crisptrutski]

It's probably more efficient to flip this around and skip the join by doing something like [:not [:in :dc.dashboard_id the-really-small-number-of-sample-dashboard-ids]]

[calherries]

Does it matter? This query gets every dashcard in the app DB anyway

[crisptrutski]

My intuition for how databases work, and the related costs, is very rotted, so I may be way off.

I suspect that a table scan with a join is a lot more expensive than a plain table scan, I'd expect it to build a temporary table or do sub-selects.

I don't think the query analyzer has access to statistics that would suggest doing this inversion internally, and it's not any more complicated. Not gonna block on it, this namespace is ripe for broader optimizations.

[calherries]

@crisptrutski Sorry, I updated the PR because I decided to also include the Metabase Analytics changes. Can you review again?

[crisptrutski]

According to all the blood in the last CI run, Collection doesn't have such a column 😢

I'm guessing you can't abuse :personal_owner_id without hiding it from other users, so I'm not sure what you can do here.

One dubious option would be to remove collections which only contain sample cards 🥴

[calherries]

Yeah, I've updated this. I had some local state tricking me into thinking there was a creator_id column.

See the latest version https://github.com/metabase/metabase/pull/41519/files#diff-96de37bff643ff1187fa863f5a14541a6c03a6c2c20deaa7e043a0fcf8f54a08R305

[crisptrutski]

Cool 👍

[crisptrutski]

Re-ticking ✅ ✅

A slightly philosophical question - at what point does sample content become user content?

e.g.

• a customer renames and adds new cards to a collection
• a customer edits a card substantially and adds it to another dashboard

etc.

[calherries]

Made a clone because I accidentally merged this into the parent branch #41522

[calherries]

A slightly philosophical question - at what point does sample content become user content?

Good question, it won't ever change. Once something is created as sample content it stays sample content. A sample content card that is renamed will stay sample content. A new card/dashboard in a sample content collection will not be counted as sample content though.

I think this is ok since this is just for our own internal statistics and probably only need to be like 99% correct to be useful.

[crisptrutski]

Yeah it's probably better to keep things simple than to navigate such a grey area.

If we were going to dip our toes into refining things, I was thinking we could compare created and updated times.