Persisted models schema (#21109)

* dir locals for api/let-404

* Driver supports persisted model

* PersistedInfo model

far easier to develop this model with the following sql:

```sql
create table persisted_info (
   id serial primary key not null
  ,db_id int references metabase_database(id) not null
  ,card_id int references report_card(id) not null
  ,question_slug text not null
  ,query_hash text not null
  ,table_name text not null
  ,active bool not null
  ,state text not null
  ,UNIQUE (db_id, card_id)
)

```
and i'll make the migration later. Way easier to just dorp table, \i
persist.sql and keep developing without worrying about the migration
having changed so it can't rollback, SHAs, etc

* Persisting api (not making/deleting tables yet)

http POST "localhost:3000/api/card/4075/persist" Cookie:$COOKIE -pb
http DELETE "localhost:3000/api/card/4075/persist" Cookie:$COOKIE -pb

useful from commandline (this is httpie)

* Pull format-name into ddl.i

* Postgres ddl

* Hook up endpoints

* move schema-name into interface

* better jdbc connection management

* Hotswap peristed tables into qp

* clj-kondo fixes

* docstrings

* bad alias in test infra

* goodbye testing format-name function

left over. everything uses ddl.i/format-name and this rump was left

* keep columns in persisted info

columns that are in the persisted query. I thought about a tuple of
[col-name type] instead of just the col-name. I didn't do this this type
because I want to ensure that we compute the db-type in ONLY ONE WAY
ever and i wasn't ready to commit to that yet. I'm not sure this is
necessary in the future so it remains out now.

Context: we hot-swap the persisted table in for the original
query. Match up on query hash remaining the same. It continues to use
the metadata from the original query and just `select cols from table`

* Add migration for persisted_info table

also removes the db_id. Don't know why i was thinking that was
necessary. also means we don't need another unique constraint on (db_id,
card_id) since we can just mark the card_id as unique. no idea what i
was thinking.

* fix ns in a sad manner :(

far better to just have no alias to indicate it is required for side
effects.

* Dont hardcode a card-id :(:(:( my B

* copy the PersistedInfo

* ns cleanup, wrong alias, reflection warning

* Check that state of persisted_info is persisted

* api to enable persistence on a db

i'm not wild about POST /api/database/:id/persist and POST
/api/database/:id/unpersist but carrying on. left a note about it.

So now you can enable persistence on a db, enable persistence on a model
by posting to api/card/:id/persist and everything works.

What does not work yet is the unpersisting or re-persisting of models
when using the db toggle.

* Add refresh_begin and refresh_end to persisted_info

This information helps us with two bits:
- when we need to chunk refreshing models, this lets us order by
staleness so we can refresh a few models and pick up later
- if we desire, we can look at the previous elapsed time of refreshes
and try to gauge amount of work we want. This gives us a bit of
look-ahead. We can of course track our progress as we go but there's no
way to know if the next refresh might take an hour. This gives us a bit
of insight.

* Refresh tables every 8 hours ("0 0 0/8 * * ? *")

Tables are refreshed every 8 hours. There is one single job doing this
named "metabase.task.PersistenceRefresh.job" but it has 0 triggers by
default. Each database with persisted models will add a trigger to this
to refresh those models every 8 hours.

When you unpersist a model, it will immediately remove the table and
then delete the persisted_info record.

When you mark a database as persist false, it will immediately mark all
persisted_info rows as inactive and deleteable, and unschedule its
trigger. A background thread will then start removing the tables.

* Schedule refreshing on startup, watching for already scheduled

does not allow for schedule changes but that's a future endeavor

* appease our linter overlords

* Dynamic var to inbhit persistence when refreshing

also, it checked the state against "active" instead of "persisted" which
is really freaky. how has this worked in the past if thats the case?

* api docstrings on card persist

* docstring

* Don't sync the persisted schemas

* Fix bad sql when no deleteable rows

getting error with bad sql when there were no ids

* TaskHistory for refreshing

* Add created_at to persist_info table

helpful if this ever ends up in the audit section

* works on redshift

hooked up the hierarchy and redshift is close enought that it just works

* Remove persist_info record after deleting "deleteable"

* Better way to check that something exists

* POST /api/<card-id>/refresh

api to refresh a model's persisted record

* return a 204 from refreshing

* Add buttons to persist/unpersist a database and a model for PoC (#21344)

* Redshift and postgres report true for persist-models

there are separate notions of persistence is possible vs persistence is
enabled. Seems like we're just gonna check details for enabled and rely
on the driver multimethod for whether it is possible.

* feature for enabled, hydrate card with persisted

two features: :persist-models for which dbs support it, and
:persist-models-enabled for when that option is enabled.

POST to api/<card-id>/unpersist

hydrate persisted on cards so FE can display persist/unpersist for
models

* adjust migration number

* remove deferred-tru 



* conditionally hydrate persisted on models only

* Look in right spot for persist-models-enabled

* Move persist enabled into options not details

changing details recomposes the pool, which is especially bad now that
we have refresh tasks going on reusing the same connection

* outdated comment

* Clean up source queries from persisted models

their metadata might have had [:field 19 nil] field_refs and we should
substitute just [:field "the-name" {:base-type :type/Whatever-type}
since it will be a select from a native query.

Otherwise you get the following:

```
2022-03-31 15:52:11,579 INFO api.dataset :: Source query for this query is Card 4,088
2022-03-31 15:52:11,595 WARN middleware.fix-bad-references :: Bad :field clause [:field 4070 nil] for field "category.catid" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,596 WARN middleware.fix-bad-references :: Bad :field clause [:field 4068 nil] for field "category.catgroup" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,596 WARN middleware.fix-bad-references :: Bad :field clause [:field 4071 nil] for field "category.catname" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,596 WARN middleware.fix-bad-references :: Bad :field clause [:field 4069 nil] for field "category.catdesc" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,611 WARN middleware.fix-bad-references :: Bad :field clause [:field 4070 nil] for field "category.catid" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,611 WARN middleware.fix-bad-references :: Bad :field clause [:field 4068 nil] for field "category.catgroup" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,611 WARN middleware.fix-bad-references :: Bad :field clause [:field 4071 nil] for field "category.catname" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,611 WARN middleware.fix-bad-references :: Bad :field clause [:field 4069 nil] for field "category.catdesc" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,622 WARN middleware.fix-bad-references :: Bad :field clause [:field 4070 nil] for field "category.catid" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,622 WARN middleware.fix-bad-references :: Bad :field clause [:field 4068 nil] for field "category.catgroup" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,622 WARN middleware.fix-bad-references :: Bad :field clause [:field 4071 nil] for field "category.catname" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,623 WARN middleware.fix-bad-references :: Bad :field clause [:field 4069 nil] for field "category.catdesc" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
```
I think its complaining that that table is not joined in the query and
giving up.

While doing this i see we are hitting the database a lot:

```
2022-03-31 22:52:18,838 INFO api.dataset :: Source query for this query is Card 4,111
2022-03-31 22:52:18,887 INFO middleware.fetch-source-query :: Substituting cached query for card 4,111 from metabase_cache_1e483_229.model_4111_redshift_c
2022-03-31 22:52:18,918 INFO middleware.fetch-source-query :: Substituting cached query for card 4,111 from metabase_cache_1e483_229.model_4111_redshift_c
2022-03-31 22:52:18,930 INFO middleware.fetch-source-query :: Substituting cached query for card 4,111 from metabase_cache_1e483_229.model_4111_redshift_c
```

I tried to track down why we are doing this so much but couldn't get
there.

I think I need to ensure that we are using the query store annoyingly :(

* Handle native queries

didn't nest the vector in the `or` clause correctly. that was truthy
only when the mbql-query local was truthy. Can't put the vector `[false
mbql-query]` there and rely on that behavior

* handle datetimetz in persisting

* Errors saved into persisted_info

* Reorder migrations to put v43.00-047 before 048

* correct arity mismatch in tests

* comment in refresh task

* GET localhost:3000/api/persist

Returns persisting information:
- most information from the `persist_info` table. Excludes a few
columns (query_hash, question_slug, created_at)
- adds database name and card name
- adds next fire time from quartz scheduling

```shell
❯ http GET "localhost:3000/api/persist" Cookie:$COOKIE -pb
[
    {
        "active": false,
        "card_name": "hooking reviews to events",
        "columns": [
            "issue__number",
            "actor__login",
            "user__login",
            "submitted_at",
            "state"
        ],
        "database_id": 19,
        "database_name": "pg-testing",
        "error": "No method in multimethod 'field-base-type->sql-type' for dispatch value: [:postgres :type/DateTimeWithLocalTZ]",
        "id": 4,
        "next-fire-time": "2022-04-06T08:00:00.000Z",
        "refresh_begin": "2022-04-05T20:16:54.654283Z",
        "refresh_end": "2022-04-05T20:16:54.687377Z",
        "schema_name": "metabase_cache_1e483_19",
        "state": "error",
        "table_name": "model_4077_hooking_re"
    },
    {
        "active": true,
        "card_name": "redshift Categories",
        "columns": [
            "catid",
            "catgroup",
            "catname",
            "catdesc"
        ],
        "database_id": 229,
        "database_name": "redshift",
        "error": null,
        "id": 3,
        "next-fire-time": "2022-04-06T08:00:00.000Z",
        "refresh_begin": "2022-04-06T00:00:01.242505Z",
        "refresh_end": "2022-04-06T00:00:01.825512Z",
        "schema_name": "metabase_cache_1e483_229",
        "state": "persisted",
        "table_name": "model_4088_redshift_c"
    }
]

```

* include card_id in /api/persist

* drop table if exists

* Handle rescheduling refresh intervals

There is a single global value for the refresh interval. The API
requires it to be 1<=value<=23. There is no validation if someone
changes the value in the db or with an env variable. Setting this to a
nonsensical value could cause enormous load on the db so they shouldn't
do that.

On startup, unschedule all tasks and then reschedule them to make sure
that they have the latest value.

One thing to note: there is a single global value but i'm making a task
for each database. Seems like an obvious future enhancement so I don't
want to deal with migrations. Figure this gives us the current spec
behavior to have a trigger for each db with the same value and lets us
get more interesting using the `:options` on the database in the
future.

* Mark as admin not internal

lets it show up in `api/setting/` . I'm torn on how special this value
is. Is it the setting code's requirement to invoke the reschedule
refresh triggers or should that be on the setting itself.

It feels "special" and can do a lot of work from such just setting an
integer. There's a special endpoint to set it which is aware, and thus
would be a bit of an error to set this setting through the more
traditional setting endpoint

* Allow for "once a day" refresh interval

* Global setting to enable/disable

post api/persist/enable
post api/persist/disable

enable allows for other scheduling operations (enabling on a db, and
then on a model).

Disable will
- update each enabled database and disable in options
- update each persisted_info record and set it inactive and state
deleteable
- unschedule triggers to refresh
- schedule task to unpersist each model (deleting table and associated
pesisted_info row)

* offset and limits on persisted info list

```shell
http get "localhost:3000/api/persist?limit=1&offset=1" Cookie:$COOKIE -pb
{
    "data": [
        {
            "active": true,
            "card_id": 4114,
            "card_name": "Categories from redshift",
            "columns": [
                "catid",
                "catgroup",
                "catname",
                "catdesc"
            ],
            "database_id": 229,
            "database_name": "redshift",
            "error": null,
            "id": 12,
            "next-fire-time": "2022-04-08T00:00:00.000Z",
            "refresh_begin": "2022-04-07T22:12:49.209997Z",
            "refresh_end": "2022-04-07T22:12:49.720232Z",
            "schema_name": "metabase_cache_1e483_229",
            "state": "persisted",
            "table_name": "model_4114_categories"
        }
    ],
    "limit": 1,
    "offset": 1,
    "total": 2
}
```

* Include collection id, name, and authority level

* Include creator on persisted-info records

* Add settings to manage model persistence globally (#21546)

* Common machinery for running steps

* Add model cache refreshes monitoring page (#21551)

* don't do shenanigans

* Refresh persisted and error persisted_info rows

* Remarks on migration column

* Lint nits (sorted-ns and docstrings)

* Clean up unused function, docstring

* Use `onChanged` prop to call extra endpoints (#21593)

* Tests for persist-refresh

* Reorder requires

* Use quartz for individual refreshing for safety

switch to using one-off jobs to refresh individual tables. Required
adding some job context so we know which type to run.

Also, cleaned up the interface between ddl.interface and the
implementations. The common behaviors of advancing persisted-info state,
setting active, duration, etc are in a public `persist!` function which
then calls to the multimethod `persist!*` function for just the
individual action on the cached table.

Still more work to be done:
- do we want creating and deleting to be put into this type of system?
Quite possible
- we still don't know if a query is running against the cached table
that can prevent dropping the table. Perhaps using some delay to give
time for any running query to finish. I don't think we can easily solve
this in general because another instance in the cluster could be
querying against it and we don't have any quick pub/sub type of
information sharing. DB writes would be quite heavy.
- clean up the ddl.i/unpersist method in the same way we did refresh and
persist. Not quite clear what to do about errors, return values, etc.

* Update tests with more job-info in context

* Fix URL type conflicts

* Whoops get rid of our Thread/sleep test :)

* Some tests for the new job-data, clean up task history saving

* Fix database model persistence button states (#21636)

* Use plain database instance on form

* Fix DB model persistence toggle button state

* Add common `getSetting` selector

* Don't show caching button when turned off globally

* Fix text issue

* Move button from "Danger zone"

* Fix unit test

* Skip default setting update request for model persistence settings (#21669)

* Add a way to skip default setting update request

* Skip default setting update for persistence

* Add changes for front end persistence

- Order by refresh_begin descending
- Add endpoint /persist/:persisted-info-id for fetching a single entry.

* Move PersistInfo creation into interface function

* Hide model cache monitoring page when caching is turned off (#21729)

* Add persistence setting keys to `SettingName` type

* Conditionally hide "Tools" from admin navigation

* Conditionally hide caching Tools tab

* Add route guard for Tools

* Handle missing settings during init

* Add route for fetching persistence by card-id

* Wrangling persisted-info states

Make quartz jobs handle any changes to database.
Routes mark persisted-info state and potentially trigger jobs.
Job read persisted-info state.

Jobs

- Prune
-- deletes PersistedInfo `deleteable`
-- deletes cache table

- Refresh
-- ignores `deletable`
-- update PersistedInfo `refreshing`
-- drop/create/populate cache table

Routes

card/x/persist
- creates the PersistedInfo `creating`
- trigger individual refresh

card/x/unpersist
- marks the PersistedInfo `deletable`

database/x/unpersist
- marks the PersistedInfos `deletable`
- stops refresh job

database/x/persist
- starts refresh job

/persist/enable
- starts prune job

/persist/disable
- stops prune job
- stops refresh jobs
- trigger prune once

* Save the definition on persist info

This removes the columns and query_hash columns in favor of definition.

This means, that if the persisted understanding of the model is
different than the actual model during fetch source query we won't
substitute.

This makes sure we keep columns and datatypes in line.

* Remove columns from api call

* Add a cache section to model details sidebar (#21771)

* Extract `ModelCacheRefreshJob` type

* Add model cache section to sidebar

* Use `ModelCacheRefreshStatus` type name

* Add endpoint to fetch persistence info by model ID

* Use new endpoint at QB

* Use `CardId` from `metabase-types/api`

* Remove console.log

* Fix `getPersistedModelInfoByModelId` selector

* Use `t` instead of `jt`

* Provide seam for prune testing

- Fix spelling of deletable

* Include query hash on persisted_info

we thought we could get away with just checking the definition but that
is schema shaped. So if you changed a where clause we should invalidate
but the definition would be the same (same table name, columns with
types).

* Put random hash in PersistedInfo test defaults

* Fixing linters

* Use new endpoint for model cache refresh modal (#21742)

* Use new endpoint for cache status modal

* Update refresh timestamps on refresh

* Move migration to 44

* Dispatch on initialized driver

* Side effects get bangs!

* batch hydrate :persisted on cards

* bang on `models.persisted-info/make-ready!`

* Clean up a doc string

* Random fixes: docstrings, make private, etc

* Bangs on side effects

* Rename global setting to `persisted-models-enabled`

felt awkward (enabled-persisted-models) and renamed to make it a bit
more natural. If you are developing you need to set the new value to
true and then your state will stay the same

* Rename parameter for site-uuid-str for clarity

* Lint cleanups

interesting that the compojure one is needed for clj-kondo. But i guess
it makes sense since there is a raw `GET` in `defendpoint`.

* Docstring help

* Unify type :type/DateTimeWithTZ and :type/DateTimeWithLocalTZ

both are "TIMESTAMP WITH TIME ZONE". I had got an error and saw that the
type was timestamptz so i used that. They are synonyms although it might
require an extension.

* Make our old ns linter happy

Co-authored-by: Alexander Polyankin <alexander.polyankin@metabase.com>
Co-authored-by: Anton Kulyk <kuliks.anton@gmail.com>
Co-authored-by: Case Nelson <case@metabase.com>