issue_comments

I made the changes locally and tested them with Python 3.6 like so: cd /tmp mkdir v cd v pipenv shell --python=python3.6 cd ~/Dropbox/Development/datasette pip install -e '.[test]' pytest tests/test_asyncdi.py

E File "/home/runner/work/datasette/datasette/datasette/utils/vendored_graphlib.py", line 56 E if (result := self._node2info.get(node)) is None: E ^ E SyntaxError: invalid syntax Oh no - the vendored code I use has := so doesn't work on Python 3.6! Will have to backport it more thoroughly.

Conclusion: using a giant convoluted CTE and UNION ALL query to attempt to calculate facets at the same time as retrieving rows is a net LOSS for performance! Very surprised to see that.

One very interesting difference between the two: on the single giant query page:

json { "request_duration_ms": 376.4317020000476, "sum_trace_duration_ms": 370.0828700000329, "num_traces": 5 } And on the page that uses separate queries: json { "request_duration_ms": 819.012272000009, "sum_trace_duration_ms": 201.52852100000018, "num_traces": 19 } The separate pages page takes 819ms total to render the page, but spends 201ms across 19 SQL queries.

The single big query takes 376ms total to render the page, spending 370ms in 5 queries

All of that additional non-SQL overhead must be stuff relating to Python and template rendering code running on the page. I'm really surprised at how much overhead that is! This is worth researching separately.

I tried adding cases > 10000 but the SQL query now takes too long - so moving this to my laptop.

cd /tmp wget https://covid-19.datasettes.com/covid.db datasette covid.db \ --setting facet_time_limit_ms 10000 \ --setting sql_time_limit_ms 10000 \ --setting trace_debug 1 http://127.0.0.1:8006/covid/ny_times_us_counties?_trace=1&_facet_size=3&_size=2&cases__gt=10000 shows in the traces:

json [ { "type": "sql", "start": 12.693033525, "end": 12.694056904, "duration_ms": 1.0233789999993803, "traceback": [ " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/base.py\", line 262, in get\n return await self.view_get(\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/base.py\", line 477, in view_get\n response_or_template_contexts = await self.data(\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/table.py\", line 705, in data\n results = await db.execute(sql, params, truncate=True, **extra_args)\n" ], "database": "covid", "sql": "select rowid, date, county, state, fips, cases, deaths from ny_times_us_counties where \"cases\" > :p0 order by rowid limit 3", "params": { "p0": 10000 } }, { "type": "sql", "start": 12.694285093, "end": 12.814936275, "duration_ms": 120.65118200000136, "traceback": [ " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/base.py\", line 262, in get\n return await self.view_get(\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/base.py\", line 477, in view_get\n response_or_template_contexts = await self.data(\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/table.py\", line 723, in data\n count_rows = list(await db.execute(count_sql, from_sql_params))\n" ], "database": "covid", "sql": "select count(*) from ny_times_us_counties where \"cases\" > :p0", "params": { "p0": 10000 } }, { "type": "sql", "start": 12.818812089, "end": 12.851172544, "duration_ms": 32.360455000000954, "traceback": [ " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/table.py\", line 856, in data\n suggested_facets.extend(await facet.suggest())\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/facets.py\", line 164, in suggest\n distinct_values = await self.ds.execute(\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/app.py\", line 634, in execute\n return await self.databases[db_name].execute(\n" ], "database": "covid", "sql": "select county, count(*) as n from (\n select rowid, date, county, state, fips, cases, deaths from ny_times_us_counties where \"cases\" > :p0 \n ) where county is not null\n group by county\n limit 4", "params": { "p0": 10000 } }, { "type": "sql", "start": 12.851418868, "end": 12.871268359, "duration_ms": 19.84949100000044, "traceback": [ " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/table.py\", line 856, in data\n suggested_facets.extend(await facet.suggest())\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/facets.py\", line 164, in suggest\n distinct_values = await self.ds.execute(\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/app.py\", line 634, in execute\n return await self.databases[db_name].execute(\n" ], "database": "covid", "sql": "select state, count(*) as n from (\n select rowid, date, county, state, fips, cases, deaths from ny_times_us_counties where \"cases\" > :p0 \n ) where state is not null\n group by state\n limit 4", "params": { "p0": 10000 } }, { "type": "sql", "start": 12.871497655, "end": 12.897715027, "duration_ms": 26.217371999999628, "traceback": [ " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/views/table.py\", line 856, in data\n suggested_facets.extend(await facet.suggest())\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/facets.py\", line 164, in suggest\n distinct_values = await self.ds.execute(\n", " File \"/usr/local/Cellar/datasette/0.58.1/libexec/lib/python3.9/site-packages/datasette/app.py\", line 634, in execute\n return await self.databases[db_name].execute(\n" ], "database": "covid", "sql": "select fips, count(*) as n from (\n select rowid, date, county, state, fips, cases, deaths from ny_times_us_counties where \"cases\" > :p0 \n ) where fips is not null\n group by fips\n limit 4", "params": { "p0": 10000 } } ] So that's: fetch rows: 1.0233789999993803 ms count: 120.65118200000136 ms facet county: 32.360455000000954 ms facet state: 19.84949100000044 ms facet fips: 26.217371999999628 ms = 200.1 ms total

Compared to: http://127.0.0.1:8006/covid?sql=with+cte+as+(%0D%0A++select+rowid%2C+date%2C+county%2C+state%2C+fips%2C+cases%2C+deaths%0D%0A++from+ny_times_us_counties%0D%0A)%2C%0D%0Atruncated+as+(%0D%0A++select+null+as+_facet%2C+null+as+facet_name%2C+null+as+facet_count%2C+rowid%2C+date%2C+county%2C+state%2C+fips%2C+cases%2C+deaths%0D%0A++from+cte+order+by+date+desc+limit+4%0D%0A)%2C%0D%0Astate_facet+as+(%0D%0A++select+%27state%27+as+_facet%2C+state+as+facet_name%2C+count(*)+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A)%2C%0D%0Afips_facet+as+(%0D%0A++select+%27fips%27+as+_facet%2C+fips+as+facet_name%2C+count(*)+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A)%2C%0D%0Acounty_facet+as+(%0D%0A++select+%27county%27+as+_facet%2C+county+as+facet_name%2C+count(*)+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A)%0D%0Aselect+*+from+truncated%0D%0Aunion+all+select+*+from+state_facet%0D%0Aunion+all+select+*+from+fips_facet%0D%0Aunion+all+select+*+from+county_facet&_trace=1

Which is 353ms total.

The separate queries ran faster! Really surprising result there.

One last experiment: I'm going to try running an expensive query in the CTE portion.

With trace.

https://covid-19.datasettes.com/covid/ny_times_us_counties?_trace=1&_facet_size=3&_size=2&_trace=1 shows the following:

fetch rows: 0.41762600005768036 ms facet state: 284.30423800000426 ms facet county: 273.2565999999679 ms facet fips: 197.80996999998024 ms = 755.78843400001ms total

It didn't run a count because that's the homepage and the count is cached. So I dropped the count from the query and ran it:

https://covid-19.datasettes.com/covid?sql=with+cte+as+(%0D%0A++select+rowid%2C+date%2C+county%2C+state%2C+fips%2C+cases%2C+deaths%0D%0A++from+ny_times_us_counties%0D%0A)%2C%0D%0Atruncated+as+(%0D%0A++select+null+as+_facet%2C+null+as+facet_name%2C+null+as+facet_count%2C+rowid%2C+date%2C+county%2C+state%2C+fips%2C+cases%2C+deaths%0D%0A++from+cte+order+by+date+desc+limit+4%0D%0A)%2C%0D%0Astate_facet+as+(%0D%0A++select+%27state%27+as+_facet%2C+state+as+facet_name%2C+count()+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A)%2C%0D%0Afips_facet+as+(%0D%0A++select+%27fips%27+as+_facet%2C+fips+as+facet_name%2C+count()+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A)%2C%0D%0Acounty_facet+as+(%0D%0A++select+%27county%27+as+_facet%2C+county+as+facet_name%2C+count()+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A)%0D%0Aselect++from+truncated%0D%0Aunion+all+select++from+state_facet%0D%0Aunion+all+select++from+fips_facet%0D%0Aunion+all+select+*+from+county_facet&_trace=1

Shows 649.4359889999259 ms for the query - compared to 755.78843400001ms for the separate. So it saved about 100ms.

Still not a huge difference though!

One disadvantage to this approach: if you have a SQL time limit of 1s and it takes 0.9s to return the rows but then 0.5s to calculate each of the requested facets the entire query will exceed the time limit.

Could work around this by catching that error and then re-running the query just for the rows, but that would result in the user having to wait longer for the results.

Could try to remember if that has happened using an in-memory Python data structure and skip the faceting optimization if it's caused problems in the past? That seems a bit gross.

Maybe this becomes an opt-in optimization you can request in your metadata.json setting for that table, which massively increases the time limit? That's a bit weird too - now there are two separate implementations of the faceting logic, which had better have a REALLY big pay-off to be worth maintaining.

What if we kept the query that returns the rows to be displayed on the page separate from the facets, but then executed all of the facets together using this method such that the cte only (presumably) has to be calculated once? That would still lead to multiple facets potentially exceeding the SQL time limit when single facets would not have.

Maybe a better optimization would be to move facets to happening via fetch() calls from the client, so the user gets to see their rows instantly and the facets then appear as and when they are available (though it would cause page jank).

https://covid-19.datasettes.com/covid is 805.2MB

https://covid-19.datasettes.com/covid/ny_times_us_counties?_trace=1&_facet_size=3&_size=2

Equivalent SQL:

https://covid-19.datasettes.com/covid?sql=with+cte+as+%28%0D%0A++select+rowid%2C+date%2C+county%2C+state%2C+fips%2C+cases%2C+deaths%0D%0A++from+ny_times_us_counties%0D%0A%29%2C%0D%0Atruncated+as+%28%0D%0A++select+null+as+_facet%2C+null+as+facet_name%2C+null+as+facet_count%2C+rowid%2C+date%2C+county%2C+state%2C+fips%2C+cases%2C+deaths%0D%0A++from+cte+order+by+date+desc+limit+4%0D%0A%29%2C%0D%0Astate_facet+as+%28%0D%0A++select+%27state%27+as+_facet%2C+state+as+facet_name%2C+count%28%29+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A%29%2C%0D%0Afips_facet+as+%28%0D%0A++select+%27fips%27+as+_facet%2C+fips+as+facet_name%2C+count%28%29+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A%29%2C%0D%0Acounty_facet+as+%28%0D%0A++select+%27county%27+as+_facet%2C+county+as+facet_name%2C+count%28%29+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A%29%2C%0D%0Atotal_count+as+%28%0D%0A++select+%27COUNT%27+as+_facet%2C+%27%27+as+facet_name%2C+count%28%29+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte%0D%0A%29%0D%0Aselect++from+truncated%0D%0Aunion+all+select++from+state_facet%0D%0Aunion+all+select++from+fips_facet%0D%0Aunion+all+select++from+county_facet%0D%0Aunion+all+select+*+from+total_count

sql with cte as ( select rowid, date, county, state, fips, cases, deaths from ny_times_us_counties ), truncated as ( select null as _facet, null as facet_name, null as facet_count, rowid, date, county, state, fips, cases, deaths from cte order by date desc limit 4 ), state_facet as ( select 'state' as _facet, state as facet_name, count(*) as facet_count, null, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ), fips_facet as ( select 'fips' as _facet, fips as facet_name, count(*) as facet_count, null, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ), county_facet as ( select 'county' as _facet, county as facet_name, count(*) as facet_count, null, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ), total_count as ( select 'COUNT' as _facet, '' as facet_name, count(*) as facet_count, null, null, null, null, null, null, null from cte ) select * from truncated union all select * from state_facet union all select * from fips_facet union all select * from county_facet union all select * from total_count

_facet | facet_name | facet_count | rowid | date | county | state | fips | cases | deaths -- | -- | -- | -- | -- | -- | -- | -- | -- | --   |   |   | 1917344 | 2021-11-15 | Autauga | Alabama | 1001 | 10407 | 154   |   |   | 1917345 | 2021-11-15 | Baldwin | Alabama | 1003 | 37875 | 581   |   |   | 1917346 | 2021-11-15 | Barbour | Alabama | 1005 | 3648 | 79   |   |   | 1917347 | 2021-11-15 | Bibb | Alabama | 1007 | 4317 | 92 state | Texas | 148028 |   |   |   |   |   |   |   state | Georgia | 96249 |   |   |   |   |   |   |   state | Virginia | 79315 |   |   |   |   |   |   |   fips |   | 17580 |   |   |   |   |   |   |   fips | 53061 | 665 |   |   |   |   |   |   |   fips | 17031 | 662 |   |   |   |   |   |   |   county | Washington | 18666 |   |   |   |   |   |   |   county | Unknown | 15840 |   |   |   |   |   |   |   county | Jefferson | 15637 |   |   |   |   |   |   |   COUNT |   | 1920593 |   |   |   |   |   |   |  

(One thing I really like about this pattern is that it should work exactly the same when used to facet the results of arbitrary SQL queries as it does when faceting results from the table page.)

It's going to take another 15 minutes for the build to finish and deploy the version with _trace=1: https://github.com/simonw/covid-19-datasette/actions/runs/1469150112

Trace now enabled: https://global-power-plants.datasettes.com/global-power-plants/global-power-plants?_facet_size=3&_size=2&_nocount=1&_trace=1

Here are the relevant traces: json [ { "type": "sql", "start": 31.214430154, "end": 31.214817089, "duration_ms": 0.3869350000016425, "traceback": [ " File \"/usr/local/lib/python3.8/site-packages/datasette/views/base.py\", line 262, in get\n return await self.view_get(\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/views/base.py\", line 477, in view_get\n response_or_template_contexts = await self.data(\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/views/table.py\", line 705, in data\n results = await db.execute(sql, params, truncate=True, **extra_args)\n" ], "database": "global-power-plants", "sql": "select rowid, country, country_long, name, gppd_idnr, capacity_mw, latitude, longitude, primary_fuel, other_fuel1, other_fuel2, other_fuel3, commissioning_year, owner, source, url, geolocation_source, wepp_id, year_of_capacity_data, generation_gwh_2013, generation_gwh_2014, generation_gwh_2015, generation_gwh_2016, generation_gwh_2017, generation_data_source, estimated_generation_gwh from [global-power-plants] order by rowid limit 3", "params": {} }, { "type": "sql", "start": 31.215234586, "end": 31.220110342, "duration_ms": 4.875756000000564, "traceback": [ " File \"/usr/local/lib/python3.8/site-packages/datasette/views/table.py\", line 760, in data\n ) = await facet.facet_results()\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/facets.py\", line 212, in facet_results\n facet_rows_results = await self.ds.execute(\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/app.py\", line 634, in execute\n return await self.databases[db_name].execute(\n" ], "database": "global-power-plants", "sql": "select country_long as value, count(*) as count from (\n select rowid, country, country_long, name, gppd_idnr, capacity_mw, latitude, longitude, primary_fuel, other_fuel1, other_fuel2, other_fuel3, commissioning_year, owner, source, url, geolocation_source, wepp_id, year_of_capacity_data, generation_gwh_2013, generation_gwh_2014, generation_gwh_2015, generation_gwh_2016, generation_gwh_2017, generation_data_source, estimated_generation_gwh from [global-power-plants] \n )\n where country_long is not null\n group by country_long order by count desc, value limit 4", "params": [] }, { "type": "sql", "start": 31.221062485, "end": 31.228968364, "duration_ms": 7.905878999999061, "traceback": [ " File \"/usr/local/lib/python3.8/site-packages/datasette/views/table.py\", line 760, in data\n ) = await facet.facet_results()\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/facets.py\", line 212, in facet_results\n facet_rows_results = await self.ds.execute(\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/app.py\", line 634, in execute\n return await self.databases[db_name].execute(\n" ], "database": "global-power-plants", "sql": "select owner as value, count(*) as count from (\n select rowid, country, country_long, name, gppd_idnr, capacity_mw, latitude, longitude, primary_fuel, other_fuel1, other_fuel2, other_fuel3, commissioning_year, owner, source, url, geolocation_source, wepp_id, year_of_capacity_data, generation_gwh_2013, generation_gwh_2014, generation_gwh_2015, generation_gwh_2016, generation_gwh_2017, generation_data_source, estimated_generation_gwh from [global-power-plants] \n )\n where owner is not null\n group by owner order by count desc, value limit 4", "params": [] }, { "type": "sql", "start": 31.229809757, "end": 31.253902162, "duration_ms": 24.09240499999754, "traceback": [ " File \"/usr/local/lib/python3.8/site-packages/datasette/views/table.py\", line 760, in data\n ) = await facet.facet_results()\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/facets.py\", line 212, in facet_results\n facet_rows_results = await self.ds.execute(\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/app.py\", line 634, in execute\n return await self.databases[db_name].execute(\n" ], "database": "global-power-plants", "sql": "select primary_fuel as value, count(*) as count from (\n select rowid, country, country_long, name, gppd_idnr, capacity_mw, latitude, longitude, primary_fuel, other_fuel1, other_fuel2, other_fuel3, commissioning_year, owner, source, url, geolocation_source, wepp_id, year_of_capacity_data, generation_gwh_2013, generation_gwh_2014, generation_gwh_2015, generation_gwh_2016, generation_gwh_2017, generation_data_source, estimated_generation_gwh from [global-power-plants] \n )\n where primary_fuel is not null\n group by primary_fuel order by count desc, value limit 4", "params": [] }, { "type": "sql", "start": 31.255699745, "end": 31.256243889, "duration_ms": 0.544143999999136, "traceback": [ " File \"/usr/local/lib/python3.8/site-packages/datasette/facets.py\", line 145, in suggest\n row_count = await self.get_row_count()\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/facets.py\", line 132, in get_row_count\n await self.ds.execute(\n", " File \"/usr/local/lib/python3.8/site-packages/datasette/app.py\", line 634, in execute\n return await self.databases[db_name].execute(\n" ], "database": "global-power-plants", "sql": "select count(*) from (select rowid, country, country_long, name, gppd_idnr, capacity_mw, latitude, longitude, primary_fuel, other_fuel1, other_fuel2, other_fuel3, commissioning_year, owner, source, url, geolocation_source, wepp_id, year_of_capacity_data, generation_gwh_2013, generation_gwh_2014, generation_gwh_2015, generation_gwh_2016, generation_gwh_2017, generation_data_source, estimated_generation_gwh from [global-power-plants] )", "params": [] } ] fetch rows: 0.3869350000016425 ms facet country_long: 4.875756000000564 ms facet owner: 7.905878999999061 ms facet primary_fuel: 24.09240499999754 ms count: 0.544143999999136 ms Total = 37.8ms

I modified the query to include the total count as well: https://global-power-plants.datasettes.com/global-power-plants?sql=with+cte+as+%28%0D%0A++select+rowid%2C+country%2C+country_long%2C+name%2C+owner%2C+primary_fuel%0D%0A++from+%5Bglobal-power-plants%5D%0D%0A%29%2C%0D%0Atruncated+as+%28%0D%0A++select+null+as+_facet%2C+null+as+facet_name%2C+null+as+facet_count%2C+rowid%2C+country%2C+country_long%2C+name%2C+owner%2C+primary_fuel%0D%0A++from+cte+order+by+rowid+limit+4%0D%0A%29%2C%0D%0Acountry_long_facet+as+%28%0D%0A++select+%27country_long%27+as+_facet%2C+country_long+as+facet_name%2C+count%28%29+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A%29%2C%0D%0Aowner_facet+as+%28%0D%0A++select+%27owner%27+as+_facet%2C+owner+as+facet_name%2C+count%28%29+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A%29%2C%0D%0Aprimary_fuel_facet+as+%28%0D%0A++select+%27primary_fuel%27+as+_facet%2C+primary_fuel+as+facet_name%2C+count%28%29+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte+group+by+facet_name+order+by+facet_count+desc+limit+3%0D%0A%29%2C%0D%0Atotal_count+as+%28%0D%0A++select+%27COUNT%27+as+_facet%2C+%27%27+as+facet_name%2C+count%28%29+as+facet_count%2C%0D%0A++null%2C+null%2C+null%2C+null%2C+null%2C+null%0D%0A++from+cte%0D%0A%29%0D%0Aselect++from+truncated%0D%0Aunion+all+select++from+country_long_facet%0D%0Aunion+all+select++from+owner_facet%0D%0Aunion+all+select++from+primary_fuel_facet%0D%0Aunion+all+select+*+from+total_count&_trace=1

sql with cte as ( select rowid, country, country_long, name, owner, primary_fuel from [global-power-plants] ), truncated as ( select null as _facet, null as facet_name, null as facet_count, rowid, country, country_long, name, owner, primary_fuel from cte order by rowid limit 4 ), country_long_facet as ( select 'country_long' as _facet, country_long as facet_name, count(*) as facet_count, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ), owner_facet as ( select 'owner' as _facet, owner as facet_name, count(*) as facet_count, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ), primary_fuel_facet as ( select 'primary_fuel' as _facet, primary_fuel as facet_name, count(*) as facet_count, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ), total_count as ( select 'COUNT' as _facet, '' as facet_name, count(*) as facet_count, null, null, null, null, null, null from cte ) select * from truncated union all select * from country_long_facet union all select * from owner_facet union all select * from primary_fuel_facet union all select * from total_count The trace says that query took 34.801436999998714 ms.

To my huge surprise, this convoluted optimization only shaves the sum query time down from 37.8ms to 34.8ms!

That entire database file is just 11.1 MB though. Maybe it would make a meaningful difference on something larger?

The query takes 42.794ms to run.

Here's the equivalent page using separate queries: https://global-power-plants.datasettes.com/global-power-plants/global-power-plants?_facet_size=3&_size=2&_nocount=1

Annoyingly I can't disable facet suggestions but keep facets.

I'm going to turn on tracing so I can see how long the separate queries took.

I came up with the following query which seems to work!

sql with cte as ( select rowid, country, country_long, name, owner, primary_fuel from [global-power-plants] ), truncated as ( select null as _facet, null as facet_name, null as facet_count, rowid, country, country_long, name, owner, primary_fuel from cte order by rowid limit 4 ), country_long_facet as ( select 'country_long' as _facet, country_long as facet_name, count(*) as facet_count, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ), owner_facet as ( select 'owner' as _facet, owner as facet_name, count(*) as facet_count, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ), primary_fuel_facet as ( select 'primary_fuel' as _facet, primary_fuel as facet_name, count(*) as facet_count, null, null, null, null, null, null from cte group by facet_name order by facet_count desc limit 3 ) select * from truncated union all select * from country_long_facet union all select * from owner_facet union all select * from primary_fuel_facet (Limits should be 101, 31, 31, 31 but I reduced size to get a shorter example table).

Results look like this:

_facet | facet_name | facet_count | rowid | country | country_long | name | owner | primary_fuel -- | -- | -- | -- | -- | -- | -- | -- | --   |   |   | 1 | AFG | Afghanistan | Kajaki Hydroelectric Power Plant Afghanistan |   | Hydro   |   |   | 2 | AFG | Afghanistan | Kandahar DOG |   | Solar   |   |   | 3 | AFG | Afghanistan | Kandahar JOL |   | Solar   |   |   | 4 | AFG | Afghanistan | Mahipar Hydroelectric Power Plant Afghanistan |   | Hydro country_long | United States of America | 8688 |   |   |   |   |   |   country_long | China | 4235 |   |   |   |   |   |   country_long | United Kingdom | 2603 |   |   |   |   |   |   owner |   | 14112 |   |   |   |   |   |   owner | Lightsource Renewable Energy | 120 |   |   |   |   |   |   owner | Cypress Creek Renewables | 109 |   |   |   |   |   |   primary_fuel | Solar | 9662 |   |   |   |   |   |   primary_fuel | Hydro | 7155 |   |   |   |   |   |   primary_fuel | Wind | 5188 |   |   |   |   |   |  

This is a neat proof of concept.

I've decided to make the clever asyncio dependency injection opt-in - so you can either decorate with @inject or you can set inject_all = True on the class - for example: ```python import asyncio from datasette.utils.asyncdi import AsyncBase, inject

class Simple(AsyncBase): def init(self): self.log = []

@inject
async def two(self):
    self.log.append("two")

@inject
async def one(self, two):
    self.log.append("one")
    return self.log

async def not_inject(self, one, two):
    return one + two

class Complex(AsyncBase): inject_all = True

def __init__(self):
    self.log = []

async def b(self):
    self.log.append("b")

async def a(self, b):
    self.log.append("a")

async def go(self, a):
    self.log.append("go")
    return self.log

```

I'm going to continue working on this in a PR.

Wrote a TIL about what I learned using TopologicalSorter: https://til.simonwillison.net/python/graphlib-topologicalsorter

New test: ```python

class Complex(AsyncBase): def init(self): self.log = []

async def d(self):
    await asyncio.sleep(random() * 0.1)
    print("LOG: d")
    self.log.append("d")

async def c(self):
    await asyncio.sleep(random() * 0.1)
    print("LOG: c")
    self.log.append("c")

async def b(self, c, d):
    print("LOG: b")
    self.log.append("b")

async def a(self, b, c):
    print("LOG: a")
    self.log.append("a")

async def go(self, a):
    print("LOG: go")
    self.log.append("go")
    return self.log

@pytest.mark.asyncio async def test_complex(): result = await Complex().go() # 'c' should only be called once assert tuple(result) in ( # c and d could happen in either order ("c", "d", "b", "a", "go"), ("d", "c", "b", "a", "go"), ) And this code passes it:python import asyncio from functools import wraps import inspect

try: import graphlib except ImportError: from . import vendored_graphlib as graphlib

class AsyncMeta(type): def new(cls, name, bases, attrs): # Decorate any items that are 'async def' methods registry = {} new_attrs = {"_registry": _registry} for key, value in attrs.items(): if inspect.iscoroutinefunction(value) and not value.__name__ == "resolve": new_attrs[key] = make_method(value) _registry[key] = new_attrs[key] else: new_attrs[key] = value # Gather graph for later dependency resolution graph = { key: { p for p in inspect.signature(method).parameters.keys() if p != "self" and not p.startswith("") } for key, method in _registry.items() } new_attrs["_graph"] = graph return super().new(cls, name, bases, new_attrs)

def make_method(method): parameters = inspect.signature(method).parameters.keys()

@wraps(method)
async def inner(self, _results=None, **kwargs):
    print("\n{}.{}({}) _results={}".format(self, method.__name__, kwargs, _results))

    # Any parameters not provided by kwargs are resolved from registry
    to_resolve = [p for p in parameters if p not in kwargs and p != "self"]
    missing = [p for p in to_resolve if p not in self._registry]
    assert (
        not missing
    ), "The following DI parameters could not be found in the registry: {}".format(
        missing
    )

    results = {}
    results.update(kwargs)
    if to_resolve:
        resolved_parameters = await self.resolve(to_resolve, _results)
        results.update(resolved_parameters)
    return_value = await method(self, **results)
    if _results is not None:
        _results[method.__name__] = return_value
    return return_value

return inner

class AsyncBase(metaclass=AsyncMeta): async def resolve(self, names, results=None): print("\n resolve: ", names) if results is None: results = {}

    # Come up with an execution plan, just for these nodes
    ts = graphlib.TopologicalSorter()
    to_do = set(names)
    done = set()
    while to_do:
        item = to_do.pop()
        dependencies = self._graph[item]
        ts.add(item, *dependencies)
        done.add(item)
        # Add any not-done dependencies to the queue
        to_do.update({k for k in dependencies if k not in done})

    ts.prepare()
    plan = []
    while ts.is_active():
        node_group = ts.get_ready()
        plan.append(node_group)
        ts.done(*node_group)

    print("plan:", plan)

    results = {}
    for node_group in plan:
        awaitables = [
            self._registry[name](
                self,
                _results=results,
                **{k: v for k, v in results.items() if k in self._graph[name]},
            )
            for name in node_group
        ]
        print("    results = ", results)
        print("    awaitables: ", awaitables)
        awaitable_results = await asyncio.gather(*awaitables)
        results.update(
            {p[0].__name__: p[1] for p in zip(awaitables, awaitable_results)}
        )

    print("  End of resolve(), returning", results)
    return {key: value for key, value in results.items() if key in names}

```

But that does seem to be the plan that TopographicalSorter provides: ```python graph = {"go": {"a"}, "a": {"b", "c"}, "b": {"c", "d"}}

ts = TopologicalSorter(graph) ts.prepare() while ts.is_active(): nodes = ts.get_ready() print(nodes) ts.done(*nodes) Outputs: ('c', 'd') ('b',) ('a',) ('go',) Also:python graph = {"go": {"d", "e", "f"}, "d": {"b", "c"}, "b": {"c"}}

ts = TopologicalSorter(graph) ts.prepare() while ts.is_active(): nodes = ts.get_ready() print(nodes) ts.done(nodes) Gives: ('e', 'f', 'c') ('b',) ('d',) ('go',) `` I'm confident thatTopologicalSorteris the way to do this. I think I need to rewrite my code to call it once to get that plan, thenawait asyncio.gather(nodes)` in turn to execute it.

My goal here is to calculate the most efficient way to resolve the different nodes, running them in parallel where possible.

So for this class:

```python class Complex(AsyncBase): async def d(self): pass

async def c(self):
    pass

async def b(self, c, d):
    pass

async def a(self, b, c):
    pass

async def go(self, a):
    pass

`` A call togo()` should do this:

• c and d in parallel
• b
• a
• go

What should be happening here instead is it should resolve the full graph and notice that c is depended on by both b and a - so it should run c first, then run the next ones in parallel.

So maybe the algorithm I'm inheriting from https://docs.python.org/3/library/graphlib.html isn't the correct algorithm?

This code is really fiddly. I just got to this version: ```python import asyncio from functools import wraps import inspect

try: import graphlib except ImportError: from . import vendored_graphlib as graphlib

class AsyncMeta(type): def new(cls, name, bases, attrs): # Decorate any items that are 'async def' methods registry = {} new_attrs = {"_registry": _registry} for key, value in attrs.items(): if inspect.iscoroutinefunction(value) and not value.__name__ == "resolve": new_attrs[key] = make_method(value) _registry[key] = new_attrs[key] else: new_attrs[key] = value # Gather graph for later dependency resolution graph = { key: { p for p in inspect.signature(method).parameters.keys() if p != "self" and not p.startswith("") } for key, method in _registry.items() } new_attrs["_graph"] = graph return super().new(cls, name, bases, new_attrs)

def make_method(method): @wraps(method) async def inner(self, _results=None, kwargs): print("inner - _results=", _results) parameters = inspect.signature(method).parameters.keys() # Any parameters not provided by kwargs are resolved from registry to_resolve = [p for p in parameters if p not in kwargs and p != "self"] missing = [p for p in to_resolve if p not in self._registry] assert ( not missing ), "The following DI parameters could not be found in the registry: {}".format( missing ) results = {} results.update(kwargs) if to_resolve: resolved_parameters = await self.resolve(to_resolve, _results) results.update(resolved_parameters) return_value = await method(self, results) if _results is not None: _results[method.name] = return_value return return_value

return inner

class AsyncBase(metaclass=AsyncMeta): async def resolve(self, names, results=None): print("\n resolve: ", names) if results is None: results = {}

    # Resolve them in the correct order
    ts = graphlib.TopologicalSorter()
    for name in names:
        ts.add(name, *self._graph[name])
    ts.prepare()

    async def resolve_nodes(nodes):
        print("    resolve_nodes", nodes)
        print("    (current results = {})".format(repr(results)))
        awaitables = [
            self._registry[name](
                self,
                _results=results,
                **{k: v for k, v in results.items() if k in self._graph[name]},
            )
            for name in nodes
            if name not in results
        ]
        print("    awaitables: ", awaitables)
        awaitable_results = await asyncio.gather(*awaitables)
        results.update(
            {p[0].__name__: p[1] for p in zip(awaitables, awaitable_results)}
        )

    if not ts.is_active():
        # Nothing has dependencies - just resolve directly
        print("    no dependencies, resolve directly")
        await resolve_nodes(names)
    else:
        # Resolve in topological order
        while ts.is_active():
            nodes = ts.get_ready()
            print("    ts.get_ready() returned nodes:", nodes)
            await resolve_nodes(nodes)
            for node in nodes:
                ts.done(node)

    print("  End of resolve(), returning", results)
    return {key: value for key, value in results.items() if key in names}

With this test:python class Complex(AsyncBase): def init(self): self.log = []

async def c(self):
    print("LOG: c")
    self.log.append("c")

async def b(self, c):
    print("LOG: b")
    self.log.append("b")

async def a(self, b, c):
    print("LOG: a")
    self.log.append("a")

async def go(self, a):
    print("LOG: go")
    self.log.append("go")
    return self.log

@pytest.mark.asyncio async def test_complex(): result = await Complex().go() # 'c' should only be called once assert result == ["c", "b", "a", "go"] ``` This test sometimes passes, and sometimes fails!

Output for a pass: ``` tests/test_asyncdi.py inner - _results= None

resolve: ['a'] ts.get_ready() returned nodes: ('c', 'b') resolve_nodes ('c', 'b') (current results = {}) awaitables: [<coroutine object Complex.c at 0x1074ac890>, <coroutine object Complex.b at 0x1074ac820>] inner - _results= {} LOG: c inner - _results= {'c': None}

resolve: ['c'] ts.get_ready() returned nodes: ('c',) resolve_nodes ('c',) (current results = {'c': None}) awaitables: [] End of resolve(), returning {'c': None} LOG: b ts.get_ready() returned nodes: ('a',) resolve_nodes ('a',) (current results = {'c': None, 'b': None}) awaitables: [<coroutine object Complex.a at 0x1074ac7b0>] inner - _results= {'c': None, 'b': None} LOG: a End of resolve(), returning {'c': None, 'b': None, 'a': None} LOG: go Output for a fail: tests/test_asyncdi.py inner - _results= None

resolve: ['a'] ts.get_ready() returned nodes: ('b', 'c') resolve_nodes ('b', 'c') (current results = {}) awaitables: [<coroutine object Complex.b at 0x10923c890>, <coroutine object Complex.c at 0x10923c820>] inner - _results= {}

resolve: ['c'] ts.get_ready() returned nodes: ('c',) resolve_nodes ('c',) (current results = {}) awaitables: [<coroutine object Complex.c at 0x10923c6d0>] inner - _results= {} LOG: c inner - _results= {'c': None} LOG: c End of resolve(), returning {'c': None} LOG: b ts.get_ready() returned nodes: ('a',) resolve_nodes ('a',) (current results = {'c': None, 'b': None}) awaitables: [<coroutine object Complex.a at 0x10923c6d0>] inner - _results= {'c': None, 'b': None} LOG: a End of resolve(), returning {'c': None, 'b': None, 'a': None} LOG: go F

=================================================================================================== FAILURES =================================================================================================== _______________ test_complex _________________

@pytest.mark.asyncio
async def test_complex():
    result = await Complex().go()
    # 'c' should only be called once

  assert result == ["c", "b", "a", "go"]

E AssertionError: assert ['c', 'c', 'b', 'a', 'go'] == ['c', 'b', 'a', 'go'] E At index 1 diff: 'c' != 'b' E Left contains one more item: 'go' E Use -v to get the full diff

tests/test_asyncdi.py:48: AssertionError ================== short test summary info ================================ FAILED tests/test_asyncdi.py::test_complex - AssertionError: assert ['c', 'c', 'b', 'a', 'go'] == ['c', 'b', 'a', 'go'] ``` I figured out why this is happening.

a requires b and c

b also requires c

The code decides to run b and c in parallel.

If c completes first, then when b runs it gets to use the already-calculated result for c - so it doesn't need to call c again.

If b gets to that point before c does it also needs to call c.

Here's the latest version of my weird dependency injection async class: ```python import inspect

class AsyncMeta(type): def new(cls, name, bases, attrs): # Decorate any items that are 'async def' methods _registry = {} new_attrs = {"_registry": _registry} for key, value in attrs.items(): if inspect.iscoroutinefunction(value) and not value.name == "resolve": new_attrs[key] = make_method(value) _registry[key] = new_attrs[key] else: new_attrs[key] = value

    # Topological sort of _registry by parameter dependencies
    graph = {
        key: {
            p for p in inspect.signature(method).parameters.keys()
            if p != "self" and not p.startswith("_")
        }
        for key, method in _registry.items()
    }
    new_attrs["_graph"] = graph
    return super().__new__(cls, name, bases, new_attrs)

def make_method(method): @wraps(method) async def inner(self, kwargs): parameters = inspect.signature(method).parameters.keys() # Any parameters not provided by kwargs are resolved from registry to_resolve = [p for p in parameters if p not in kwargs and p != "self"] missing = [p for p in to_resolve if p not in self._registry] assert ( not missing ), "The following DI parameters could not be found in the registry: {}".format( missing ) results = {} results.update(kwargs) results.update(await self.resolve(to_resolve)) return await method(self, results)

return inner

bad = [0]

class AsyncBase(metaclass=AsyncMeta): async def resolve(self, names): print(" resolve({})".format(names)) results = {} # Resolve them in the correct order ts = TopologicalSorter() ts2 = TopologicalSorter() print(" names = ", names) print(" self._graph = ", self._graph) for name in names: if self._graph[name]: ts.add(name, self._graph[name]) ts2.add(name, self._graph[name]) print(" static_order =", tuple(ts2.static_order())) ts.prepare() while ts.is_active(): print(" is_active, i = ", bad[0]) bad[0] += 1 if bad[0] > 20: print(" Infinite loop?") break nodes = ts.get_ready() print(" Do nodes:", nodes) awaitables = [self._registryname for name in nodes] print(" awaitables: ", awaitables) awaitable_results = await asyncio.gather(*awaitables) results.update({ p[0].name: p[1] for p in zip(awaitables, awaitable_results) }) print(results) for node in nodes: ts.done(node)

    return results

Example usage:python class Foo(AsyncBase): async def graa(self, boff): print("graa") return 5 async def boff(self): print("boff") return 8 async def other(self, boff, graa): print("other") return 5 + boff + graa

foo = Foo() await foo.other() Output: resolve(['boff', 'graa']) names = ['boff', 'graa'] self._graph = {'graa': {'boff'}, 'boff': set(), 'other': {'graa', 'boff'}} static_order = ('boff', 'graa') is_active, i = 0 Do nodes: ('boff',) awaitables: [<coroutine object Foo.boff at 0x10bd81a40>] resolve([]) names = [] self._graph = {'graa': {'boff'}, 'boff': set(), 'other': {'graa', 'boff'}} static_order = () boff {'boff': 8} is_active, i = 1 Do nodes: ('graa',) awaitables: [<coroutine object Foo.graa at 0x10d66b340>] resolve([]) names = [] self._graph = {'graa': {'boff'}, 'boff': set(), 'other': {'graa', 'boff'}} static_order = () graa {'boff': 8, 'graa': 5} other 18 ```

I'm going to take another look at this: - https://github.com/simonw/datasette/issues/878

OK, I'm ready to start working on this today.

I'm going to go with a default representation that looks like this:

json { "rows": [ {"id": 1, "name": "One"}, {"id": 2, "name": "Two"} ], "next_url": null } Note that there's no count - all it provides is the current selection of results and an indication as to how the next can be retrieved (null if there are no more results).

I'll implement ?_extra= to provide everything else.

This is mainly happening here: - https://github.com/simonw/datasette/issues/782

Tests are failing and I think it's because the facets come back in different orders, need a tie-breaker. https://github.com/simonw/datasette/runs/4219325197?check_suite_focus=true

I'm inclined to create a Sphinx reference documentation page for this, as I did for sqlite-utils here: https://sqlite-utils.datasette.io/en/stable/reference.html

Requested here:

• https://github.com/pypa/trove-classifiers/pull/85

I'd been wondering how to get new classifiers into Trove - thanks, I'll give this a go.

See: - #830

Also note that this demo data is using a SQL view to create the JSON arrays - the view is defined as such:

sql CREATE VIEW ads_with_targets as select ads.*, json_group_array(targets.name) as target_names from ads join ad_targets on ad_targets.ad_id = ads.id join targets on ad_targets.target_id = targets.id group by ad_targets.ad_id; So running JSON faceting on top of that view is a pretty big ask!

Actually with the cache warmed up it looks like the facet query is taking 150ms which is good enough.

I tried this and it seems to work correctly: python for source_and_config in self.get_configs(): config = source_and_config["config"] source = source_and_config["source"] column = config.get("column") or config["simple"] facet_sql = """ with inner as ({sql}), deduped_array_items as ( select distinct j.value, inner.* from json_each([inner].{col}) j join inner ) select value as value, count(*) as count from deduped_array_items group by value order by count(*) desc limit {limit} """.format( col=escape_sqlite(column), sql=self.sql, limit=facet_size + 1 ) The queries are very slow though - I had to bump up to 2s time limit even against only a view returning 3,499 rows.

This looks like it might work: sql with inner as ( select * from ads_with_targets where :p0 in ( select value from json_each([ads_with_targets].[target_names]) ) ), deduped_array_items as ( select distinct j.value, inner.* from json_each([inner].[target_names]) j join inner ) select value, count(*) from deduped_array_items group by value order by count(*) desc

It uses a CTE which were introduced in SQLite 3.8 - and AWS Lambda Python 3.9 still provides 3.7 - but I've checked and I can use pysqlite3-binary to work around that there so I'm OK relying on CTEs for this.