Get Datasette compatible with Pyodide

[simonw]

I've already got this working as a prototype. Here are the changes I had to make:

• Replace the two dependencies that don't publish pure Python wheels to PyPI: click-default-group and python-baseconv
• Get Datasette to work without threading - which it turns out is exclusively used for database connections
• Make the uvicorn dependency optional (only needed when Datasette runs in the CLI)

TODO:

• Switch to click-default-group-wheel
• Remove python-baseconv dependency #1734
• Work around uvicorn import error
• Datasette setting to disable threading (for Pyodide) #1735
• Automated test for Pyodide compatibility #1737

Goal is to be able to do the following directly in https://pyodide.org/en/stable/console.html

import micropip
await micropip.install("datasette")
from datasette.app import Datasette
ds = Datasette()
await ds.client.get("/.json")

[simonw]

I released a click-default-group-wheel package to solve that dependency issue. I've already upgraded sqlite-utils to that, so now you can use that in Pyodide:

• Depend on click-default-group-wheel sqlite-utils#429

python-baseconv is only used for actor cookie expiration times:

datasette/datasette/actor_auth_cookie.py

Lines 16 to 20 in 0a7621f

	expires_at = decoded.get("e")
	if expires_at:
	timestamp = int(baseconv.base62.decode(expires_at))
	if time.time() > timestamp:
	return None

Datasette never actually sets that cookie itself - it instead encourages plugins to set it in the authentication documentation here: https://docs.datasette.io/en/0.61.1/authentication.html#including-an-expiry-time

[simonw]

I was going to vendor baseconv.py, but then I reconsidered - what if there are plugins out there that expect import baseconv to work because they have dependend on Datasette?

I used https://cs.github.com/ and as far as I can tell there aren't any!

So I'm going to remove that dependency and work out a smarter way to do this - probably by providing a utility function within Datasette itself.

[simonw]

Here's the full diff I applied to Datasette to get it fully working in Pyodide:

94a3171...8af32bc

And as a visible diff:

diff --git a/datasette/app.py b/datasette/app.py
index d269372..6c0c5fc 100644
--- a/datasette/app.py
+++ b/datasette/app.py
@@ -15,7 +15,6 @@ import pkg_resources
 import re
 import secrets
 import sys
-import threading
 import traceback
 import urllib.parse
 from concurrent import futures
@@ -26,7 +25,6 @@ from itsdangerous import URLSafeSerializer
 from jinja2 import ChoiceLoader, Environment, FileSystemLoader, PrefixLoader
 from jinja2.environment import Template
 from jinja2.exceptions import TemplateNotFound
-import uvicorn
 
 from .views.base import DatasetteError, ureg
 from .views.database import DatabaseDownload, DatabaseView
@@ -813,7 +811,6 @@ class Datasette:
             },
             "datasette": datasette_version,
             "asgi": "3.0",
-            "uvicorn": uvicorn.__version__,
             "sqlite": {
                 "version": sqlite_version,
                 "fts_versions": fts_versions,
@@ -854,23 +851,7 @@ class Datasette:
         ]
 
     def _threads(self):
-        threads = list(threading.enumerate())
-        d = {
-            "num_threads": len(threads),
-            "threads": [
-                {"name": t.name, "ident": t.ident, "daemon": t.daemon} for t in threads
-            ],
-        }
-        # Only available in Python 3.7+
-        if hasattr(asyncio, "all_tasks"):
-            tasks = asyncio.all_tasks()
-            d.update(
-                {
-                    "num_tasks": len(tasks),
-                    "tasks": [_cleaner_task_str(t) for t in tasks],
-                }
-            )
-        return d
+        return {"num_threads": 0, "threads": []}
 
     def _actor(self, request):
         return {"actor": request.actor}
diff --git a/datasette/database.py b/datasette/database.py
index ba594a8..b50142d 100644
--- a/datasette/database.py
+++ b/datasette/database.py
@@ -4,7 +4,6 @@ from pathlib import Path
 import janus
 import queue
 import sys
-import threading
 import uuid
 
 from .tracer import trace
@@ -21,8 +20,6 @@ from .utils import (
 )
 from .inspect import inspect_hash
 
-connections = threading.local()
-
 AttachedDatabase = namedtuple("AttachedDatabase", ("seq", "name", "file"))
 
 
@@ -43,12 +40,12 @@ class Database:
         self.hash = None
         self.cached_size = None
         self._cached_table_counts = None
-        self._write_thread = None
-        self._write_queue = None
         if not self.is_mutable and not self.is_memory:
             p = Path(path)
             self.hash = inspect_hash(p)
             self.cached_size = p.stat().st_size
+        self._read_connection = None
+        self._write_connection = None
 
     @property
     def cached_table_counts(self):
@@ -134,60 +131,17 @@ class Database:
         return results
 
     async def execute_write_fn(self, fn, block=True):
-        task_id = uuid.uuid5(uuid.NAMESPACE_DNS, "datasette.io")
-        if self._write_queue is None:
-            self._write_queue = queue.Queue()
-        if self._write_thread is None:
-            self._write_thread = threading.Thread(
-                target=self._execute_writes, daemon=True
-            )
-            self._write_thread.start()
-        reply_queue = janus.Queue()
-        self._write_queue.put(WriteTask(fn, task_id, reply_queue))
-        if block:
-            result = await reply_queue.async_q.get()
-            if isinstance(result, Exception):
-                raise result
-            else:
-                return result
-        else:
-            return task_id
-
-    def _execute_writes(self):
-        # Infinite looping thread that protects the single write connection
-        # to this database
-        conn_exception = None
-        conn = None
-        try:
-            conn = self.connect(write=True)
-            self.ds._prepare_connection(conn, self.name)
-        except Exception as e:
-            conn_exception = e
-        while True:
-            task = self._write_queue.get()
-            if conn_exception is not None:
-                result = conn_exception
-            else:
-                try:
-                    result = task.fn(conn)
-                except Exception as e:
-                    sys.stderr.write("{}\n".format(e))
-                    sys.stderr.flush()
-                    result = e
-            task.reply_queue.sync_q.put(result)
+        # We always treat it as if block=True now
+        if self._write_connection is None:
+            self._write_connection = self.connect(write=True)
+            self.ds._prepare_connection(self._write_connection, self.name)
+        return fn(self._write_connection)
 
     async def execute_fn(self, fn):
-        def in_thread():
-            conn = getattr(connections, self.name, None)
-            if not conn:
-                conn = self.connect()
-                self.ds._prepare_connection(conn, self.name)
-                setattr(connections, self.name, conn)
-            return fn(conn)
-
-        return await asyncio.get_event_loop().run_in_executor(
-            self.ds.executor, in_thread
-        )
+        if self._read_connection is None:
+            self._read_connection = self.connect()
+            self.ds._prepare_connection(self._read_connection, self.name)
+        return fn(self._read_connection)
 
     async def execute(
         self,
diff --git a/setup.py b/setup.py
index 7f0562f..c41669c 100644
--- a/setup.py
+++ b/setup.py
@@ -44,20 +44,20 @@ setup(
     install_requires=[
         "asgiref>=3.2.10,<3.6.0",
         "click>=7.1.1,<8.2.0",
-        "click-default-group~=1.2.2",
+        # "click-default-group~=1.2.2",
         "Jinja2>=2.10.3,<3.1.0",
         "hupper~=1.9",
         "httpx>=0.20",
         "pint~=0.9",
         "pluggy>=1.0,<1.1",
-        "uvicorn~=0.11",
+        # "uvicorn~=0.11",
         "aiofiles>=0.4,<0.9",
         "janus>=0.6.2,<1.1",
         "asgi-csrf>=0.9",
         "PyYAML>=5.3,<7.0",
         "mergedeep>=1.1.1,<1.4.0",
         "itsdangerous>=1.1,<3.0",
-        "python-baseconv==1.2.2",
+        # "python-baseconv==1.2.2",
     ],
     entry_points="""
         [console_scripts]

[simonw]

Maybe I can leave uvicorn as a dependency? Installing it works OK, it only generates errors when you try to import it:

Welcome to the Pyodide terminal emulator 🐍
Python 3.10.2 (main, Apr  9 2022 20:52:01) on WebAssembly VM
Type "help", "copyright", "credits" or "license" for more information.
>>> import micropip
>>> await micropip.install("uvicorn")
>>> import uvicorn
Traceback (most recent call last):
  File "<console>", line 1, in <module>
  File "/lib/python3.10/site-packages/uvicorn/__init__.py", line 1, in <module>
    from uvicorn.config import Config
  File "/lib/python3.10/site-packages/uvicorn/config.py", line 8, in <module>
    import ssl
  File "/lib/python3.10/ssl.py", line 98, in <module>
    import _ssl             # if we can't import it, let the error propagate
ModuleNotFoundError: No module named '_ssl'
>>> import ssl
>>> import uvicorn
Traceback (most recent call last):
  File "<console>", line 1, in <module>
  File "/lib/python3.10/site-packages/uvicorn/__init__.py", line 2, in <module>
    from uvicorn.main import Server, main, run
  File "/lib/python3.10/site-packages/uvicorn/main.py", line 24, in <module>
    from uvicorn.supervisors import ChangeReload, Multiprocess
  File "/lib/python3.10/site-packages/uvicorn/supervisors/__init__.py", line 3, in <module>
    from uvicorn.supervisors.basereload import BaseReload
  File "/lib/python3.10/site-packages/uvicorn/supervisors/basereload.py", line 12, in <module>
    from uvicorn.subprocess import get_subprocess
  File "/lib/python3.10/site-packages/uvicorn/subprocess.py", line 14, in <module>
    multiprocessing.allow_connection_pickling()
  File "/lib/python3.10/multiprocessing/context.py", line 170, in allow_connection_pickling
    from . import connection
  File "/lib/python3.10/multiprocessing/connection.py", line 21, in <module>
    import _multiprocessing
ModuleNotFoundError: No module named '_multiprocessing'
>>> import multiprocessing
>>> import uvicorn
Traceback (most recent call last):
  File "<console>", line 1, in <module>
  File "/lib/python3.10/site-packages/uvicorn/__init__.py", line 2, in <module>
    from uvicorn.main import Server, main, run
  File "/lib/python3.10/site-packages/uvicorn/main.py", line 24, in <module>
    from uvicorn.supervisors import ChangeReload, Multiprocess
  File "/lib/python3.10/site-packages/uvicorn/supervisors/__init__.py", line 3, in <module>
    from uvicorn.supervisors.basereload import BaseReload
  File "/lib/python3.10/site-packages/uvicorn/supervisors/basereload.py", line 12, in <module>
    from uvicorn.subprocess import get_subprocess
  File "/lib/python3.10/site-packages/uvicorn/subprocess.py", line 14, in <module>
    multiprocessing.allow_connection_pickling()
  File "/lib/python3.10/multiprocessing/context.py", line 170, in allow_connection_pickling
    from . import connection
  File "/lib/python3.10/multiprocessing/connection.py", line 21, in <module>
    import _multiprocessing
ModuleNotFoundError: No module named '_multiprocessing'
>>> 

Since the import ssl trick fixed the _ssl error I was hopeful that import multiprocessing could fix the _multiprocessing one, but sadly it did not.

But it looks like i can address this issue just by making import uvicorn in app.py an optional import.

[simonw]

I'm going to add a Datasette setting to disable threading entirely, designed for usage in this particular case.

I thought about adding a new setting, then I noticed this:

datasette mydatabase.db --setting num_sql_threads 10

I'm going to let users set that to 0 to disable threaded execution of SQL queries.

[simonw]

I'll release this as a 0.62a0 as soon as it's ready, so I can start testing it out in Pyodide for real.