Weird save() crash and bad error from Gel

[1st1]

this script

from __future__ import annotations
import asyncio
import uuid, json, typing, pydantic, gel  # noqa: E401, F401
from typing import TypeVar, Generic, Any
from models import default
import models.std as std
import models as m
from rich import print


client = gel.create_client()._with_debug(save_postcheck=True)

alice = client.get(default.User.filter(name="Alice"))
tsl = default.TestSingleLinks(
    req_wprop_friend=default.TestSingleLinks.req_wprop_friend.link(alice, strength=100),
    req_friend=alice,
    opt_friend=alice,
    opt_wprop_friend=default.TestSingleLinks.opt_wprop_friend.link(alice, strength=50),
)

client.save(tsl)

billie = client.get(default.User.filter(name="Billie"))
tsl2 = client.get(
    default.TestSingleLinks.select(
        "*",
        req_wprop_friend=True,
        req_friend=True,
        opt_friend=True,
        opt_wprop_friend=True,
        comp_req_wprop_friend=True,
        comp_req_friend=True,
        comp_opt_friend=True,
        comp_opt_wprop_friend=True,
    ).filter(id=tsl.id)
)

tsl2.req_wprop_friend = default.TestSingleLinks.req_wprop_friend.link(billie)
tsl2.opt_wprop_friend = default.TestSingleLinks.opt_wprop_friend.link(billie)

# Save the changes
client.save(tsl2)

crashes with

Traceback (most recent call last):
  File "/Users/yury/dev/edge/geltest/t.py", line 42, in <module>
    client.save(tsl2)
  File "/Users/yury/dev/edge/edgedb-python/gel/blocking_client.py", line 675, in save
    batch_ids = tx.wait()
                ^^^^^^^^^
  File "/Users/yury/dev/edge/edgedb-python/gel/blocking_client.py", line 610, in wait
    return iter_coroutine(self._wait())
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/yury/dev/edge/edgedb-python/gel/blocking_client.py", line 76, in iter_coroutine
    coro.send(None)
  File "/Users/yury/dev/edge/edgedb-python/gel/blocking_client.py", line 616, in _wait
    return await self._connection.batch_query(ops)  # type: ignore [no-any-return]
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/yury/dev/edge/edgedb-python/gel/base_client.py", line 305, in batch_query
    rv = await self._protocol.batch_execute(ctxs)
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "gel/protocol/protocol.pyx", line 886, in batch_execute
  File "gel/protocol/protocol.pyx", line 782, in _execute_batch
gel.errors.InternalServerError: more than one row returned by a subquery used as an expression

on this query

                with __query := (
                    with __all_data := <array<tuple<array<std::uuid>,std::uuid,std::uuid>>>$0
                    for __data in std::array_unpack(__all_data) union (
                        (                update <default::TestSingleLinks>__data.2
                set { opt_wprop_friend := <default::User><std::uuid>(  select std::array_unpack(__data.0) limit 1), req_wprop_friend := <default::User><std::uuid>__data.1 }
            )
                    )
                ) select __query{*}

with these args

             [([UUID('b4e784f8-7286-11f0-8040-df194aa8f2b4')], UUID('b4e784f8-7286-11f0-8040-df194aa8f2b4'), UUID('53bca92e-7321-11f0-bfef-cf0403df7d1b'))]

[1st1]

source PR #839

it's the new xfail test test_modelgen_save_reload_linkprops_01

[1st1]

(I suggest making a new python project, gel init it, copy orm.gel into it, then run orm.edgeql in $ gel, and then use it as a testbed as running tests is slow.

I also usually do gel instance backup and then I can

gel instance restore --backup-id 01987ddd-bcf9-7811-b2b0-9abb61852550 --non-interactive && python t.py

and run t.py on properly init db free from other t.py runs

cc @aljazerzen @dnwpark @vpetrovykh

[1st1]

this is what i use to debug save() queries usually

diff --git a/gel/blocking_client.py b/gel/blocking_client.py
index 9e25c4e..bf5e242 100644
--- a/gel/blocking_client.py
+++ b/gel/blocking_client.py
@@ -668,13 +668,14 @@ class Client(
         for tx in self._batch():
             with tx:
                 executor = make_executor()
-
                 for batches in executor:
                     for batch in batches:
+                        print(batch.query.__edgeql__()[1], batch.args)
                         tx.send_query(batch.query, batch.args)
-                    batch_ids = tx.wait()
-                    for ids, batch in zip(batch_ids, batches, strict=True):
-                        batch.feed_db_data(ids)
+                        batch_ids = tx.wait()
+                        for ids, batch in zip(batch_ids, batches, strict=True):
+                            batch.feed_db_data(ids)
+                        print("\n\n\n\n\n==============")

                 executor.commit()

[msullivan]

test_modelgen_save_reload_linkprops_01

I think you mean test_modelgen_save_reload_links_04?

The crash is fixed by geldata/gel#8932, which was released in 6.10.

The test still fails due to:

  File "/home/sully/src/e/edgedb-python/tests/test_model_generator.py", line 3973, in test_modelgen_save_reload_links_04
    with self.assertRaisesRegex(
  File "/usr/local/lib/python3.12/contextlib.py", line 144, in __exit__
    next(self.gen)
  File "/home/sully/src/e/edgedb-python/gel/_testbase.py", line 389, in assertRaisesRegex
    with super().assertRaisesRegex(exception, regex, msg=msg):
  File "/usr/local/lib/python3.12/unittest/case.py", line 263, in __exit__
    self._raiseFailure("{} not raised".format(exc_name))
  File "/usr/local/lib/python3.12/unittest/case.py", line 200, in _raiseFailure
    raise self.test_case.failureException(msg)
AssertionError: AttributeError not raised

[1st1]

yes, the test is outdated, here's a diff of what should be working

diff --git a/tests/test_model_generator.py b/tests/test_model_generator.py
index 364a6ba..1ae55de 100644
--- a/tests/test_model_generator.py
+++ b/tests/test_model_generator.py
@@ -3862,9 +3862,6 @@ class TestModelGenerator(tb.ModelTestCase):
         # Was alice reloaded? That's the object used in tlt.user
         self.assertEqual(alice.nickname, "magical unicorn")

-    @tb.xfail
-    # error from saving tsl2
-    # more than one row returned by a subquery used as an expression
     def test_modelgen_save_reload_links_04(self):
         from models.orm import default

@@ -3885,7 +3882,7 @@ class TestModelGenerator(tb.ModelTestCase):

         self.client.sync(tsl)

-        # Computed links should not be set after save
+        # Computed links should not be set -- we never fetched them
         with self.assertRaisesRegex(
             AttributeError, "'comp_req_wprop_friend' is not set"
         ):
@@ -3971,26 +3968,11 @@ class TestModelGenerator(tb.ModelTestCase):
         self.assertEqual(tsl2.opt_friend.name, "Billie")
         self.assertEqual(tsl2.opt_wprop_friend.name, "Billie")

-        # But computed links should no longer be set after the save
-        with self.assertRaisesRegex(
-            AttributeError, "'comp_req_wprop_friend' is not set"
-        ):
-            assert tsl2.comp_req_wprop_friend is not None  # access field
-
-        with self.assertRaisesRegex(
-            AttributeError, "'comp_req_friend' is not set"
-        ):
-            assert tsl2.comp_req_friend is not None  # access field
-
-        with self.assertRaisesRegex(
-            AttributeError, "'comp_opt_friend' is not set"
-        ):
-            assert tsl2.comp_opt_friend is not None  # access field
-
-        with self.assertRaisesRegex(
-            AttributeError, "'comp_opt_wprop_friend' is not set"
-        ):
-            assert tsl2.comp_opt_wprop_friend is not None  # access field
+        # Computed links should be up to date after the save
+        self.assertEqual(tsl2.comp_req_wprop_friend.name, "Billie")
+        self.assertEqual(tsl2.comp_req_friend.name, "Billie")
+        self.assertEqual(tsl2.comp_opt_friend.name, "Billie")
+        self.assertEqual(tsl2.comp_opt_wprop_friend.name, "Billie")

     def test_modelgen_save_reload_links_05(self):
         from models.orm import default

But it's now broken in a different place: tsl2.comp_req_wprop_friend.name computed shows 'Alice' instead of 'Billie' -- sync() call did not refetch the computed

[1st1]

#866

[msullivan]

I think this is fixed now, then?