fix(alembic/009): make enum reconcile dynamic — handles ALL postgres enums, not just 3

Original 009 only reconciled agentrole/team/taskstatus. Production hit LookupError: 'CELL' is not among defined channeltype values during seed bootstrap — channeltype (and ~16 other enums declared in 001) had uppercase members from the original create_all bootstrap that 009 never touched.

Rewrite queries pg_enum at migration time to find every enum with uppercase members, then for each: read current members, lowercase them, add desired_additions for the 3 enums with new ORM members, find every (table, column) referencing the enum via pg_attribute, RENAME-old/CREATE-new/ALTER-USING-lower/DROP-old.
This commit is contained in:
Renn F
2026-05-02 03:38:58 +02:00
parent 1d688a302c
commit 702c14eb2f
+108 -116
View File
@@ -1,19 +1,23 @@
"""Reconcile enum values with the ORM and add missing members.
"""Reconcile every postgres enum with the ORM (lowercase) + add new members.
Two adjustments to bring postgres enum types in line with the StrEnum
classes the ORM serializes:
1. Add missing values that were introduced after migration 001:
- agentrole.system (used internally for orchestrator-owned operations)
- team.fullstack (cross-team work)
- taskstatus.quarantined (safe-park state for problematic tasks)
1. Add missing members that were introduced after migration 001:
- agentrole.system (orchestrator-owned operations)
- team.fullstack (cross-team work)
- taskstatus.quarantined (safe-park state for problematic tasks)
2. If the database was previously bootstrapped via Base.metadata.create_all
(which uses the StrEnum member NAME — uppercase), reconcile the enum
values to lowercase so they match alembic 001's declared values and
the new ORM (Enum(..., values_callable=...)) serialization. This is a
conditional rebuild: if the enum already has lowercase members the
block is a no-op.
2. If any enum was previously bootstrapped via Base.metadata.create_all
(which uses the StrEnum member NAME — uppercase), reconcile every
affected enum to lowercase so values match alembic 001's declared
members and the new ORM serialization (Enum(..., values_callable=...)).
The reconcile path is dynamic: any enum found with uppercase members is
rebuilt by renaming the old type, creating a fresh lowercase type with
the same members (lower-cased) plus any desired additions, ALTER-ing
every column that references the type with `USING lower(col::text)::T`,
then dropping the old type. Repeats per affected enum.
Revision ID: 009_enum_reconcile
Revises: 008_align_skills
@@ -22,139 +26,127 @@ Create Date: 2026-05-02
from __future__ import annotations
from typing import TYPE_CHECKING
from alembic import op
from sqlalchemy import text
if TYPE_CHECKING:
from sqlalchemy.sql.elements import TextClause
revision = "009_enum_reconcile"
down_revision = "008_align_skills"
branch_labels = None
depends_on = None
# Per-enum desired member set (matches the StrEnum `.value` lists in
# roboco/models/base.py + roboco/models/a2a.py + work_session.py).
_DESIRED: dict[str, tuple[str, ...]] = {
"agentrole": (
"system",
"ceo",
"product_owner",
"head_marketing",
"auditor",
"main_pm",
"cell_pm",
"developer",
"qa",
"documenter",
),
"team": (
"backend",
"frontend",
"ux_ui",
"fullstack",
"main_pm",
"board",
"marketing",
),
"taskstatus": (
"backlog",
"pending",
"claimed",
"in_progress",
"blocked",
"paused",
"verifying",
"needs_revision",
"awaiting_qa",
"awaiting_documentation",
"awaiting_pm_review",
"awaiting_ceo_approval",
"completed",
"cancelled",
"quarantined",
),
# Members the ORM defines that may not be in the alembic-declared enum.
# Every value here is a literal already used by the ORM today; adding
# them to the postgres enum is what unblocks the next-write path.
_DESIRED_ADDITIONS: dict[str, tuple[str, ...]] = {
"agentrole": ("system",),
"team": ("fullstack",),
"taskstatus": ("quarantined",),
}
# (enum_name, table_name, column_name) — used by the conditional rebuild.
_USAGES: tuple[tuple[str, str, str], ...] = (
("agentrole", "agents", "role"),
("team", "agents", "team"),
("team", "tasks", "team"),
("team", "projects", "assigned_cell"),
("taskstatus", "tasks", "status"),
)
def upgrade() -> None:
"""Add missing values; rebuild if uppercase drift is detected."""
"""Add missing values; rebuild any enum found with uppercase members."""
bind = op.get_bind()
# Step 1: detect drift. If any enum has uppercase members, the DB was
# bootstrapped via create_all and needs a full rebuild for ALL the
# enums we use. If everything is lowercase already, just ADD missing
# values one by one (the cheap path).
drifted = bool(
bind.execute(
_drift_query(),
).scalar()
)
# Step 1: find every enum that has at least one uppercase member.
drifted = [
row[0]
for row in bind.execute(
text(
"""
SELECT DISTINCT t.typname
FROM pg_enum e
JOIN pg_type t ON e.enumtypid = t.oid
WHERE e.enumlabel ~ '[A-Z]'
ORDER BY t.typname
"""
)
)
]
if not drifted:
# Cheap path: ADD missing values per enum.
for enum_name, members in _DESIRED.items():
for value in members:
# Cheap path: add the missing values for known additions.
for enum_name, additions in _DESIRED_ADDITIONS.items():
for value in additions:
op.execute(f"ALTER TYPE {enum_name} ADD VALUE IF NOT EXISTS '{value}'")
return
# Drift path: rebuild the affected enums. Each rebuild is a
# rename-old / create-new / alter-column / drop-old sequence. We
# USING lower(col::text)::new_enum to convert uppercase data.
for enum_name, members in _DESIRED.items():
# Step 2: rebuild each drifted enum.
# For each enum:
# - read current members
# - construct new member list as lowercase(current) plus desired_additions
# - rename old, create new, ALTER every (table, column) using it,
# drop old.
for enum_name in drifted:
current = [
row[0]
for row in bind.execute(
text(
"""
SELECT e.enumlabel
FROM pg_enum e
JOIN pg_type t ON e.enumtypid = t.oid
WHERE t.typname = :name
ORDER BY e.enumsortorder
"""
),
{"name": enum_name},
)
]
new_members: list[str] = []
seen: set[str] = set()
for member in current:
lowered = member.lower()
if lowered not in seen:
new_members.append(lowered)
seen.add(lowered)
for addition in _DESIRED_ADDITIONS.get(enum_name, ()):
if addition not in seen:
new_members.append(addition)
seen.add(addition)
usages = [
(row[0], row[1])
for row in bind.execute(
text(
"""
SELECT n.nspname || '.' || c.relname AS table_qualified, a.attname
FROM pg_attribute a
JOIN pg_class c ON a.attrelid = c.oid
JOIN pg_namespace n ON c.relnamespace = n.oid
JOIN pg_type t ON a.atttypid = t.oid
WHERE t.typname = :name
AND a.attnum > 0
AND NOT a.attisdropped
AND c.relkind = 'r'
"""
),
{"name": enum_name},
)
]
op.execute(f"ALTER TYPE {enum_name} RENAME TO {enum_name}_old")
members_sql = ", ".join(f"'{v}'" for v in members)
members_sql = ", ".join(f"'{v}'" for v in new_members)
op.execute(f"CREATE TYPE {enum_name} AS ENUM ({members_sql})")
for enum_name, table, column in _USAGES:
op.execute(
f"ALTER TABLE {table} "
f"ALTER COLUMN {column} TYPE {enum_name} "
f"USING lower({column}::text)::{enum_name}"
)
for table_qualified, column in usages:
op.execute(
f"ALTER TABLE {table_qualified} "
f"ALTER COLUMN {column} TYPE {enum_name} "
f"USING lower({column}::text)::{enum_name}"
)
for enum_name in _DESIRED:
op.execute(f"DROP TYPE {enum_name}_old")
def downgrade() -> None:
"""Remove the values added by upgrade.
"""Intentional no-op.
Postgres has no DROP VALUE primitive; the only way to remove an enum
member is the rebuild dance from upgrade. For the no-drift path we
simply leave the added values in place — they cause no harm and
backing them out would be a destructive rebuild on healthy data. For
the drift path we cannot recover the prior uppercase values without
losing referential integrity, so we likewise leave the rebuild in
place. This downgrade is therefore intentionally a no-op.
Postgres has no DROP VALUE primitive removing an enum member would
require the same rebuild dance from upgrade. We can't recover the
prior uppercase shape after the data was already lowercased without
losing referential integrity, so we leave the reconciled state in
place.
"""
return None
def _drift_query() -> TextClause:
"""SELECT true iff any tracked enum has uppercase members."""
enum_list = ", ".join(f"'{name}'" for name in _DESIRED)
return text(
f"""
SELECT EXISTS (
SELECT 1
FROM pg_enum e
JOIN pg_type t ON e.enumtypid = t.oid
WHERE t.typname IN ({enum_list})
AND e.enumlabel ~ '[A-Z]'
)
"""
)