ref: handle old malformed ArrayField values in export / import

src/sentry/backup/imports.py

@@ -26,6 +26,7 @@
 )
 from sentry.backup.helpers import Filter, ImportFlags, Printer
 from sentry.backup.scopes import ImportScope
+from sentry.backup.services.import_export.impl import fixup_array_fields
 from sentry.backup.services.import_export.model import (
     RpcFilter,
     RpcImportError,
@@ -148,9 +149,7 @@ def _import(
     # wasteful - in the future, we should explore chunking strategies to enable a smaller memory
     # footprint when processing super large (>100MB) exports.
     content: bytes | str = (
-        decrypt_encrypted_tarball(src, decryptor)
-        if decryptor is not None
-        else src.read().decode("utf-8")
+        decrypt_encrypted_tarball(src, decryptor) if decryptor is not None else src.read()
     )
 
     if len(DELETED_MODELS) > 0 or len(DELETED_FIELDS) > 0:
@@ -170,6 +169,8 @@ def _import(
         # Return the content to byte form, as that is what the Django deserializer expects.
         content = orjson.dumps(content_as_json)
 
+    content = fixup_array_fields(content)
+
     filters = []
     if filter_by is not None:
         filters.append(filter_by)

src/sentry/backup/services/import_export/impl.py

@@ -3,10 +3,13 @@
 # in modules such as this one where hybrid cloud data models or service classes are
 # defined, because we want to reflect on type annotations and avoid forward references.
 
+import ast
 import logging
 import traceback
 
 import sentry_sdk
+from django.apps import apps
+from django.contrib.postgres.fields.array import ArrayField
 from django.core.exceptions import ValidationError as DjangoValidationError
 from django.core.serializers import deserialize, serialize
 from django.core.serializers.base import DeserializationError
@@ -51,6 +54,7 @@
 from sentry.users.models.user import User
 from sentry.users.models.userpermission import UserPermission
 from sentry.users.models.userrole import UserRoleUser
+from sentry.utils import json
 
 logger = logging.getLogger(__name__)
 
@@ -88,6 +92,28 @@ def get_existing_import_chunk(
     )
 
 
+def fixup_array_fields[T: (str, str | bytes)](json_data: T) -> T:
+    # preserve for 3 versions as per https://docs.sentry.io/concepts/migration/#version-support-window
+    # so probably 2025.09 this can go away?
+    try:
+        contents = json.loads(json_data)
+    except Exception:  # let the actual import/export produce a better message
+        return json_data
+
+    for dct in contents:
+        model = apps.get_model(dct["model"])
+        for k, v in dct["fields"].items():
+            if isinstance(model._meta.get_field(k), ArrayField) and isinstance(v, str):
+                try:
+                    json.loads(v)
+                except Exception:
+                    # old ArrayField: value was not properly encoded as json
+                    dct["fields"][k] = json.dumps(ast.literal_eval(v))
+                else:
+                    pass
+    return json.dumps(contents)
+
+
 class UniversalImportExportService(ImportExportService):
     """
     This implementation is universal regardless of which mode (CONTROL, REGION, or MONOLITH) it is
@@ -208,6 +234,9 @@ def import_by_model(
                 min_inserted_pk: int | None = None
                 max_inserted_pk: int | None = None
                 last_seen_ordinal = min_ordinal - 1
+
+                json_data = fixup_array_fields(json_data)
+
                 for deserialized_object in deserialize(
                     "json", json_data, use_natural_keys=False, ignorenonexistent=True
                 ):

tests/sentry/backup/test_rpc.py

@@ -12,7 +12,7 @@
 from sentry.backup.dependencies import NormalizedModelName, get_model_name
 from sentry.backup.helpers import ImportFlags
 from sentry.backup.services.import_export import import_export_service
-from sentry.backup.services.import_export.impl import get_existing_import_chunk
+from sentry.backup.services.import_export.impl import fixup_array_fields, get_existing_import_chunk
 from sentry.backup.services.import_export.model import (
     RpcExportError,
     RpcExportErrorKind,
@@ -501,3 +501,9 @@ def test_bad_unspecified_scope(self):
 
         assert isinstance(result, RpcExportError)
         assert result.get_kind() == RpcExportErrorKind.UnspecifiedScope
+
+
+def test_fixup_array_fields() -> None:
+    before = '[{"model":"sentry.dashboardwidgetquery","fields":{"aggregates":"[\'a\',\'b\']"}}]'
+    expect = '[{"model":"sentry.dashboardwidgetquery","fields":{"aggregates":"[\\"a\\",\\"b\\"]"}}]'
+    assert fixup_array_fields(before) == expect