def test_is_nullable():
assert json_schema.is_nullable({
'type': ['array', 'null'],
'items': {
'type': ['boolean']
}
})
assert json_schema.is_nullable({'type': ['integer', 'null']})
assert not json_schema.is_nullable({'type': ['string']})
assert not json_schema.is_nullable({})
def _denest_schema(table_path,
table_json_schema,
key_prop_schemas,
subtables,
level=-1):
new_properties = {}
for prop, item_json_schema in _denest_schema__singular_schemas(
table_json_schema):
if json_schema.is_object(item_json_schema):
_denest_schema_helper(table_path + (prop, ), (prop, ),
item_json_schema,
json_schema.is_nullable(item_json_schema),
new_properties, key_prop_schemas, subtables,
level)
elif json_schema.is_iterable(item_json_schema):
_create_subtable(table_path + (prop, ), item_json_schema,
key_prop_schemas, subtables, level + 1)
elif json_schema.is_literal(item_json_schema):
if (prop, ) in new_properties:
new_properties[(prop, )]['anyOf'].append(item_json_schema)
else:
new_properties[(prop, )] = {'anyOf': [item_json_schema]}
table_json_schema['properties'] = new_properties
def test_complex_objects__logical_statements():
every_type = {
'type':
['null', 'integer', 'number', 'boolean', 'string', 'array', 'object'],
'items': {
'type': 'integer'
},
'format':
'date-time',
'properties': {
'a': {
'type': 'integer'
},
'b': {
'type': 'number'
},
'c': {
'type': 'boolean'
}
}
}
assert json_schema.is_iterable(every_type)
assert json_schema.is_nullable(every_type)
assert json_schema.is_iterable(every_type)
assert json_schema.is_object(every_type)
def test__anyOf__schema__implicit_any_of():
denested = error_check_denest(
{
'properties': {
'every_type': {
'type': ['integer', 'null', 'number', 'boolean', 'string', 'array', 'object'],
'items': {'type': 'integer'},
'format': 'date-time',
'properties': {
'i': {'type': 'integer'},
'n': {'type': 'number'},
'b': {'type': 'boolean'}
}
}
}
},
[],
[])
assert 2 == len(denested)
table_batch = _get_table_batch_with_path(denested, tuple())
denested_props = table_batch['streamed_schema']['schema']['properties']
assert 4 == len(denested_props)
anyof_schemas = denested_props[('every_type',)]['anyOf']
assert 4 == len(anyof_schemas)
assert 4 == len([x for x in anyof_schemas if json_schema.is_literal(x)])
assert 4 == len([x for x in anyof_schemas if json_schema.is_nullable(x)])
assert 1 == len([x for x in anyof_schemas if json_schema.is_datetime(x)])
def test_simplify__allOf__nullable():
assert json_schema.is_nullable(
json_schema.simplify(
{'allOf': [{
'type': ['integer']
}, {
'type': ['string', 'null']
}]}))
def denest_schema_helper(self, table_name, table_json_schema, not_null,
top_level_schema, current_path, key_prop_schemas,
subtables, level):
for prop, item_json_schema in table_json_schema['properties'].items():
next_path = current_path + self.NESTED_SEPARATOR + prop
if json_schema.is_object(item_json_schema):
self.denest_schema_helper(table_name, item_json_schema,
not_null, top_level_schema,
next_path, key_prop_schemas,
subtables, level)
elif json_schema.is_iterable(item_json_schema):
self.create_subtable(table_name + self.NESTED_SEPARATOR + prop,
item_json_schema, key_prop_schemas,
subtables, level + 1)
else:
if not_null and json_schema.is_nullable(item_json_schema):
item_json_schema['type'].remove('null')
elif not json_schema.is_nullable(item_json_schema):
item_json_schema['type'].append('null')
top_level_schema[next_path] = item_json_schema
def _literal_only_schema(schema):
ret_types = json_schema.get_type(schema)
if json_schema.is_object(schema):
ret_types.remove(json_schema.OBJECT)
if json_schema.is_iterable(schema):
ret_types.remove(json_schema.ARRAY)
if json_schema.is_nullable(schema):
ret_types.remove(json_schema.NULL)
ret_schemas = []
for t in ret_types:
s = deepcopy(schema)
s['type'] = [t]
if json_schema.is_nullable(schema):
s = json_schema.make_nullable(s)
ret_schemas.append(s)
return {'anyOf': ret_schemas}
def _denest_schema_helper(table_path, table_json_schema, nullable,
top_level_schema, key_prop_schemas, subtables,
level):
for prop, item_json_schema in table_json_schema['properties'].items():
if json_schema.is_object(item_json_schema):
_denest_schema_helper(table_path + (prop, ), item_json_schema,
nullable, top_level_schema, key_prop_schemas,
subtables, level)
if json_schema.is_iterable(item_json_schema):
_create_subtable(table_path + (prop, ), item_json_schema,
key_prop_schemas, subtables, level + 1)
if json_schema.is_literal(item_json_schema):
if nullable and not json_schema.is_nullable(item_json_schema):
item_json_schema['type'].append('null')
top_level_schema[table_path +
(prop, )] = _literal_only_schema(item_json_schema)
def merge_put_schemas(self, cur, table_schema, table_name, existing_schema,
new_schema):
new_properties = new_schema['properties']
existing_properties = existing_schema['schema']['properties']
for name, schema in new_properties.items():
## Mapping exists
if self.get_mapping(existing_schema, name, schema) is not None:
pass
## New column
elif name not in existing_properties:
existing_properties[name] = schema
self.add_column(cur, table_schema, table_name, name, schema)
## Existing column non-nullable, new column is nullable
elif not json_schema.is_nullable(existing_properties[name]) \
and json_schema.get_type(schema) \
== json_schema.get_type(json_schema.make_nullable(existing_properties[name])):
existing_properties[name] = json_schema.make_nullable(
existing_properties[name])
self.make_column_nullable(cur, table_schema, table_name, name)
## Existing column, types compatible
elif json_schema.to_sql(json_schema.make_nullable(schema)) \
== json_schema.to_sql(json_schema.make_nullable(existing_properties[name])):
pass
## Column type change
elif self.mapping_name(name, schema) not in existing_properties \
and self.mapping_name(name, existing_properties[name]) not in existing_properties:
self.split_column(cur, table_schema, table_name, name, schema,
existing_properties)
## Error
else:
raise PostgresError(
'Cannot handle column type change for: {}.{} columns {} and {}. Name collision likely.'
.format(table_schema, table_name, name,
self.mapping_name(name, schema)))
def test__anyOf__schema__stitch_date_times():
denested = error_check_denest(
{'properties': {
'a': {
"anyOf": [
{
"type": "string",
"format": "date-time"
},
{"type": ["string", "null"]}]}}},
[],
[])
table_batch = _get_table_batch_with_path(denested, tuple())
anyof_schemas = table_batch['streamed_schema']['schema']['properties'][('a',)]['anyOf']
assert 2 == len(anyof_schemas)
assert 2 == len([x for x in anyof_schemas if json_schema.is_literal(x)])
assert 2 == len([x for x in anyof_schemas if json_schema.is_nullable(x)])
assert 1 == len([x for x in anyof_schemas if json_schema.is_datetime(x)])
def add_column(self, cur, table_schema, table_name, column_name,
column_schema):
data_type = json_schema.to_sql(column_schema)
if not json_schema.is_nullable(column_schema) \
and not self.is_table_empty(cur, table_schema, table_name):
self.logger.warning(
'Forcing new column `{}.{}.{}` to be nullable due to table not empty.'
.format(table_schema, table_name, column_name))
data_type = json_schema.to_sql(
json_schema.make_nullable(column_schema))
to_execute = sql.SQL('ALTER TABLE {table_schema}.{table_name} ' +
'ADD COLUMN {column_name} {data_type};').format(
table_schema=sql.Identifier(table_schema),
table_name=sql.Identifier(table_name),
column_name=sql.Identifier(column_name),
data_type=sql.SQL(data_type))
cur.execute(to_execute)
def _denest_schema(table_path,
table_json_schema,
key_prop_schemas,
subtables,
level=-1):
new_properties = {}
for prop, item_json_schema in table_json_schema['properties'].items():
if json_schema.is_object(item_json_schema):
_denest_schema_helper(table_path + (prop, ), item_json_schema,
json_schema.is_nullable(item_json_schema),
new_properties, key_prop_schemas, subtables,
level)
if json_schema.is_iterable(item_json_schema):
_create_subtable(table_path + (prop, ), item_json_schema,
key_prop_schemas, subtables, level + 1)
if json_schema.is_literal(item_json_schema):
new_properties[(prop, )] = _literal_only_schema(item_json_schema)
table_json_schema['properties'] = new_properties
def upsert_table_helper(self,
connection,
schema,
metadata,
log_schema_changes=True):
"""
Upserts the `schema` to remote by:
- creating table if necessary
- adding columns
- adding column mappings
- migrating data from old columns to new, etc.
:param connection: remote connection, type left to be determined by implementing class
:param schema: TABLE_SCHEMA(local)
:param metadata: additional information necessary for downstream operations,
:param log_schema_changes: defaults to True, set to false to disable logging of table level schema changes
:return: TABLE_SCHEMA(remote)
"""
table_path = schema['path']
_metadata = deepcopy(metadata)
_metadata['schema_version'] = CURRENT_SCHEMA_VERSION
table_name = self.add_table_mapping(connection, table_path, _metadata)
existing_schema = self._get_table_schema(connection, table_path,
table_name)
if existing_schema is None:
self.add_table(connection, table_name, _metadata)
existing_schema = self._get_table_schema(connection, table_path,
table_name)
self.add_key_properties(connection, table_name,
schema.get('key_properties', None))
## Only process columns which have single, nullable, types
single_type_columns = []
for column_name__or__path, column_schema in schema['schema'][
'properties'].items():
column_path = column_name__or__path
if isinstance(column_name__or__path, str):
column_path = (column_name__or__path, )
single_type_column_schema = deepcopy(column_schema)
column_types = json_schema.get_type(single_type_column_schema)
make_nullable = json_schema.is_nullable(column_schema)
for type in column_types:
if type == json_schema.NULL:
continue
single_type_column_schema['type'] = [type]
if make_nullable:
single_type_columns.append(
(column_path,
json_schema.make_nullable(single_type_column_schema)))
else:
single_type_columns.append(
(column_path, deepcopy(single_type_column_schema)))
## Process new columns against existing
raw_mappings = existing_schema.get('mappings', {})
mappings = []
for to, m in raw_mappings.items():
mapping = json_schema.simple_type(m)
mapping['from'] = tuple(m['from'])
mapping['to'] = to
mappings.append(mapping)
table_empty = self.is_table_empty(connection, table_name)
for column_path, column_schema in single_type_columns:
upsert_table_helper__start__column = time.monotonic()
canonicalized_column_name = self._canonicalize_column_identifier(
column_path, column_schema, mappings)
nullable_column_schema = json_schema.make_nullable(column_schema)
def log_message(msg):
if log_schema_changes:
self.LOGGER.info(
'Table Schema Change [`{}`.`{}`:`{}`] {} (took {} millis)'
.format(
table_name, column_path, canonicalized_column_name,
msg,
_duration_millis(
upsert_table_helper__start__column)))
## NEW COLUMN
if not column_path in [m['from'] for m in mappings]:
upsert_table_helper__column = "New column"
### NON EMPTY TABLE
if not table_empty:
upsert_table_helper__column += ", non empty table"
self.LOGGER.warning(
'NOT EMPTY: Forcing new column `{}` in table `{}` to be nullable due to table not empty.'
.format(column_path, table_name))
column_schema = nullable_column_schema
self.add_column(connection, table_name,
canonicalized_column_name, column_schema)
self.add_column_mapping(connection, table_name, column_path,
canonicalized_column_name,
column_schema)
mapping = json_schema.simple_type(column_schema)
mapping['from'] = column_path
mapping['to'] = canonicalized_column_name
mappings.append(mapping)
log_message(upsert_table_helper__column)
continue
## EXISTING COLUMNS
### SCHEMAS MATCH
if [
True for m in mappings if m['from'] == column_path
and self.json_schema_to_sql_type(
m) == self.json_schema_to_sql_type(column_schema)
]:
continue
### NULLABLE SCHEMAS MATCH
### New column _is not_ nullable, existing column _is_
if [
True for m in mappings if m['from'] == column_path
and self.json_schema_to_sql_type(m) ==
self.json_schema_to_sql_type(nullable_column_schema)
]:
continue
### NULL COMPATIBILITY
### New column _is_ nullable, existing column is _not_
non_null_original_column = [
m for m in mappings
if m['from'] == column_path and json_schema.shorthand(m) ==
json_schema.shorthand(column_schema)
]
if non_null_original_column:
## MAKE NULLABLE
self.make_column_nullable(connection, table_name,
canonicalized_column_name)
self.drop_column_mapping(connection, table_name,
canonicalized_column_name)
self.add_column_mapping(connection, table_name, column_path,
canonicalized_column_name,
nullable_column_schema)
mappings = [
m for m in mappings
if not (m['from'] == column_path and json_schema.shorthand(
m) == json_schema.shorthand(column_schema))
]
mapping = json_schema.simple_type(nullable_column_schema)
mapping['from'] = column_path
mapping['to'] = canonicalized_column_name
mappings.append(mapping)
log_message(
"Made existing column nullable. New column is nullable, existing column is not"
)
continue
### FIRST MULTI TYPE
### New column matches existing column path, but the types are incompatible
duplicate_paths = [m for m in mappings if m['from'] == column_path]
if 1 == len(duplicate_paths):
existing_mapping = duplicate_paths[0]
existing_column_name = existing_mapping['to']
if existing_column_name:
self.drop_column_mapping(connection, table_name,
existing_column_name)
## Update existing properties
mappings = [m for m in mappings if m['from'] != column_path]
mapping = json_schema.simple_type(nullable_column_schema)
mapping['from'] = column_path
mapping['to'] = canonicalized_column_name
mappings.append(mapping)
existing_column_new_normalized_name = self._canonicalize_column_identifier(
column_path, existing_mapping, mappings)
mapping = json_schema.simple_type(
json_schema.make_nullable(existing_mapping))
mapping['from'] = column_path
mapping['to'] = existing_column_new_normalized_name
mappings.append(mapping)
## Add new columns
### NOTE: all migrated columns will be nullable and remain that way
#### Table Metadata
self.add_column_mapping(
connection, table_name, column_path,
existing_column_new_normalized_name,
json_schema.make_nullable(existing_mapping))
self.add_column_mapping(connection, table_name, column_path,
canonicalized_column_name,
nullable_column_schema)
#### Columns
self.add_column(connection, table_name,
existing_column_new_normalized_name,
json_schema.make_nullable(existing_mapping))
self.add_column(connection, table_name,
canonicalized_column_name,
nullable_column_schema)
## Migrate existing data
self.migrate_column(connection, table_name,
existing_mapping['to'],
existing_column_new_normalized_name)
## Drop existing column
self.drop_column(connection, table_name,
existing_mapping['to'])
upsert_table_helper__column = "Splitting `{}` into `{}` and `{}`. New column matches existing column path, but the types are incompatible.".format(
existing_column_name, existing_column_new_normalized_name,
canonicalized_column_name)
## REST MULTI TYPE
elif 1 < len(duplicate_paths):
## Add new column
self.add_column_mapping(connection, table_name, column_path,
canonicalized_column_name,
nullable_column_schema)
self.add_column(connection, table_name,
canonicalized_column_name,
nullable_column_schema)
mapping = json_schema.simple_type(nullable_column_schema)
mapping['from'] = column_path
mapping['to'] = canonicalized_column_name
mappings.append(mapping)
upsert_table_helper__column = "Adding new column to split column `{}`. New column matches existing column's path, but no types were compatible.".format(
column_path)
## UNKNOWN
else:
raise Exception(
'UNKNOWN: Cannot handle merging column `{}` (canonicalized as: `{}`) in table `{}`.'
.format(column_path, canonicalized_column_name,
table_name))
log_message(upsert_table_helper__column)
return self._get_table_schema(connection, table_path, table_name)
def upsert_table_helper(self, connection, schema, metadata):
"""
Upserts the `schema` to remote by:
- creating table if necessary
- adding columns
- adding column mappings
- migrating data from old columns to new, etc.
:param connection: remote connection, type left to be determined by implementing class
:param schema: TABLE_SCHEMA(local)
:param metadata: additional information necessary for downstream operations
:return: TABLE_SCHEMA(remote)
"""
table_path = schema['path']
table_name = self.add_table_mapping(connection, table_path, metadata)
existing_schema = self.get_table_schema(connection, table_path,
table_name)
if existing_schema is None:
self.add_table(connection, table_name, metadata)
existing_schema = self.get_table_schema(connection, table_path,
table_name)
self.add_key_properties(connection, table_name,
schema.get('key_properties', None))
## Only process columns which have single, nullable, types
single_type_columns = []
for column_name__or__path, column_schema in schema['schema'][
'properties'].items():
column_path = column_name__or__path
if isinstance(column_name__or__path, str):
column_path = (column_name__or__path, )
single_type_column_schema = deepcopy(column_schema)
column_types = json_schema.get_type(single_type_column_schema)
make_nullable = json_schema.is_nullable(column_schema)
for type in column_types:
if type == json_schema.NULL:
continue
single_type_column_schema['type'] = [type]
if make_nullable:
single_type_columns.append(
(column_path,
json_schema.make_nullable(single_type_column_schema)))
else:
single_type_columns.append(
(column_path, single_type_column_schema))
## Process new columns against existing
raw_mappings = existing_schema.get('mappings', {})
mappings = []
for to, m in raw_mappings.items():
mappings.append({
'from': tuple(m['from']),
'to': to,
'type': m['type']
})
table_empty = self.is_table_empty(connection, table_name)
for column_path, column_schema in single_type_columns:
canonicalized_column_name = self._canonicalize_column_identifier(
column_path, column_schema, mappings)
nullable_column_schema = json_schema.make_nullable(column_schema)
## NEW COLUMN
if not column_path in [m['from'] for m in mappings]:
### NON EMPTY TABLE
if not table_empty:
self.LOGGER.warning(
'NOT EMPTY: Forcing new column `{}` in table `{}` to be nullable due to table not empty.'
.format(column_path, table_name))
column_schema = nullable_column_schema
self.add_column(connection, table_name,
canonicalized_column_name, column_schema)
self.add_column_mapping(connection, table_name, column_path,
canonicalized_column_name,
column_schema)
mappings.append({
'from': column_path,
'to': canonicalized_column_name,
'type': json_schema.get_type(column_schema)
})
continue
## EXISTING COLUMNS
### SCHEMAS MATCH
if [
True for m in mappings if m['from'] == column_path and
json_schema.to_sql(m) == json_schema.to_sql(column_schema)
]:
continue
### NULLABLE SCHEMAS MATCH
### New column _is not_ nullable, existing column _is_
if [
True for m in mappings
if m['from'] == column_path and json_schema.to_sql(m) ==
json_schema.to_sql(nullable_column_schema)
]:
continue
### NULL COMPATIBILITY
### New column _is_ nullable, existing column is _not_
non_null_original_column = [
m for m in mappings
if m['from'] == column_path and json_schema.sql_shorthand(m) ==
json_schema.sql_shorthand(column_schema)
]
if non_null_original_column:
## MAKE NULLABLE
self.make_column_nullable(connection, table_name,
canonicalized_column_name)
self.drop_column_mapping(connection, table_name,
canonicalized_column_name)
self.add_column_mapping(connection, table_name, column_path,
canonicalized_column_name,
nullable_column_schema)
mappings = [
m for m in mappings if not (
m['from'] == column_path and json_schema.sql_shorthand(
m) == json_schema.sql_shorthand(column_schema))
]
mappings.append({
'from':
column_path,
'to':
canonicalized_column_name,
'type':
json_schema.get_type(nullable_column_schema)
})
continue
### FIRST MULTI TYPE
### New column matches existing column path, but the types are incompatible
duplicate_paths = [m for m in mappings if m['from'] == column_path]
if 1 == len(duplicate_paths):
existing_mapping = duplicate_paths[0]
existing_column_name = existing_mapping['to']
if existing_column_name:
self.drop_column_mapping(connection, table_name,
existing_column_name)
## Update existing properties
mappings = [m for m in mappings if m['from'] != column_path]
mappings.append({
'from':
column_path,
'to':
canonicalized_column_name,
'type':
json_schema.get_type(nullable_column_schema)
})
existing_column_new_normalized_name = self._canonicalize_column_identifier(
column_path, existing_mapping, mappings)
mappings.append({
'from':
column_path,
'to':
existing_column_new_normalized_name,
'type':
json_schema.get_type(
json_schema.make_nullable(existing_mapping))
})
## Add new columns
### NOTE: all migrated columns will be nullable and remain that way
#### Table Metadata
self.add_column_mapping(
connection, table_name, column_path,
existing_column_new_normalized_name,
json_schema.make_nullable(existing_mapping))
self.add_column_mapping(connection, table_name, column_path,
canonicalized_column_name,
nullable_column_schema)
#### Columns
self.add_column(connection, table_name,
existing_column_new_normalized_name,
json_schema.make_nullable(existing_mapping))
self.add_column(connection, table_name,
canonicalized_column_name,
nullable_column_schema)
## Migrate existing data
self.migrate_column(connection, table_name,
existing_mapping['to'],
existing_column_new_normalized_name)
## Drop existing column
self.drop_column(connection, table_name,
existing_mapping['to'])
## REST MULTI TYPE
elif 1 < len(duplicate_paths):
## Add new column
self.add_column_mapping(connection, table_name, column_path,
canonicalized_column_name,
nullable_column_schema)
self.add_column(connection, table_name,
canonicalized_column_name,
nullable_column_schema)
mappings.append({
'from':
column_path,
'to':
canonicalized_column_name,
'type':
json_schema.get_type(nullable_column_schema)
})
## UNKNOWN
else:
raise Exception(
'UNKNOWN: Cannot handle merging column `{}` (canonicalized as: `{}`) in table `{}`.'
.format(column_path, canonicalized_column_name,
table_name))
return self.get_table_schema(connection, table_path, table_name)
