The oslo_db.sqlalchemy.update_match Module

exception oslo_db.sqlalchemy.update_match.CantUpdateException

Bases: exceptions.Exception

exception oslo_db.sqlalchemy.update_match.MultiRowsMatched

Bases: oslo_db.sqlalchemy.update_match.CantUpdateException

exception oslo_db.sqlalchemy.update_match.NoRowsMatched

Bases: oslo_db.sqlalchemy.update_match.CantUpdateException

oslo_db.sqlalchemy.update_match.manufacture_criteria(mapped, values)

Given a mapper/class and a namespace of values, produce a WHERE clause.

The class should be a mapped class and the entries in the dictionary correspond to mapped attribute names on the class.

A value may also be a tuple in which case that particular attribute will be compared to a tuple using IN. The scalar value or tuple can also contain None which translates to an IS NULL, that is properly joined with OR against an IN expression if appropriate.

Parameters:
  • cls – a mapped class, or actual Mapper object.
  • values – dictionary of values.
oslo_db.sqlalchemy.update_match.manufacture_entity_criteria(entity, include_only=None, exclude=None)

Given a mapped instance, produce a WHERE clause.

The attributes set upon the instance will be combined to produce a SQL expression using the mapped SQL expressions as the base of comparison.

Values on the instance may be set as tuples in which case the criteria will produce an IN clause. None is also acceptable as a scalar or tuple entry, which will produce IS NULL that is properly joined with an OR against an IN expression if appropriate.

Parameters:
  • entity – a mapped entity.
  • include_only – optional sequence of keys to limit which keys are included.
  • exclude – sequence of keys to exclude
oslo_db.sqlalchemy.update_match.manufacture_persistent_object(session, specimen, values=None, primary_key=None)

Make an ORM-mapped object persistent in a Session without SQL.

The persistent object is returned.

If a matching object is already present in the given session, the specimen is merged into it and the persistent object returned. Otherwise, the specimen itself is made persistent and is returned.

The object must contain a full primary key, or provide it via the values or primary_key parameters. The object is peristed to the Session in a “clean” state with no pending changes.

Parameters:
  • session – A Session object.
  • specimen – a mapped object which is typically transient.
  • values – a dictionary of values to be applied to the specimen, in addition to the state that’s already on it. The attributes will be set such that no history is created; the object remains clean.
  • primary_key – optional tuple-based primary key. This will also be applied to the instance if present.
oslo_db.sqlalchemy.update_match.update_on_match(query, specimen, surrogate_key, values=None, attempts=3, include_only=None, process_query=None, handle_failure=None)

Emit an UPDATE statement matching the given specimen.

E.g.:

with enginefacade.writer() as session:
    specimen = MyInstance(
        uuid='ccea54f',
        interface_id='ad33fea',
        vm_state='SOME_VM_STATE',
    )

    values = {
        'vm_state': 'SOME_NEW_VM_STATE'
    }

    base_query = model_query(
        context, models.Instance,
        project_only=True, session=session)

    hostname_query = model_query(
            context, models.Instance, session=session,
            read_deleted='no').
        filter(func.lower(models.Instance.hostname) == 'SOMEHOSTNAME')

    surrogate_key = ('uuid', )

    def process_query(query):
        return query.where(~exists(hostname_query))

    def handle_failure(query):
        try:
            instance = base_query.one()
        except NoResultFound:
            raise exception.InstanceNotFound(instance_id=instance_uuid)

        if session.query(hostname_query.exists()).scalar():
            raise exception.InstanceExists(
                name=values['hostname'].lower())

        # try again
        return False

    persistent_instance = base_query.update_on_match(
        specimen,
        surrogate_key,
        values=values,
        process_query=process_query,
        handle_failure=handle_failure
    )

The UPDATE statement is constructed against the given specimen using those values which are present to construct a WHERE clause. If the specimen contains additional values to be ignored, the include_only parameter may be passed which indicates a sequence of attributes to use when constructing the WHERE.

The UPDATE is performed against an ORM Query, which is created from the given Session, or alternatively by passing the `query parameter referring to an existing query.

Before the query is invoked, it is also passed through the callable sent as process_query, if present. This hook allows additional criteria to be added to the query after it is created but before invocation.

The function will then invoke the UPDATE statement and check for “success” one or more times, up to a maximum of that passed as attempts.

The initial check for “success” from the UPDATE statement is that the number of rows returned matches 1. If zero rows are matched, then the UPDATE statement is assumed to have “failed”, and the failure handling phase begins.

The failure handling phase involves invoking the given handle_failure function, if any. This handler can perform additional queries to attempt to figure out why the UPDATE didn’t match any rows. The handler, upon detection of the exact failure condition, should throw an exception to exit; if it doesn’t, it has the option of returning True or False, where False means the error was not handled, and True means that there was not in fact an error, and the function should return successfully.

If the failure handler is not present, or returns False after attempts number of attempts, then the function overall raises CantUpdateException. If the handler returns True, then the function returns with no error.

The return value of the function is a persistent version of the given specimen; this may be the specimen itself, if no matching object were already present in the session; otherwise, the existing object is returned, with the state of the specimen merged into it. The returned persistent object will have the given values populated into the object.

The object is is returned as “persistent”, meaning that it is associated with the given Session and has an identity key (that is, a real primary key value).

In order to produce this identity key, a strategy must be used to determine it as efficiently and safely as possible:

  1. If the given specimen already contained its primary key attributes fully populated, then these attributes were used as criteria in the UPDATE, so we have the primary key value; it is populated directly.
  2. If the target backend supports RETURNING, then when the update() query is performed with a RETURNING clause so that the matching primary key is returned atomically. This currently includes Postgresql, Oracle and others (notably not MySQL or SQLite).
  3. If the target backend is MySQL, and the given model uses a single-column, AUTO_INCREMENT integer primary key value (as is the case for Nova), MySQL’s recommended approach of making use of LAST_INSERT_ID(expr) is used to atomically acquire the matching primary key value within the scope of the UPDATE statement, then it fetched immediately following by using SELECT LAST_INSERT_ID(). http://dev.mysql.com/doc/refman/5.0/en/information- functions.html#function_last-insert-id
  4. Otherwise, for composite keys on MySQL or other backends such as SQLite, the row as UPDATED must be re-fetched in order to acquire the primary key value. The surrogate_key parameter is used for this in order to re-fetch the row; this is a column name with a known, unique value where the object can be fetched.
oslo_db.sqlalchemy.update_match.update_returning_pk(query, values, surrogate_key)

Perform an UPDATE, returning the primary key of the matched row.

The primary key is returned using a selection of strategies:

  • if the database supports RETURNING, RETURNING is used to retrieve the primary key values inline.
  • If the database is MySQL and the entity is mapped to a single integer primary key column, MySQL’s last_insert_id() function is used inline within the UPDATE and then upon a second SELECT to get the value.
  • Otherwise, a “refetch” strategy is used, where a given “surrogate” key value (typically a UUID column on the entity) is used to run a new SELECT against that UUID. This UUID is also placed into the UPDATE query to ensure the row matches.
Parameters:
  • query – a Query object with existing criterion, against a single entity.
  • values – a dictionary of values to be updated on the row.
  • surrogate_key – a tuple of (attrname, value), referring to a UNIQUE attribute that will also match the row. This attribute is used to retrieve the row via a SELECT when no optimized strategy exists.
Returns:

the primary key, returned as a tuple. Is only returned if rows matched is one. Otherwise, CantUpdateException is raised.