Table Of Contents

Previous topic

Bonnie Client API

Next topic

Calendaring in (Large) Environments with Discretionary Access Control

This Page

Application Logic and Database Design Considerations

Database Technology

NoSQL storing key value pairs, trick is to store as many keys as is necessary to get to the related value.

Relationships problematic. Possibly for data only, use ORM for relationships.

Object Relationship Manager

  • User

    An individual human being with physical presence (in the here and now, past and/or future).

  • Group

    A group of individual User objects.

  • Role

    A role attached to one or more User objects, functionally the inverse of a Group.

  • IMAP Folder

    • METADATA, including:
      • Unique ID (persistent)
      • Shared seen, important to Read/Unread status tracking,
    • ACL
    • Payload

Change Recording in ORM

Rather than recording the changes to objects explicitly, it is more effective to define the objects themselves as volatile objects, in such a way that updates to them imply a changelog record be created.

In summary, it is better to reduce the processor’s workload and code base from:

digraph event_notification {
        rankdir = LR;
        splines = true;
        overlab = prism;

        edge [color=gray50, fontname=Calibri, fontsize=11]
        node [shape=record, fontname=Calibri, fontsize=11]

        "object record";
        "object changelog record";

        "change" -> "processor";

        "processor" -> "object record" [label="updates"];
        "processor" -> "object changelog record" [label="creates"];



digraph event_notification {
        rankdir = LR;
        splines = true;
        overlab = prism;

        edge [color=gray50, fontname=Calibri, fontsize=11]
        node [shape=record, fontname=Calibri, fontsize=11]

        "object record";
        "object changelog record";

        "change" -> "processor";

        "processor" -> "object record" [label="updates"];
        "object record" -> "object changelog record" [label="record change implies creation of"];


The basis of this implementation is simple (in Python):

  • Declare a class for individual change records,
  • Declare a declarative class inherited by individual object table definitions,
  • Use the declarative class in addition to the declarative base class in the definition of the individual object table class,
  • Use __setattr__() to catch any changes to existing objects,
  • Process the current value of the object that is being changed, if any,
  • Record the object name, object id, current value (if any), and the value it is about to change to,
  • Insert the new changelog record in to the current transaction.
class Change(DeclarativeBase):
        This object represents an entry of a ChangeLog-type table.

    __tablename__ = 'changes'

    # Depending on the size of these tables, adjust the following
    # table name that updates automatically. This example does so monthly.
    # Note that this only affects every initialization of the engine.

    #__tablename__ = eval(
    #        '"changes_%s"' % (
    #                datetime.strftime(datetime.utcnow(), "%Y_%m")
    #            )
    #    )

    id = Column(Integer, primary_key=True)
    object_name = Column(String(64))
    object_id = Column(Integer)
    value_from = Column(Text)
    value_to = Column(Text)

    # Add:
    #   - who
    #   - when

class ChangeRecordDeclarativeBase(object):
        This abstract base class must be used for DeclarativeBase class
        definitions for which we want to record changes to individual entries.

    def __setattr__(self, key, value):
        current_value = None

        if hasattr(self, key):
            current_value = getattr(self, key)

        if not current_value == None and not current_value == value:
            # Record the change
            change = Change()
            change.object_name = self.__class__.__name__
            change.object_id =
            change.value_from = current_value
            change.value_to = value


        DeclarativeBase.__setattr__(self, key, value)

class Folder(ChangeRecordDeclarativeBase, DeclarativeBase):
        An IMAP folder.

    __tablename__ = 'folders'

    id = Column(Integer, primary_key=True)
    path = Column(Text, nullable=False)
    uniqueid = Column(String(16))
    created = Column(DateTime)

    _metadata = relation("FolderMetadata")
    _acl = relation("FolderACL")

Users are Volatile and Groups do not Exist

Usernames as issued by Cyrus IMAP 2.5 notifications are volatile, in that the same physical human being ( could change email addresses for any of many unrelated causes (

It is therefore mandatory to:

  • resolve IMAP login usernames to canonified IMAP login usernames,

    User jdoe2 could in fact be the same physical human being as and

  • relate canonified IMAP login usernames to persistent user attribute values,

  • relate mail folder names, paths and URIs in personal namespaces to persistent user attribute values,

  • resolve IMAP ACE subject entries to their persistent attribute values, for both users and groups,

  • store membership information about groups at the time of an event,

  • store roles attached to users.

This needs to happen in a timely fashion, for intermediate changes to the authoritative, canonical user and group information database, in the period of time between the event notification and the collection of information, could invalidate the permanent record.

digraph bonnie_user {
        splines = true;
        overlap = prism;

        edge [color=gray50, fontname=Calibri, fontsize=11]
        node [shape=record, fontname=Calibri, fontsize=11]

        subgraph cluster_dbuser {
                label = "User (Database)";
                dbuser_id [label="ID", color=blue, fontcolor=blue];
                dbuser_uniqueid [label="UniqueID", color=blue, fontcolor=blue];

        subgraph cluster_ldapuser {
                label = "User (LDAP)";
                ldapuser_dn [label="Entry DN", color=blue, fontcolor=blue];
                ldapuser_uniqueid [label="UniqueID", color=blue, fontcolor=blue];

        subgraph cluster_dbdata {
                label = "Database Data";
                dbcolumn_dbuser_id [label="UserID", color=blue, fontcolor=blue];

        dbuser_id -> dbuser_uniqueid [label="resolves to"];
        dbuser_id -> dbcolumn_dbuser_id [label="FOREIGN KEY",dir=back];
        dbuser_uniqueid -> ldapuser_uniqueid [label="equals"];