Federation: Efficiency Improvements

• don't send superfluous activities (i.e. update does not affect receiver)
• in receiver don't spend longer than necessary finding out nothing needs to be done

Related issues

• #249 (closed) Activity Queue (avoids sending duplicate updates)
• #228 saves loop execution (searching for object_model) on receiver methods

changed time estimate to 2h

mentioned in issue management/core-team#3

mentioned in issue #249 (closed)

changed the description

changed the description

changed the description

changed title from Federation: don't send more activities than necessary to Federation: Efficiency Improvements

marked the checklist item in receiver don't spend longer than necessary finding out nothing needs to be done as completed

The remaining task on this issue is that bandwidth usage could be reduced if the sender is able to know when an update will affect the receiver

For example if I do this:

cm = CircleMember.objects.create(user=distant_u, circle=local_circle)
cm.save()
cm.save()

it will result in a two Update activities, one for each cm.save()

Alternatively the sender could parse through the content of sent activity(s) to check if content has already been delivered (#249 (closed) ) before sending. This would be O(N) to the number of activities which are parsed, although it could be limited drastically by checking only those sent to the target inbox, within some time frame.

EDIT: I've thought of an O(1) solution which may be a lot simpler:

old_activity = get_most_recent_delivered_activity(target, activity_type)
new_activity = build_activity(object, target)

if no_new_changes(old_activity, new_activity):
    # no need to send an activity
    return

send_activity(new_activity)

no_new_changes would require parsing the activity body into JSON and comparing the old_activity.object state to the new_activity.object

I also considered storing past state of the model in the database but I don't think it makes sense because it implies nothing about the receiver

Each check would require parsing the activity body into JSON

You could use an upper tailored layer that will have target and resource field and an unique.

I don't follow?

The activity could for example be an update on circle owner, serialized roughly like so:

old_activity = {
    '@id': 'https://circle-id',
    'owner': {
        '@id': 'https://distant-user'
    }
}

I think I've thought of an improvement on my initial design, I've edited my comment above with an explanation

You mean that you save every things that change on a json? Because you can have two time the same activity, but to two or more different recipients.

Yes - https://git.startinblox.com/djangoldp-packages/djangoldp/blob/5db285727ff93b5519e3c30bba9c88421b261742/djangoldp/models.py#L254

marked the checklist item don't send superfluous activities (i.e. update does not affect receiver) as completed

mentioned in merge request !160 (merged)

mentioned in commit e94ea652

closed via merge request !160 (merged)