Open-source live-chat, email support, omni-channel desk. An alternative to Intercom, Zendesk, Salesforce Service Cloud etc. π₯π¬
Ruby actioncablechat-widgetconversationcustomer-supportdashboarddesigndockerdocker-imageherokuintercomjavascriptlivechatopensourcerailsrubysassvuejswhatsappzendesk
class AutoAssignment::AgentAssignmentService
# Allowed agent ids: array
# This is the list of agents from which an agent can be assigned to this conversation
# examples: Agents with assignment capacity, Agents who are members of a team etc
pattr_initialize [:conversation!, :allowed_agent_ids!]
def find_assignee
round_robin_manage_service.available_agent(allowed_agent_ids: allowed_online_agent_ids)
end
# Locks the row to serialize with concurrent assignment writers, then sets the assignee
# on the in-memory conversation without saving. Called from the conversation's own
# before_save so status and assignee commit as one change-set and both stay visible to
# the after_commit callbacks. Returns the new assignee, or nil when nothing changed.
def assign_under_lock
locked = Conversation.lock.find_by(id: conversation.id)
return unless locked
discard_already_applied_status_change(locked)
return unless reassignment_still_needed?(locked)
new_assignee = find_assignee
return unless new_assignee
conversation.assignee_id = locked.assignee_id
conversation.clear_attribute_changes([:assignee_id])
conversation.assignee = new_assignee
end
def perform
# Standalone assignment with its own save (create path). Write through conversation
# itself so its already-registered after_commit callbacks actually fire.
Conversation.transaction do
conversation.save if assign_under_lock
end
end
private
# A concurrent writer may have committed the same status transition while we waited for
# the lock; keeping our stale copy dirty would announce it a second time (duplicate
# conversation.opened events, activities and reporting rows), so drop it and let the
# save carry only changes that are genuinely ours.
def discard_already_applied_status_change(locked_conversation)
return unless conversation.will_save_change_to_status? && conversation.status == locked_conversation.status
conversation.clear_attribute_changes([:status])
end
# Fields the in-flight save is writing commit at their pending value (e.g. bot_handoff!
# clears the agent bot in the same save that opens), so the locked row only decides
# fields this save leaves untouched β the ones a concurrent writer would win.
def reassignment_still_needed?(locked_conversation)
bot_source = conversation.will_save_change_to_assignee_agent_bot_id? ? conversation : locked_conversation
return false if bot_source.assignee_agent_bot_id.present?
assignee = conversation.will_save_change_to_assignee_id? ? conversation.assignee : locked_conversation.assignee
assignee.blank? || locked_conversation.inbox.members.exclude?(assignee)
end
def online_agent_ids
online_agents = OnlineStatusTracker.get_available_users(conversation.account_id)
online_agents.select { |_key, value| value.eql?('online') }.keys if online_agents.present?
end
def allowed_online_agent_ids
# We want to perform roundrobin only over online agents
# Hence taking an intersection of online agents and allowed member ids
# the online user ids are string, since its from redis, allowed member ids are integer, since its from active record
@allowed_online_agent_ids ||= online_agent_ids & allowed_agent_ids&.map(&:to_s)
end
def round_robin_manage_service
@round_robin_manage_service ||= AutoAssignment::InboxRoundRobinService.new(inbox: conversation.inbox)
end
def round_robin_key
format(::Redis::Alfred::ROUND_ROBIN_AGENTS, inbox_id: conversation.inbox_id)
end
end