SHARE

Slides

• SHARE
• Slides
• Problem
• Key Requirements
• Solution
• SHARE Node
• Conceptual Data Model and Flow
• SHARE Data Update
• Detailed Data Model
• Formulas
• Data Resynchronization Protocol
• Detailed Data Resynchronization
• Web Client
• Provider Data Import / Export
• Actors
• TBD

Problem

• task/incident related hierarchical reporting across heterogenous, isolated org entities (providers)
• rapid, non-disruptive rollout/tear down
• dynamic provider- and data sets
• fail-operational

Key Requirements

• non-intrusive: no modification of provider machines/network, no end user installation
• fail-operational: no single point of failure causing data loss
• self-repairing: automatic data sync upon (re)connection
• flexible: provider lists and data sets can be modified at runtime without disrupting operation
• uniform: can be used across several hierarchy levels
• extensible: optional provider specific automation

Solution

• ad hoc overlay using dedicated, pre-configured SHARE nodes inside provider networks

SHARE Node

• uniform node design (same SW, different config files)
• 4 optional functions: user-server, node-server, node-client, provider-import/export
• provider-local data display and entry through user-server (browser)
• sync with upstream and downstream through node-server/node-client
• upstream/downstream/user clients: JSON over websockets (allowing non-RACE/SHARE endpoints)

Conceptual Data Model and Flow

• distributed, replicated, filtered spreadsheet with typed rows (int,real,bool,string,intList,realList)
• columns owned by providers (write access)
• cells hold value and time
• owner + date resolve conflicts (CRDT)
• supports formulas (value- and time triggered)
• sync through connector upstream node

SHARE Data Update

• configured (semi-) static structure: ColumnList, RowList
• dynamic data: ColumnData (morphable key-value maps)
• change stimulus: ColumnDataChange, time-triggered formulas
• complete state snapshot: invariant Node object

Detailed Data Model

Formulas

Data Resynchronization Protocol

• based on column ownership (provider node) and ColumnData dates
• dates transmitted with NodeState message (per-CD for ext. columns, per-cell for own)
• newer own data transmitted with normal ColumnDataChange messages
• protocol can run in cycles until fixpoint is reached, i.e. does not need to halt local op

Detailed Data Resynchronization

Web Client

Provider Data Import / Export

• utilizing comprehensive RACE Import/Export actor infrastructure
• this is RACEs native domain (http, JMS, Kafka, sockets, ..., XML, Json, Protobuf, binary, ...)
• supports soft-realtime with > 1000 msg/sec
• configured actors that communicate via publish/subscribe channels
• imports processed as normal ColumnDataChange events
• can be gradually introduced/extended

Actors

• SHARe is generic RACE application
• using actors as concurrent execution units

TBD

• demo / presentation slides
• NTP implementation for clock sync
• runtime RowList/ColumnList swap
• user auth (pw change)
• more clients (Flutter)
• specs (variants, Fret?)
• global monitoring (Mesa?)