DACI: How to implement a repository with history?
Tips and info
Recommendations
Contributors
Contributors: I am seeking the right people to get involved in the decision. Add your comments to this page, let's get the conversation started.
Please add:
- The people directly impacted by this so we can include them.
- Any references to previous work and investigations that we can leverage.
- Any constraints and challenges we need to consider to make this decision and following action plan.
- Any additional options we should consider before making the decision.
Background
A common scenario we come across with almost all metadata repositories we have seen is that they lack the ability to store historical information about metadata and respond to point-in-time inquiries. While Egeria's type system and APIs have been built from the beginning to support such history, we have not yet implemented a backend storage option that implements history.
Considering this comes up frequently as a common need, even to augment existing metadata repositories, providing such a historical store for metadata could be a somewhat narrow but nonetheless extremely common adoption point for Egeria.
Current state
We are currently considering implementation options for an initial approach to such a repository.
Data for decision support
- Identification of potential technologies to use as the backing store for such a repository.
Options considered
| Option 1: bi-temporal RDBMS | Option 2: bi-temporal graph | Option 3: search index | |
|---|---|---|---|
Description | Using a bi-temporal relational database like DB2 | Using a bi-temporal graph store like Crux | Using a search index like Elastic |
Rollout plan | |||
| Pros and cons | Native Handles historical information natively at the storage layer, so should be simpler to implement point-in-time inquiry. New approach Takes a new approach to a backing store (relational) compared to our existing implementations (graph-based) Commercial We are unaware of any open source, native bi-temporal RDBMS, so this would put a dependency on licensed commercial software. Schema Requires a fixed schema, which raises questions about how to both handle efficient queries (not storing things as unqueryable blobs) but also manage history when the type system itself (schema?) may have changed over the course of that history (ie. deprecated attributes and types) | Native Handles historical information natively at the storage layer, so should be simpler to implement point-in-time inquiry. Similar to existing Close alignment with our current repository approaches that are more graph-focused than relational. Embedded option Provides a simple option to run in an embedded capacity, which could be useful for demonstration purposes (not requiring additional infrastructure and components). Pluggable backends Implemented using pluggable characteristics for its own backends, including both open source and commercial options. Schemaless It sounds like each document in Crux is essentially schema-less (tuples / triples-based), so it may be feasible to store multiple versions of a type across the history of a given instance of metadata | |
Risks | Scalability The resource requirements that might be necessary for a "true production" rollout are unclear, or the volume to which it can scale. (We heard mention of "16 TB" (sounds plenty) but also "10 million triples" (with history, and one triple per attribute value, per instance, this sounds small?) | ||
Estimated cost and effort |
FAQ
Q1.
A1.
References
Follow-up action items
Learn more: https://www.atlassian.com/team-playbook/plays/daci
Copyright © 2016 Atlassian
This work is licensed under a Creative Commons Attribution-Non Commercial-Share Alike 4.0 International License.