Date

14 Nov 2017

7am PST / 4pm CET

Meeting Link

https://meetings.webex.com/collabs/#/meetings/detail?uuid=M0UUCAJ68OYLPMKAROOE9PJSBW-3O29&rnd=897752.12165

Toll: +1-415-655-0001

Number: 199 948 686

Attendees

Goals

Feedback on Demo-data loading approaches
Review error message handling from dev forum

Discussion items

Time	Item	Who	Notes
5m	Action Item review from last time	Josh Zamor (Deactivated)
30m	Demo data loading	Chongsun Ahn (Unlicensed)	Problem statement and options for solution
	Review dev forum post(s) on error messages	Paweł Albecki (Deactivated) and Brandon Bowersox-Johnson	Dev forum post: https://groups.google.com/forum/#!topic/openlmis-dev/iwGo2mLR3wk
5m	Multi-cloud / evaluating Azure	Ben Leibert	Note: may not be ready for this yet
	Non-functional requirement progress - performance?		Sort of did this in showcase
	IDs in DTO's equals and hashcode - effects on etags	Chongsun Ahn (Unlicensed)	Note: May not be ready for this quite yet

Notes

Demo Data

Current: on wiki - Dev writes JSON files which are turned into SQL files (generated)
Biggest pain points:
- Dev has to generate their own UUID
- Many tools use CSV, people can use CSV, but CSV to JSON is currently a slow process
- Dev has to ensure loading-order is correct
Bottom line: Developer is needed, non-devs can't realistically manage this directly. Even for a dev this is slow.

New approaches:

Merge ref data seed tool into each service
1. Some great advantages (hit the major pain points)
2. Difficult to put into a service
3. In our startup of a service, there is a timing issue - many testing tools rely on demo-data to exist before starting
4. Code references (for ref integrity) are configurable in the mapping file
Use the performance data approach
1. CSV files - generate with Mockaroo
2. Hits all the major pain points
Nifi
1. https://nifi.apache.org/
2. Has the same startup issue

UUID's generated:

Contract tests and performance tests use those UUID
Even some manual QA tests use UUID (maybe legacy issue - we want to re-write these using the admin UI)
So: changing existing UUID's could be a big problem
- If the UUID's could make the tests smarter - wouldn't fix QA tests but then we wouldn't have this hard-coded / human made UUID issue
Could Excel do this!?

Error responses - consistency

Dev forum had already focused on option 3 - use the top level error message with some section below that. Keeps backward compatability.

Further we discussed the idea for nesting (details within details) which seems like something we could do if needed. Ideas for usage was something along the lines of semantically structuring error messages (e.g. there's an error, there's an error in the budget, there's an error in budget line item 5) or directly down to a line item.

Between fieldErrors or renaming it to details we felt there's value in just keeping it a direct "fieldErrors" name.

Azure

$5000 credit / year for any non-profit in certain countries (more the developed countries - USA is in the list)
Opening two data centers in Africa in 2018.
Non-technical decision makers within countries are very hesitant to move their data out of their country. AWS is unknown outside of the technical community and therefore untrusted. By contrast, stakeholders are familiar andcomfortable with Office365. Selling them on the use of Microsoft’s cloud, which they probably already use, is likely a lot easier than selling them on AWS.
Intuitive and modern UI – great for organizations which are less technical but nevertheless interested in maintaining their own OpenLMIS instance.
Users can schedule recurring/automated backups for the days/times of their choice.
Comes with built-in monitoring and alerting - may obviate the added complexity of Scalyr.
Not sure about competitor to Amazon ECS or Kubernetes / Swarm mode

Action Items

Sebastian Brudziński: followup with Chongsun on dmeo-data loading and ref data seed tool
Paweł Albecki (Deactivated): writeup decision on error messages and put into our style guide (RTD)