2019-07-24 TC Meeting notes

Date

Jul 24, 2019

7am PST / 4pm CEST

Time	Item	Who	Notes

Time	Item	Who	Notes
5m	Agenda and action item review	@Josh Zamor (Deactivated)
20m	De-centralized lot management	@Sebastian Brudziński	Angola implementation cannot rely on centralized lot management and needs storeroom managers to be able to add lots in some way Any plans, designs or ideas for anything similar ever existed in core? MVP that would be accepted in core? It would be worth to design something that could be incorporated in core
15m	Performance tests	@Sebastian Brudziński	More and more are failing We cannot rely on results too much - Difference between the performance test results Do we have resources to continue fixing and taking care of them?
20m	Look at streaming from Casper	@Chongsun Ahn (Unlicensed)	Brief intro of goals of project: proof of concept to show a gradual v2→v3 migration focus on streaming data What we've learned about: considerations in building for streaming as opposed to batch lessons from technology involved what should OpenLMIS understand from this?
3m	UI caching design - call for reviews	@Sebastian Brudziński	UI Caching Design Comments are welcome - we want to get to implementation next week

Current implementation is to manage items and lots centrally, there's one admin right for both of these.
- data quality was the big reason to choose this
Angola needs:
- not centralized
- each user managing stock at the facility could manage lots as they go
  - through the physical inventory and receive processes
  - search first, then add lot as they go if it doesn't exist
- Need something before next core release
What's the minimum working version that'd be accepted in core
- configuration option to select either centralized or not-centralized
- keep the boundaries between ref data and stock mgmt correct
  - in Angola it'd be a UI and right change (split), in core we'd likely want this to be more comprehensive
  - seems like lots would need to be created before a stock event could be fully processed, as a lot id is part of the stock event data
- strategy for de-duplication:
  - report for mgmt to find dupes
  - stock events and stock line items are written in stone - they can't change. We should keep this in mind that UPDATE on certain tables are not allowed.

They've been failing since forever
Sometimes a test starts to fail, it could be a problem in starting ref data, could be a red herring, could be an actual failure of not meeting criteria
- functional tests don't seem to be having the same issues.
Not many days that performance tests pass, see link above
- oftentimes the results vary wildly between runs
What we've done:
- warmup phase
What value do we have for these tests?
- Perhaps none
  - The small bumps every 5 builds, 10 builds, etc add up to a failure for the test suite almost all the time.
- To make them valuable:
  - We could continue playing the raise the limit game
  - We could look at trend lines instead of pass/fail indicators
Want to revisit this topic, we'll create a ticket and brainstorm ideas for improve reliability

Migration from v2 to v3 over a long period of time so that users don't face a hard-switch point
- Focused on moving Requisitions between systems
- One way data pipeline
  - could this be both ways? yes, the main work is in building the transformation steps in Nifi, so quite possible.
- Another use case could be:
  - in onboarding to v3, we could use a bi-way flow that would allow users to start using v3, and if they encounter a bug they could fall back to v2.
Using data-pumps and Nifi
- Kafka
- Debezium
- Nifi
- Kafka Connect
Saw data streaming out of v2 and into v3 in near-real-time
Want to move this tech into our reporting stack
- How resilient is the tech? Is it mature?
  - Restarting the stack is working, pieces can be brought down without the whole thing falling down
  - Could be used for reporting pretty easily

@Josh Zamor (Deactivated) to look at caching document