💾 Archived View for thebird.nl › gn-gemtext-threads › topics › systems › orchestration.gmi captured on 2023-04-26 at 13:18:15. Gemini links have been rewritten to link to archived content
View Raw
More Information
⬅️ Previous capture (2023-01-29)
-=-=-=-=-=-=-
Orchestration and fallbacks
After the Penguin2 crash in Aug. 2022 it has become increasingly clear how hard it is to deploy GeneNetwork. GNU Guix helps a great deal with dependencies, but it does not handle orchestration between machines/services well. Also we need to look at the future.
What is GN today in terms of services
- [X] Main GN2 server (Python, 20+ processes, 3+ instances: depends on all below)
- [X] Matching GN3 server and REST endpoint (Python: less dependencies)
- [X] Mariadb
- [X] redis
- [X] virtuoso (@aruni)
- [X] GN-proxy (Racket, authentication handler: redis, mariadb)
- [X] Alias proxy (Racket, gene aliases wikidata)
- [X] opar server
- [+] Jupyter, R-shiny and Julia notebooks, nb-hub server
- [X] BNW server (@efraimf)
- [+] UCSC browser (@efraimf)
- [X] GN1 instances (older python, 12 instances in principle, 2 running today)
- [ ] Access to HPC for GEMMA (coming)
- [+] Backup services (sheepdog, rsync, borg)
- [+] monitoring services (incl. systemd, gunicorn, shepherd, sheepdog)
- [ ] mail server
- [X] https certificates
- [X] http(s) proxy (nginx)
- [X] CI/CD services (with github webhooks)
- [+] git server (gitea or cgit)
- [X] file server (formerly IPFS)
- [ ] SPARQL endpoint
Somewhat decoupled services:
- [X] genecup
- [X] R/shiny power service Dave
- [ ] biohackrxiv
- [ ] hegp
- [ ] covid19
- [ ] guix publish server (runs on penguin2, needs tux02 @efraimf)
I am still missing a few! All run by a man and his diligent dog.
For the future the orchestration needs to be more robust and resilient. This means:
- A fallback for every service on a separate machine
- Improved privacy protection for (future) human data
- Separate servers serving different data sources
- Partial synchronization between data sources
The only way we *can* scale is by adding machines. But the system is not yet ready for that. Also getting rid of monolithic primary databases in favor of files helps synchronization.