---
sheet_format: 1.0
id: 2024-07-maci-107
title: "Les fondations ouvertes d'un plugin oublié poussent les groupes dans le cloud"
title_en: "The open foundations of a forgotten plugin push big groups into the cloud"
speaker: Erwan Rougeux
with: [Horacio Gonzalez (Clever Cloud, host), Antoine Sabot-Durand (SCIAM), Florentin Dubois (Clever Cloud)]
date: 2024-07-26
event: "Message à caractère informatique #107"
place: "online / studio"
format: podcast
language: fr
duration: "58:54"
video: https://www.youtube.com/watch?v=AtuMFcoR1pw
slides: null
topics: [open source foundations, kubernetes, devops, cloud migration, microservices, haproxy, 3d printing, education]
source: "Summary written by Claude (AI) from the YouTube auto-caption transcript, with statements attributed to Erwan by local identification of his voice only (other speakers are not identified). Not a verbatim transcript; check the recording before quoting."
review: reviewed
review_date: 2026-10-22
---

# Les fondations ouvertes d'un plugin oublié poussent les groupes dans le cloud

**In one sentence:** Erwan Rougeux's first MACI, recorded on 18 July 2024: he recounts running 400 microservices on Kubernetes for French schools abroad with 1.5 FTE, argues that stacked abstractions make us forget the layers below, and presents Manyfold, a 3D asset manager.

## TL;DR
Host Horacio Gonzalez is joined by Antoine Sabot-Durand (Java architect at SCIAM, formerly at Red Hat), Florentin Dubois (engineering manager, network and automation at Clever Cloud) and Erwan Rougeux, who has just joined Clever Cloud as vice-president for international growth after ten years as CIO of the Mission laïque française. This episode and #108 come from the same recording session, on 18 July 2024. A Chromium extension that exposes hardware data to `*.google.com` pages leads to Google, Kubernetes and open-source foundations (Red Hat projects moving to foundations, Clever Cloud joining Eclipse). Erwan tells how his former association rebuilt on Docker, then Kubernetes, then Clever Cloud. Antoine describes a large European bank's move from mainframes to a private cloud that reproduces its silos. Florentin closes on HAProxy 3.0 and Maglev, Erwan on Manyfold.

## Human stake
- **For whom:** small non-profit IT teams, and the colleagues in schools around the world who depended on their services.
- **What is at stake:** keeping a school network's services running with almost no staff, and getting non-technical colleagues to understand the tools.
- **The documented choice:** he recounts rebuilding the association's platform on Docker and Kubernetes after a 2015 hack and running 400 microservices with 1.5 FTE ([18:42](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1122s), [21:53](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1313s)); he says he gave the Kubernetes comic to his general management to explain the change ([43:26](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2606s)).

## Erwan's key points
- He introduces himself as a self-taught geek, for ten years CIO of the Mission laïque française, a non-profit that builds and runs French schools abroad, who has recently joined Clever Cloud as vice-president in charge of international growth ([2:27](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=147s)).
- He recounts that after the association was hacked in 2015 (by Islamic State, he says) he had to rebuild everything and chose Docker over the historical VMs: Swarm first, then Kubernetes discovered at a DockerCon through Rancher, then bare metal ([18:42](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1122s), [19:26](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1166s)).
- Like many small teams, he says, they struggled: V1 was a failure, V2 coughed hard, V3 ended up at AWS, because coding paid more than maintaining infrastructure ([19:50](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1190s)).
- He recounts meeting Quentin Adam at the bar of a TGV neither of them should have taken; Quentin said Clever Cloud could do better than his Kubernetes setup, and the association later moved there for "serious savings" ([20:14](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1214s)).
- He credits Kubernetes' documentation and plug-and-play design: at the Covid peak, with schools from San Francisco to Wuhan and none in France, they maintained 400 microservices around the clock with 1.5 FTE and a budget of 200 k a year (his figures) ([21:25](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1285s), [21:53](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1313s)).
- A home-made PaaS automated with GitLab saved a lot of time; they trained colleagues in schools worldwide, built expertise hubs and wrote documentation, often in English, sometimes in Arabic ([22:19](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1339s), [22:45](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1365s)).
- Explaining the move from monolith to microservices was long his hobby-horse; he says the Kubernetes comics, given to his general management, helped him start changing things in a very siloed education world ([43:26](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2606s)).
- He argues that piling abstraction on abstraction makes us forget the techniques underneath: "we are inventing a square wheel because we forgot we already knew how to make round ones". Kubernetes, he says, is only an abstraction of Docker orchestration, itself an abstraction of LXC ([44:18](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2658s), [44:43](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2683s)).
- On no-code, he argues the code will not be the best ever written, but it gives everyone access to building a digital product: the more people code, the more tools, the more answers to problems ([45:06](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2706s), [45:29](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2729s)).
- He presents Manyfold, a digital asset manager for 3D files, used for his tabletop role-playing miniatures since he got a 3D printer: rules auto-tag files from folder names, and he says he moved its database to PostgreSQL for his 1.2 TB of files. It runs as a pod on his home Kubernetes server, which hosts about 70 services ([54:29](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3269s), [55:38](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3338s), [56:48](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3408s)).

## Outline
- [0:16](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=16s) Introductions: Antoine Sabot-Durand (architect at SCIAM, ex-Red Hat, CDI and MicroProfile), Erwan (ex-CIO of the Mission laïque française, now international growth at Clever Cloud), Florentin Dubois (network and automation at Clever Cloud), host Horacio Gonzalez.
- [4:04](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=244s) A tweet about an extension in Chromium's code that exposes system information to every `*.google.com` site, meant for Hangouts and Meet. Brave will disable it; no answer from Google.
- [8:04](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=484s) Worries about Fuchsia and data flowing from the OS to Google; a participant's ambivalence about Google's tools and their trade-offs.
- [10:49](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=649s) Kubernetes and the CNCF as a governance guarantee; Kubernetes as a rewrite of Borg by ex-Googlers, Prometheus, Nomad.
- [13:37](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=817s) IBM and the Eclipse Foundation; Red Hat projects moving to foundations: Quarkus considering it, Kogito/KIE heading to Apache. A neutral third party reassures contributors; why the CNCF works.
- [18:18](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1098s) Horacio asks Erwan how he met Kubernetes: the 2015 hack, Docker, Swarm, Rancher, bare metal; V1, V2, V3 at AWS; the TGV meeting with Quentin Adam.
- [21:25](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1285s) 400 microservices for schools from San Francisco to Wuhan, 1.5 FTE, 200 k a year; a GitLab-driven PaaS; training colleagues worldwide; documentation in English and Arabic.
- [23:57](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1437s) Apache versus Eclipse: Apache's processes are very strict, Eclipse more flexible. Horacio: Clever Cloud has joined the Eclipse Foundation, the only one with a European branch.
- [26:03](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1563s) Antoine on a large European bank porting its accounting platform from mainframes to microservices on a private cloud built on IBM Cloud, with the old world's security layers.
- [28:44](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1724s) Big groups go to the cloud without being organised for it: silos, ops whose only goal is that production does not fall, "DevOps with a big slash in the middle". The DevOps Topologies site.
- [31:05](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1865s) A Kubernetes that is not elastic because internal billing cannot charge CPU and memory: "on-premise inside the cloud". Testing in production; pods waiting up to two hours for a security layer.
- [35:42](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2142s) The DevOps promise badly implemented; "dev that does not wake you at night". DHH's "We are leaving the cloud". Cycles: mainframes, cloud, maybe back.
- [37:39](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2259s) Antoine on application servers (WebSphere, JBoss) and the risk of rebuilding a huge one brick by brick; moduliths; one function per microservice; start monolithic, split later.
- [43:02](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2582s) Acculturating decision-makers: Erwan and the Kubernetes comics for his general management; agility in a siloed education world.
- [44:18](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2658s) Erwan: abstraction layers make us forget the layers below; the square wheel; no-code as access for everyone.
- [45:53](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2753s) Antoine's two axes: acculturate management, and have techs learn the history of their technologies. The bank's "wild" migrations of application servers.
- [49:17](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2957s) Florentin on HAProxy 3.0: Sozu v1 is out and HTTP/2 is being implemented; stick tables; Google's Maglev paper and Clever Cloud's next-generation load balancing.
- [51:41](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3101s) HTTP/2 is harder than it looks: congestion, stream prioritisation; it can degrade downloads, such as images from S3.
- [54:29](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3269s) Erwan presents Manyfold: 3D asset management for role-playing miniatures, auto-tagging, PostgreSQL, 1.2 TB of files, a pod on his home Kubernetes.
- [58:00](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3480s) Closing; the session continues in the next episode.

## Notable quotes
> "On a dû maintenir 400 microservices 24 heures sur 24 et on les a maintenus avec 1,5 ETP et pour un budget de 200 k à l'année." — Erwan Rougeux, [21:53](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1313s)
*"We had to keep 400 microservices running around the clock, and we kept them running with 1.5 FTE and a budget of 200 k a year."*

> "Kubernetes n'est qu'une abstraction au final d'une orchestration de Docker qui est lui-même une abstraction des LXC." — Erwan Rougeux, [44:43](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2683s)
*"In the end Kubernetes is only an abstraction of Docker orchestration, which is itself an abstraction of LXC."*

> "On est en train d'inventer une roue carrée parce qu'on a oublié qu'on savait déjà faire des roues rondes." — Erwan Rougeux, [44:43](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2683s)
*"We are inventing a square wheel because we forgot we already knew how to make round ones."*

> "Plus il y a de personnes qui codent, plus il y a d'outils, plus il y a d'outils, plus il y a de réponses à des problèmes." — Erwan Rougeux, [45:29](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2729s)
*"The more people code, the more tools there are; the more tools, the more answers to problems."*


## Questions this answers
**Q: How did Erwan come to Kubernetes, and then to Clever Cloud?**
A: He recounts that after a 2015 hack of the Mission laïque française he rebuilt everything on Docker, then Kubernetes via Rancher, then bare metal; the third version ended on AWS. A chance meeting with Quentin Adam in a TGV led the association to Clever Cloud, for savings ([18:42](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1122s), [20:14](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1214s)).

**Q: How much does it take to run a school network's services on Kubernetes?**
A: Erwan's figures for the Mission laïque française at the Covid peak: 400 microservices, around the clock, with 1.5 FTE and 200 k a year, thanks to Kubernetes' documentation, a GitLab-automated PaaS and trained colleagues in the schools ([21:25](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1285s), [22:19](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1339s)).

**Q: How do you explain Kubernetes and microservices to non-technical management?**
A: Erwan says the Kubernetes comics, handed to his general management, were pitched at their level and let him start changing things. Antoine adds that management acculturation is costly but necessary ([43:26](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2606s), [45:53](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2753s)).

**Q: Why do open-source projects move to foundations?**
A: The guests say a neutral third party reassures contributors about a project's future and the fate of their code, at the cost of governance; Red Hat is considering it for Quarkus. Apache is described as very strict, Eclipse as more flexible ([15:09](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=909s), [23:57](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1437s)).

**Q: Why do big companies' cloud migrations go wrong?**
A: Antoine describes a large bank that moved to a private cloud without changing its organisation: dev and ops silos, Kubernetes sized for peak because resources cannot be billed internally, security layers that make pods wait up to two hours ([28:44](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1724s), [31:05](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=1865s), [34:12](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=2052s)).

**Q: Is HTTP/2 always faster?**
A: No, according to Florentin: congestion and stream prioritisation are hard, and for some cases, such as downloading images from S3, HTTP/2 can reduce performance ([51:41](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3101s), [53:41](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3221s)).

**Q: How do you organise a large collection of 3D printing files?**
A: Erwan recommends Manyfold, a 3D digital asset manager: it auto-tags files from folder names, has a viewer, and he runs it as a pod on his home Kubernetes with PostgreSQL for 1.2 TB of files ([54:29](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3269s), [56:48](https://www.youtube.com/watch?v=AtuMFcoR1pw&t=3408s)).

## Mentioned
Clever Cloud, SCIAM, Red Hat, Mission laïque française, CDI, Jakarta EE, MicroProfile, Quarkus, Chromium, Brave, Google Meet, Fuchsia, Google Cloud Platform, Kubernetes, CNCF, Borg, Nomad, IBM, Eclipse Foundation, Kogito, Apache Software Foundation, Redis, Docker, Swarm, DockerCon, Rancher, AWS, Quentin Adam, GitLab, Wuhan, IBM Cloud, DevOps Topologies, DHH, WebSphere, JBoss, OpenShift, HAProxy, Sozu, Maglev, Cloudflare, S3, Manyfold, PostgreSQL, Kickstarter, Patreon, Horacio Gonzalez, Antoine Sabot-Durand, Florentin Dubois