co-op/support
2
0 Fork 0
Code Issues 10 Pull Requests Actions Packages Projects Releases Wiki Activity

Move Cloudron to green infrastructure #2

New Issue
Open
opened 2026-09-24 17:46:33 +00:00 by ntnsndr · 3 comments
ntnsndr commented 2026-09-24 17:46:33 +00:00
Copy link
Copy source

What's the idea?

Currently, the co-op's Cloudron system (which all our services are hosted on, but not any AI models) is running on Nathan's personal Hostinger account. We should transition to a co-op-owned system.

Because this system had to be set up before we had our email and domain ready, we couldn't attach

Why does it matter?

  • Independence from Nathan's personal account
  • Green energy infrastructure

How might it work?

The process for server migration with Cloudron is relatively straightforward, but there might be some downtime.

Potential providers

## What's the idea? Currently, the co-op's Cloudron system (which all our services are hosted on, but not any AI models) is running on Nathan's personal Hostinger account. We should transition to a co-op-owned system. Because this system had to be set up before we had our email and domain ready, we couldn't attach ## Why does it matter? * Independence from Nathan's personal account * Green energy infrastructure ## How might it work? The process for server migration with Cloudron is relatively straightforward, but there might be some downtime. ## Potential providers * Scaleway VPS * Migrate to a different Hostinger account, which might be easier - Hostinger's sustainability site: https://www.hostinger.com/sustainability
inference-bot commented 2026-09-29 18:31:20 +00:00
Owner
Copy link
Copy source

Provider analysis (researched 2026-09-29)

Ground truth from the live system: the Cloudron host is a Hostinger VPS in Campinas, Brazil (AS47583) — 2 vCPU, 8 GB RAM, 96 GB disk, currently at ~0.4 load average, 3.2 GB RAM available, disk 40% used. 13 apps, all healthy. The AI inference itself does not run here (it's at Tinfoil/GreenPT/PublicAI) — this box hosts the gateway, portal, chat, git, forum, email, and other services. So the requirements are: a reliable mid-size VPS with good latency to a mostly-US/EU membership, and a values story.

Where the candidates stand

Provider Location Energy Privacy / jurisdiction Notes Est. price
Hetzner Germany (Nuremberg/Falkenstein) or Finland (Helsinki) 100% renewable (hydro) — Germany since 2008, Finland since 2018 (sustainability page) EU / GDPR; German company Cloudron's best-documented host; strong price/performance ~€14/mo for CX32 (4 vCPU / 8 GB — double the current vCPUs)
Infomaniak Geneva, Switzerland (own datacenters, not colo) 100% renewable hydro, free-air cooling, TCO Certified Swiss jurisdiction (strongest privacy law); independent, employee-co-owned Values-perfect fit; costs roughly 2× Hetzner ~CHF 25–35/mo equivalent
1984.hosting Iceland 100% renewable (geothermal/hydro) Iceland; civil-liberties host since 2006 Great story; smaller operator; prepaid billing (1–3 yr); would want a disk/network benchmark first ~€9–20/mo
Scaleway France Good renewable claims (market-based) EU Already mentioned in the original post; reasonable varies
Hostinger (current) Brazil Grid mix undisclosed; no renewable guarantee BR; promo-renewal pricing Status quo KVM 2 renewal pricing

Sources: Hetzner sustainability, Infomaniak D3 datacenter, 1984.hosting.

Assessment

  • Hetzner (recommended default): best price/performance, 100% hydro, EU jurisdiction, Cloudron installs cleanly there. Moving Brazil → Europe reduces latency for our US members (~90–110 ms vs ~140–180 ms today) and improves latency to GreenPT (EU-based) sharply.
  • Infomaniak (values-maximal): employee-co-owned, Swiss privacy law, hydro, own DCs — the brand story is nearly perfectly aligned with a co-op. ~2× the price.
  • 1984.hosting (wildcard): cheapest green story and civil-liberties DNA; risk is relying on a small shop for all services.
  • Staying on Hostinger but moving accounts: solves the personal-account dependency but not the green/latency issues; Hostinger's own sustainability page describes efficiency measures and carbon-compensation claims rather than procured renewable electricity for this location.

Migration process (Cloudron, low downtime)

  1. Pre-stage: provision the new VPS (same or larger specs), install Cloudron, complete setup on a temporary domain.
  2. Backup: full Cloudron backup on the current server (all 13 apps + their data).
  3. Restore: on the new server, restore from that backup (Cloudron's built-in migrate path). Apps come back with data intact; internal app config travels with them.
  4. Verify: check app health, log into each critical app (portal, gateway, chat, git), run a test inference call through the gateway.
  5. DNS cutover: at domains.coop, repoint the A records (my., chat., portal., gateway., git., forum., newsletter., calendar., dashboard., panel., search., vault., email., inference.coop) to the new IP. Strip trailing dots per our DNS convention. TTL is already low.
  6. Propagate + burn-in: keep the old VPS running 24–48 h as DNS propagates, then take a final backup, cancel the Hostinger plan.

Downtime: effectively only step 5's propagation window — minutes, not hours, since we can pre-restore and switch DNS atomically per-app.

Analysis prepared by Inferencebot; happy to rerun with live benchmarks (disk/network tests) against any finalist.

## Provider analysis (researched 2026-09-29) Ground truth from the live system: the Cloudron host is a **Hostinger VPS in Campinas, Brazil** (AS47583) — 2 vCPU, 8 GB RAM, 96 GB disk, currently at ~0.4 load average, 3.2 GB RAM available, disk 40% used. 13 apps, all healthy. The AI inference itself does **not** run here (it's at Tinfoil/GreenPT/PublicAI) — this box hosts the gateway, portal, chat, git, forum, email, and other services. So the requirements are: a reliable mid-size VPS with good latency to a mostly-US/EU membership, and a values story. ### Where the candidates stand | Provider | Location | Energy | Privacy / jurisdiction | Notes | Est. price | |---|---|---|---|---|---| | **Hetzner** | Germany (Nuremberg/Falkenstein) or Finland (Helsinki) | **100% renewable (hydro)** — Germany since 2008, Finland since 2018 ([sustainability page](https://www.hetzner.com/unternehmen/nachhaltigkeit/)) | EU / GDPR; German company | Cloudron's best-documented host; strong price/performance | ~€14/mo for CX32 (4 vCPU / 8 GB — double the current vCPUs) | | **Infomaniak** | Geneva, Switzerland (own datacenters, not colo) | **100% renewable hydro**, free-air cooling, TCO Certified | **Swiss jurisdiction** (strongest privacy law); independent, employee-co-owned | Values-perfect fit; costs roughly 2× Hetzner | ~CHF 25–35/mo equivalent | | **1984.hosting** | Iceland | **100% renewable** (geothermal/hydro) | Iceland; civil-liberties host since 2006 | Great story; smaller operator; prepaid billing (1–3 yr); would want a disk/network benchmark first | ~€9–20/mo | | Scaleway | France | Good renewable claims (market-based) | EU | Already mentioned in the original post; reasonable | varies | | Hostinger (current) | Brazil | Grid mix undisclosed; no renewable guarantee | BR; promo-renewal pricing | Status quo | KVM 2 renewal pricing | Sources: [Hetzner sustainability](https://www.hetzner.com/unternehmen/nachhaltigkeit/), [Infomaniak D3 datacenter](https://www.datacenters.com/infomaniak-network-sa-d3-geneva), [1984.hosting](https://1984.hosting/). ### Assessment - **Hetzner (recommended default):** best price/performance, 100% hydro, EU jurisdiction, Cloudron installs cleanly there. Moving Brazil → Europe reduces latency for our US members (~90–110 ms vs ~140–180 ms today) and improves latency to GreenPT (EU-based) sharply. - **Infomaniak (values-maximal):** employee-co-owned, Swiss privacy law, hydro, own DCs — the brand story is nearly perfectly aligned with a co-op. ~2× the price. - **1984.hosting (wildcard):** cheapest green story and civil-liberties DNA; risk is relying on a small shop for all services. - **Staying on Hostinger but moving accounts:** solves the personal-account dependency but not the green/latency issues; Hostinger's own sustainability page describes efficiency measures and carbon-compensation claims rather than procured renewable electricity for this location. ### Migration process (Cloudron, low downtime) 1. **Pre-stage:** provision the new VPS (same or larger specs), install Cloudron, complete setup on a temporary domain. 2. **Backup:** full Cloudron backup on the current server (all 13 apps + their data). 3. **Restore:** on the new server, restore from that backup (Cloudron's built-in migrate path). Apps come back with data intact; internal app config travels with them. 4. **Verify:** check app health, log into each critical app (portal, gateway, chat, git), run a test inference call through the gateway. 5. **DNS cutover:** at domains.coop, repoint the A records (my., chat., portal., gateway., git., forum., newsletter., calendar., dashboard., panel., search., vault., email., inference.coop) to the new IP. Strip trailing dots per our DNS convention. TTL is already low. 6. **Propagate + burn-in:** keep the old VPS running 24–48 h as DNS propagates, then take a final backup, cancel the Hostinger plan. Downtime: effectively only step 5's propagation window — minutes, not hours, since we can pre-restore and switch DNS atomically per-app. *Analysis prepared by Inferencebot; happy to rerun with live benchmarks (disk/network tests) against any finalist.*
inference-bot commented 2026-09-29 21:21:23 +00:00
Owner
Copy link
Copy source

VPS Lite follow-up (researched 2026-09-29)

Infomaniak's VPS Lite line looks very attractive at first glance — here are the verified facts and the catches.

Lite tiers (annual commitment)

CHF/mo vCPU RAM Disk
2.70 1 2 GB 20 GB
5.40 2 2 GB 40 GB
7.20 2 4 GB 60 GB
9.00 4 4 GB 80 GB
18.00 4 8 GB 160 GB ← matches our current host + double the vCPUs

Same values package as the rest of Infomaniak: Swiss jurisdiction, employee-co-owned, own datacenters, 100% hydro, free-air cooling. Competitive with Hetzner's ~€14 CX32.

The four catches (from Infomaniak's own Lite-vs-Cloud comparison)

VPS Lite VPS Cloud
Bandwidth 500 Mbit/s 10 Gb/s
SLA none 99.99%
Snapshots no 1 included (expandable)
Flexibility scale up only up and down
Port 25 (outbound email) blocked granted on justified request

Why these matter for us specifically:

  1. No SLA — 13 apps serving real members with no availability guarantee. Infomaniak's own positioning says Lite is for "projects that are not yet critical."
  2. No snapshots — workable: Cloudron's built-in backups to an external target (e.g., Infomaniak's own object storage) cover recovery.
  3. Scale up only — fine for a growing co-op.
  4. Port 25 blocked — the big one. We run email.inference.coop (Cloudron's mail server) on this box; Cloudron mail delivers directly to recipient MXes over port 25. Blocked port = member email, portal notifications, and Listmonk newsletters all stop working unless we route everything through an external SMTP relay (extra dependency + deliverability puzzle).

Updated recommendation

Infomaniak VPS Cloud (entry tier: 4 vCPU / 12 GB / 250 GB NVMe) — the cheapest Infomaniak tier that keeps the 99.99% SLA, snapshots, resize-both-ways, and port 25 access. Roughly double Lite's top tier, but email alone justifies it, and 12 GB gives the 13 apps headroom as membership grows.

Alternative architecture — split: Lite top tier (CHF 18) for everything except mail, moving email to Infomaniak's own Mail Service (their core product, hydro-powered, likely better deliverability than self-run Cloudron mail). Cleaner separation, bigger migration. Worth considering at the meeting.

Cheapest acceptable: Lite top tier + external SMTP relay (e.g., an SMTP provider with good reputation). Works, but adds a third-party dependency for a co-op that values self-hosting.

Sources: VPS Lite plans (pricing + Lite-vs-Cloud table), VPS Cloud.

TL;DR: Lite is a trap for our particular stack — not because of compute, but because of email. VPS Cloud entry tier is the Infomaniak option that actually fits.

## VPS Lite follow-up (researched 2026-09-29) Infomaniak's **VPS Lite** line looks very attractive at first glance — here are the verified facts and the catches. ### Lite tiers (annual commitment) | CHF/mo | vCPU | RAM | Disk | |---|---|---|---| | 2.70 | 1 | 2 GB | 20 GB | | 5.40 | 2 | 2 GB | 40 GB | | 7.20 | 2 | 4 GB | 60 GB | | 9.00 | 4 | 4 GB | 80 GB | | **18.00** | **4** | **8 GB** | **160 GB** ← matches our current host + double the vCPUs | Same values package as the rest of Infomaniak: Swiss jurisdiction, employee-co-owned, own datacenters, 100% hydro, free-air cooling. Competitive with Hetzner's ~€14 CX32. ### The four catches (from Infomaniak's own Lite-vs-Cloud comparison) | | VPS Lite | VPS Cloud | |---|---|---| | Bandwidth | 500 Mbit/s | 10 Gb/s | | **SLA** | **none** | 99.99% | | Snapshots | no | 1 included (expandable) | | Flexibility | scale up only | up and down | | **Port 25 (outbound email)** | **blocked** | granted on justified request | Why these matter for us specifically: 1. **No SLA** — 13 apps serving real members with no availability guarantee. Infomaniak's own positioning says Lite is for "projects that are not yet critical." 2. **No snapshots** — workable: Cloudron's built-in backups to an external target (e.g., Infomaniak's own object storage) cover recovery. 3. **Scale up only** — fine for a growing co-op. 4. **Port 25 blocked** — the big one. We run **email.inference.coop** (Cloudron's mail server) on this box; Cloudron mail delivers directly to recipient MXes over port 25. Blocked port = member email, portal notifications, and Listmonk newsletters all stop working unless we route everything through an external SMTP relay (extra dependency + deliverability puzzle). ### Updated recommendation **Infomaniak VPS Cloud (entry tier: 4 vCPU / 12 GB / 250 GB NVMe)** — the cheapest Infomaniak tier that keeps the 99.99% SLA, snapshots, resize-both-ways, and port 25 access. Roughly double Lite's top tier, but email alone justifies it, and 12 GB gives the 13 apps headroom as membership grows. **Alternative architecture — split:** Lite top tier (CHF 18) for everything *except* mail, moving email to Infomaniak's own **Mail Service** (their core product, hydro-powered, likely better deliverability than self-run Cloudron mail). Cleaner separation, bigger migration. Worth considering at the meeting. **Cheapest acceptable:** Lite top tier + external SMTP relay (e.g., an SMTP provider with good reputation). Works, but adds a third-party dependency for a co-op that values self-hosting. Sources: [VPS Lite plans](https://www.infomaniak.com/en/hosting/vps-lite) (pricing + Lite-vs-Cloud table), [VPS Cloud](https://www.infomaniak.com/en/hosting/vps-cloud). *TL;DR: Lite is a trap for our particular stack — not because of compute, but because of email. VPS Cloud entry tier is the Infomaniak option that actually fits.*
ntnsndr commented 2026-09-30 19:20:42 +00:00
Author
Copy link
Copy source

Another consideration: Can a new provider help improve email deliverability? #8

Another consideration: Can a new provider help improve email deliverability? https://git.inference.coop/co-op/support/issues/8
Sign in to join this conversation.
No Branch/Tag Specified
Labels
No items
No labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
inference-admin (Admin) inference-bot (InferenceBot)
No Assignees
2 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: co-op/support#2