OpenSlides for club meetings: why our first server struggled and how we rebuilt it


Digital members' meetings look straightforward at first. Install OpenSlides, connect the video conference, invite everyone. In practice, installation isn't the hardest part. Running it reliably under real meeting conditions is.

Here's what happened at one of our annual general meetings with around 30 participants, why performance suffered and how we rebuilt the installation on a fresh VPS.


Our original setup

We were running:

  • OpenSlides in Docker on a small VPS
  • 4 vCPUs
  • 1 GB RAM
  • A separate Jitsi video conference
  • Participant lists, voting and presentation mode in use at the same time

The assumption was simple: thirty participants can't generate much load.

We were wrong.

What actually happened

Several things went wrong during the meeting:

  • Some participants couldn't log in at times
  • Pages were slow to respond
  • Voting results appeared late
  • A few browsers lost their connections

The video conference wasn't the culprit. Jitsi was hosted separately and remained stable. The trouble was inside OpenSlides.

OpenSlides is made up of several services:

  • Backend services
  • Datastore readers and writers
  • Redis message bus
  • PostgreSQL database
  • Proxy and client

When participants interact simultaneously, the system handles many small database operations and events. Voting can create especially sharp spikes.

With only 1 GB of memory, there was little room to breathe. The VPS started swapping, processes waited for I/O and the whole system became sluggish.

The common misunderstanding

The mistake is assuming few participants means little load.

What matters is:

  • How many people click at the same moment
  • How often voting is started
  • How many clients receive updates
  • Whether presentation mode is active

Thirty active participants can put more stress on a system than two hundred passive viewers.

What we learnt afterwards

Our analysis pointed to:

  • RAM as the limiting resource
  • Containers that started correctly but had no headroom
  • Short spikes causing noticeable delays
  • Missing live monitoring during the event

The architecture wasn't fundamentally broken. We'd simply sized the server too tightly.

Why we rebuilt it

Instead of endlessly patching the existing installation, we started fresh.

We wanted:

  • A reproducible setup
  • A clean separation between data and configuration
  • Headroom for spikes
  • Straightforward migration before future events

The new target was:

  • 4 vCPUs
  • 8 GB RAM
  • The standard OpenSlides Docker Compose setup
  • An external reverse proxy with TLS
  • Video conferencing kept separate

The improvement was noticeable immediately.

Why a fresh server helped

For an important live event, predictability matters more than squeezing out every last bit of efficiency.

A new server brought:

  • No accumulated leftovers
  • No hidden configuration fragments
  • A clear version baseline
  • Easier troubleshooting

In volunteer organisations, where meetings happen only occasionally, simplicity is often more valuable than elaborate optimisation.

Lessons learnt

  1. OpenSlides is a collection of services, not just a small web app.
  2. For our workload, RAM mattered more than CPU.
  3. Voting can trigger bursts of activity.
  4. Keep the video conference independent.
  5. Load-test the system before an important meeting.

What I'd recommend for clubs with up to around 50 participants

For a setup like ours, a sensible starting point is:

  • 4 vCPUs
  • 6–8 GB RAM
  • SSD storage
  • A separate video conferencing service

That's a starting point based on our experience, not an official universal minimum. Test the actual version and meeting workflow before you commit.

The next article covers the fresh VPS installation, Docker setup, migrations and the little configuration traps that cost us time.

Previous Next