Split into multiple VMs #5

New issue

Closed

opened 2023-04-13 15:27:08 +00:00 by pim · 0 comments

pim commented

2023-04-13 15:27:08 +00:00

Owner

Idea: have two VMs per physical server that participate in Docker swarm. One for services that are either stateless or have no crucial data. Let's call such a VM a "public VM". The other VM contains important data. Let's call such a VM a "private VM". Additionally, we need a "manager" VM that acts as the Docker swarm manager, but can also run some other management services.

We can thus classify each current service:

Public service: mastodon, Pim's homepage, cyberchef, jitsi meet, inbucket, kms
Private service: seafile, syncthing, freshrss, overleaf, hedgedoc
Management service: traefik, prometheus, forgejo

Given that we do not have network storage yet, we have to be careful to constraint a certain service to a certain VM. This is the case for the following services: mastodon, seafile, syncthing, freshrss, overleaf, hedgedoc, forgejo
We should do a simple measurement which currently uses the most CPU, to spread them over physical servers.

We have three physical servers. One of these will solely act as the backup server in order to have a separate fault domain and seperation of duty. We therefore have two physical machines to run VMs on. Idea: have two VMs per physical server that participate in Docker swarm. One for services that are either stateless or have no crucial data. Let's call such a VM a "public VM". The other VM contains important data. Let's call such a VM a "private VM". Additionally, we need a "manager" VM that acts as the Docker swarm manager, but can also run some other management services. We can thus classify each current service: - Public service: mastodon, Pim's homepage, cyberchef, jitsi meet, inbucket, kms - Private service: seafile, syncthing, freshrss, overleaf, hedgedoc - Management service: traefik, prometheus, forgejo Given that we do not have network storage yet, we have to be careful to constraint a certain service to a certain VM. This is the case for the following services: mastodon, seafile, syncthing, freshrss, overleaf, hedgedoc, forgejo We should do a simple measurement which currently uses the most CPU, to spread them over physical servers.