Skip to content

A machine is offline

Offline means nothing has been heard from the agent for 300 seconds. It does not mean the machine is off, and it does not mean the agent is broken — those are two of several possibilities.

  1. Try SSH. If that fails too, the problem is below Flotera — power, network or the host itself. Nothing on this page will help until SSH works.

  2. Terminal window
    systemctl status flotera-agent

    If it is stopped or restarting in a loop, look at why:

    Terminal window
    journalctl -u flotera-agent -n 50 --no-pager
  3. The agent needs outbound HTTPS to agent.flotera.com.

    Terminal window
    curl -sS -o /dev/null -w '%{http_code}\n' https://agent.flotera.com/

    This is the most common cause of a machine that was fine yesterday: a firewall rule, a new egress policy, or an upstream network change. A response — any HTTP status — means the path is open.

  4. Did something replace the driver underneath it?

    Section titled “Did something replace the driver underneath it?”

    An unattended OS upgrade can pull the GPU driver out from under a running agent. Check whether an upgrade ran recently, and turn automatic updates off — that requirement exists for exactly this reason.

  5. It is idempotent and will not touch your credentials or the command journal.

    Terminal window
    /bin/bash -c "$(curl -fsSL https://get.flotera.com)" -- --token erm_init_...

It came back on its own and then went again

Section titled “It came back on its own and then went again”

A machine that flaps is telling you something real, and muting the alert hides it. Usual causes, in order:

  • Power — a marginal PSU under GPU load, or a shared circuit.
  • Network — a consumer router that reboots itself, or DHCP churn.
  • Thermal shutdown — check the temperature history on the machine’s card around the times it dropped.

A machine that goes offline mid-rental fails that rental, and the downtime counts against your reliability score on the marketplace. This is why announcing maintenance matters — announced downtime is treated very differently from sudden downtime.