- Schedule the upgrade during a period of low platform use to limit the performance impact.
- Downgrading to an earlier release is not supported. Back up your data stores before you upgrade.
- When a release needs downtime or an extra step, its release notes say so in the Upgrade notes section.
Which upgrades does this cover?
A zero-downtime rolling upgrade applies when you move from one maintenance release to a later one in the same LTS family, for example from 5.9.6 to 5.9.7. An upgrade from one LTS family to another, such as 5.1.x to 5.9.x, is a different procedure. See the upgrade runbook. Moving from the per-service Helm charts to the singleflowx-ai chart is also not a rolling upgrade. It needs a maintenance window, see Migrate self-hosted FlowX.AI to the flowx-ai Helm chart.
How do I check whether a release supports it?
Each LTS maintenance release states it at the top of its Upgrade notes section, in one of three forms:
If you skip releases, for example from 5.9.5 to 5.9.7, read the Upgrade notes of every release in between and complete their steps too.
How do I run the rolling upgrade?
1
Read the release notes
Read the Upgrade notes and the deployment guidelines of the target release, and of every release you skip. Complete the steps they list, such as creating a new Kafka topic before the services start.
2
Back up your data stores
Take a backup of your databases and object store. Downgrading to an earlier release is not supported, so the backup is your recovery point.
3
Roll out the new release
Upgrade your Helm release to the chart version of the target platform release, with your existing values plus any changes the release notes ask for. With the
flowx-ai chart, the chart version is the platform release, and the chart sets every service’s image tag to the version that ships with it.Kubernetes replaces the pods of each Deployment gradually, and old and new pods run side by side until the rollout ends. Services can be upgraded in any order.4
Watch the rollout
Wait until every Deployment reports the new pods as available:Then sign in to FlowX.AI Designer and run a process to check the platform end to end.
What keeps the rollout free of downtime?
Theflowx-ai chart deploys every service as a Deployment with the RollingUpdate strategy, set under global.updateStrategy, and keeps the Kubernetes defaults for how many pods it adds and removes at a time. With one replica, Kubernetes starts the new pod first and removes the old pod only when the new one is ready.
A pod is ready when its readiness probe passes:
- Core services have a readiness probe on by default, on
/manage/actuator/health/readiness. - AI services have no readiness probe by default. Kubernetes counts a new AI service pod as ready as soon as its container starts, before the service has finished starting. To gate them, set
services.<name>.customReadinessProbeto a probe that fits the service.
- Don’t set
updateStrategy.typetoRecreate, globally or for a service.Recreatestops every old pod before it starts the new ones. - Run two or more replicas of the services that must stay available, and add a disruption budget. See How do I scale a service and keep it available?
RollingUpdate strategy, and the container has a readiness probe.
Related resources
Configure the flowx-ai Helm chart
Replicas, autoscaling, and disruption budgets per service
Migrate to the flowx-ai Helm chart
Move from the per-service charts, with a maintenance window

