Red Hat tells Ansible 2.7 upgraders to settle the deployment model before booking the window

The decisions Red Hat details first are where the platform runs and how many copies of it a team keeps ^[9][16]. The third is high availability and disaster recovery, introduced with the observation that automation has moved from a supportive utility to critical enterprise infrastructure ^[20]. The post does not tie the MCP server or the Vault OIDC support to a particular deployment model ^[23].

The three paths in decision one differ mostly in who operates the thing. Self-managed on Red Hat Enterprise Linux virtual machines gives deep control over infrastructure and suits teams with existing operational expertise ^[10]. Operator-managed on OpenShift or Kubernetes is the container orchestration route ^[11]. The managed service on Microsoft Azure or Amazon Web Services offloads platform operations to Red Hat site reliability engineers ^[12]. Two of the three leave the pager with the customer's own team ^[22]. Red Hat's line on this is that there is no wrong answer, only the right fit for a team's requirements ^[13].

Decision two is the one that quietly sets headcount. A team can run a single multi-tenant production platform and use automation mesh for network separation, or distribute platforms across different networks ^[16]. Red Hat wrote that teams should "Determine how many platforms provide genuine value versus the operational overhead required to maintain them" ^[17]. The same section says robust non-production testing is "no longer optional, but a prerequisite for success" as content grows in complexity ^[18], and points at Configuration as Code and ephemeral test infrastructure as the levers ^[19].

Portability matters here. Dev and production do not have to match, because execution environments keep automation consistent regardless of the underlying platform deployment ^[14]. Red Hat also recommends defining the deployments themselves as Infrastructure as Code ^[15]. For a team that has been hand-building non-production to mirror production, that changes what the second environment has to be.

One way to plan the window is to separate the decisions whose wrong answer costs a reconfiguration from the ones whose wrong answer costs a second migration. Vault OIDC and how execution environments get built are the first kind ^[3][4]. The deployment model and the number of production platforms are the second kind ^[9][16]. Red Hat says the six decisions matter especially for anyone still on an RPM-based installation of 2.4 or 2.5 ^[8]. That group answers the expensive questions even if the AI orchestration features are what brought them ^[2].