[kolla][kolla-ansible] Managing Object Storage Backend via Kolla-Ansible
Hello everyone, We are currently using Kolla-Ansible stable/2024.1 to manage our OpenStack deployment, including both core services and the Swift object storage backend. Our goal is to manage all OpenStack services, including object storage, through a unified lifecycle management tool. We understand that managing large volumes of data can affect the performance of other services. To address this, we've separated the data flow entry points that could become bottlenecks into dedicated host groups: a load balancer and a Swift proxy. Specifically, we've created two separate host groups: one for the secondary load balancer (which serves as the initial entry point) and another for the Swift proxy services. 🔹 Key Point 1: Decoupling Swift Traffic We’ve separated the Swift proxy nodes from the other OpenStack API services. This also means that the load balancer handling Swift traffic must be distinct from the main load balancer used for services like Nova, Glance, and Neutron. 🔹 Key Point 2: Keystone Endpoint Behavior Swift still registers with the existing Keystone service. Depending on the configuration: - If the second_loadbalancer is enabled, Keystone publishes the VIP (Virtual IP) of the secondary HAProxy as the Swift endpoint. - If it’s disabled, Keystone uses the direct API address of the Swift proxy node. 🔹 Key Point 3: Two Supported Configurations Kolla-Ansible supports two architecture models for Swift: Normal and Advanced. - Normal Configuration This is the default setup in stable/2024.1, where all OpenStack services, including Swift, share the same load balancer and traffic path. - Advanced Configuration This is recommended for large-scale deployments with heavy data volumes. In this setup, Swift traffic goes through a dedicated load balancer and proxy nodes, which reduces contention and improves performance. We’ve attached an architectural diagram to better illustrate the setup. We’d love to share this approach with the community and welcome any feedback or suggestions. Our goal is to make managing both compute and storage services within a single deployment framework as simple as possible. Thanks in advance for your insights!
This email is a duplicate of the previous one: https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.... I apologize for the repetition. After a pause, the moderator decided to resend it with an image.
On 2025-10-07 09:21:11 -0000 (-0000), sarah.kh09@gmail.com wrote:
This email is a duplicate of the previous one: [...] I apologize for the repetition. After a pause, the moderator decided to resend it with an image.
Apologies for not spotting that you'd resent the message to the list before I approved pending messages from moderation. Note that the automated response the list software sends to tell you that your message was held (and why) also provides instructions on how to cancel that post so it will be dropped from the moderation queue, for example if you're going to send it again differently. That's the best way to avoid winding up with duplicate posts. I try to check the moderation queue at least once a day, though I don't always get to it (especially on weekends, holidays, vacations, conference travel, et cetera). We have other moderators as well, but it will still often be many hours before a held post is seen and approved. If you plan to resend the message in that situation, please remember to follow the provided instructions for cancelling the original first. -- Jeremy Stanley on behalf of the openstack-discuss moderators
Thank you for the explanation and for your help.
participants (3)
-
Jeremy Stanley
-
Sara khosravi
-
sarah.kh09ï¼ gmail.com