Fairwinds provides Managed Kubernetes-as-a-Service & open source software to enable organizations to run & optimize mission-critical Kubernetes infrastructure
USA
- Tweets 3,357
- Following 762
- Followers 1,394
- Likes 708
Kubernetes has become the backbone for production AI, but keeping cluster configs zero-drift isn't easy.
Let's swap platform engineering stories in Booth #177 at @TheAIConference in San Francisco (next week!). Book a chat with Andy: hubs.li/Q04vKJxy0
#ai #kubernetes
38
Hardware fault isolation gets cleaner in Kubernetes 1.37. With DRA device taints reaching GA, platform engineers can now isolate or drain individual degraded GPUs & accelerators directly w/o cordoning entire multi-GPU nodes. hubs.li/Q04wMr_m0
#K8s #PlatformEngineering #GPU
30
We moved production n8n on EKS from a community Helm chart to n8n’s official chart. The cutover depended on where state lived: PostgreSQL, Redis, S3, and Secrets Manager held what the main pod needed.
See how we did it (+ get an example): hubs.li/Q04ydptt0 #n8n #Kubernetes
44
Noonlight's GPU cluster made its kOps-to-EKS migration harder. Those nodes support video services whose analytics need to run as quickly as possible, so the team had to work through deployment trade-offs late in the migration.
Learn more:
bit.ly/4xI0IWu #GPU #K8s
26
Attending @TheAIConference in SF? So are we.
Drop by Booth #177 to meet the Fairwinds team. Whether you're moving past the laptop phase or optimizing enterprise Kubernetes clusters for ML models, Andy Suderman would be happy to chat.
Schedule a 1:1 here: hubs.li/Q04vKZKT0
43
Network Observability is now live in Fairwinds Insights. Click a node in Service Map, and Live Events filters to just that connection. Learn more about how to set it up and how it can help you determine the real cause of a timeout.
hubs.li/Q04s-CPX0
#K8s #Observability
31
Self-managing Kubernetes can be the right choice. But it requires a deliberate commitment to staffing, operational coverage, incident response, lifecycle management, and the ongoing work production clusters demand.
Who owns it on your team?
hubs.li/Q04xNwyV0 #Kubernetes
25
Fairwinds CTO Andy Suderman built and ran an LLM inference endpoint live at AWS Community Day Midwest, EKS deploy to a working response. Watch the replay. hubs.li/Q04qKNcV0 #AWSCommunityDay #Kubernetes #EKS
44
Scaling inference on Kubernetes comes with unique challenges—large container images, GPU allocation limits, and unpredictable cloud costs.
Want to discuss with #K8s experts? We're in Booth #177 at @TheAIConference in SF (Sept 29–Oct 1)!
Grab time w/ Andy: hubs.li/Q04vK_cg0
33
ADP talk often skips the real test: could an agent safely use your current IDP without a human watching every step? If the answer is no, the work is still platform engineering. You don't need a new tool with a new name. bit.ly/4wLqfxr
#Kubernetes #PlatformEngineering #AI
37
We're heading to @TheAIConference in San Francisco (Sept 29–Oct 1)!
If you're building or scaling AI workloads on Kubernetes & want to talk GPU efficiency, policy enforcement, or cluster rightsizing, visit Booth #177 and/or grab time w/ Andy Suderman: hubs.li/Q04vKFvF0
1
31
Getting a Kubernetes cluster into production is a milestone, but operating it over time means keeping up with releases, add-ons, nodes, networking, monitoring, and changing workload requirements.
Who owns that work in your org?
hubs.li/Q04vJc1D0
#Kubernetes #production
23
Kubernetes 1.37 extends in-place pod resizing with init container support in GA, making zero-downtime VPA more practical. Restart-free vertical scaling requires cgroup v2 on node pools as cgroup v1 defaults tighten.
hubs.li/Q04wMsdy0
#Kubernetes #DevOps #CloudNative
1
45
For Noonlight, the availability bar is explicit: an AWS availability zone failure can't cause an outage, and a failed video or alarm creation through its APIs is unacceptable.
Read how the team approached kOps-to-EKS migration: bit.ly/4yhKvH8 #Kubernetes #availability
62
EKS, AKS, and GKE reduce control-plane responsibility. They don't eliminate the wider operational work for the Kubernetes platform, including nodes, networking, add-ons, DNS, monitoring, upgrades, security.
Learn about the cost of getting K8s ops wrong: hubs.li/Q04vJpLd0
44
Think of ADP as IDP growing up. Same platform, new users. The work is identity, permissions, quotas, and observability that still make sense when agents deploy and change things at scale.
bit.ly/4w89ATp
#IDP #AIOps #CloudNative
59
How's your Network Observability? It's now available in Fairwinds Insights, so you can see who's talking to whom, identify what's leaving the cluster, and debug dependencies in Kubernetes terms: pod, namespace, + workload owner.
Learn more:
hubs.li/Q04s-G-Q0
#Observability
35
Running AI infrastructure on Kubernetes shouldn’t drain your cloud budget on idle GPUs. We're heading to @TheAIConference in San Francisco (Sept 29–Oct 1)!
Stop by Booth #177 or book time with Andy to discuss GPU rightsizing + cluster cost controls: hubs.li/Q04vKGmP0
1
1
41
Fairwinds ran a live demo at AWS Community Day Midwest: LLM inference on EKS, cluster to first response. The recording is on YouTube now. hubs.li/Q04qLw4k0 #AWSCommunityDay #Kubernetes #EKS
49
Fairwinds retweeted
From project pavilion booth discussions to maintainer-led deep dives, KubeCon + CloudNativeCon Japan showcased the power of global collaboration.
Thank you to every attendee, speaker, and maintainer who brought their energy and expertise!
#CloudNative #KubeCon
4
13
3,658