MLOps articles

Kubernetes service mesh for ML platforms

Decide whether a service mesh fits your ML platform, then design traffic policy, identity, observability, rollout, and failure behavior deliberately.

Run Polyaxon AI workloads from GitHub Actions

Use GitHub Actions to validate and submit traceable AI, ML, and agent workloads to Polyaxon without turning CI runners into training infrastructure.

Single vs. multi-cluster Kubernetes for ML platforms

Choose between one Kubernetes cluster and multiple clusters using isolation, failure domains, data locality, accelerator access, operations, and cost.

Running Redis on Kubernetes for ML platforms

Design Redis on Kubernetes around workload semantics, persistence, topology, memory, security, recovery, and observable operation for ML systems.

When to migrate ML workloads to Kubernetes

Decide whether Kubernetes fits your ML platform, then migrate workload contracts, storage, scheduling, security, and operations in controlled stages.

Data-centric AI and MLOps solve different problems

Data-centric AI can improve model quality, but it does not replace the operational discipline needed to run machine learning systems.

Right-size Kubernetes resources for ML workloads

Set CPU, memory, ephemeral-storage, and GPU resources from measured ML workload behavior while preserving scheduling efficiency and reliability.

Troubleshoot OOMKilled in Kubernetes ML workloads

Diagnose container memory limits, node pressure, application allocation, and ML data-loading behavior before changing Kubernetes resources.

Understand Kubernetes Pod evictions for ML

Distinguish node-pressure, API-initiated, preemption, and node-failure disruptions, then design ML workloads to recover safely.

A fuller picture of model behavior during training

Metrics alone do not explain model behavior. Teams need artifacts, samples, images, logs, and lineage tied to each training run.

Manage Kubernetes Secrets for ML workloads

Protect ML credentials with encryption, least-privilege access, workload identity, controlled delivery, rotation, and Polyaxon connections.

PromQL cheat sheet for Kubernetes ML platforms

Use practical PromQL patterns for Kubernetes capacity, workload reliability, latency, and ML operations while controlling cardinality.

Should an ML platform run databases on Kubernetes?

Evaluate database placement for ML platforms across operational ownership, storage, availability, recovery, upgrades, security, and performance.

Use Kubernetes labels and annotations for ML

Design labels, selectors, and annotations that connect Kubernetes resources to ML ownership and operations without breaking controllers or metrics.

Kubernetes alerting practices for ML platforms

Design actionable Kubernetes alerts for ML services, batch operations, shared capacity, and the monitoring pipeline itself.

What Kubernetes means for ML teams

Learn the Kubernetes control model, core workload and infrastructure objects, and the capabilities an ML platform must add above the cluster.

Five shifts shaping enterprise AI platforms

Plan enterprise AI platforms around measurable outcomes, heterogeneous compute, durable agents, cost per successful task, and continuous governance.

Monitor Kubernetes ML workloads with Prometheus

Build a useful Prometheus monitoring model across Kubernetes objects, nodes, containers, applications, and ML operations without uncontrolled cardinality.

Optimize multi-cloud costs for ML workloads

Control multi-cloud ML spending with normalized allocation, workload unit economics, placement policies, elastic capacity, and reproducible efficiency measurements.

Kubernetes monitoring for ML workloads

Monitor Kubernetes control planes, nodes, containers, schedulers, applications, and ML outcomes with useful correlations and controlled cardinality.

Azure Storage best practices for ML data and artifacts

Design Azure Storage accounts for ML datasets and artifacts with clear boundaries, workload identities, network controls, lifecycle rules, and recovery tests.

Datadog vs. AppDynamics for ML platform monitoring

Evaluate Datadog and AppDynamics against application transactions, Kubernetes infrastructure, ML workflows, telemetry governance, and operating cost.

Datadog vs. New Relic for ML platform observability

Compare Datadog and New Relic for Kubernetes-based ML workloads using telemetry coverage, workflow context, investigation speed, governance, and cost.

Experiment tracking in machine learning

Building machine learning models is an experimental process that requires several iterations. In this blog post we go over how Polyaxon manages experiment tracking.