Automation / DevOps / IaC

HashiCorp has launched tfpolicy as a public beta feature within HCP Terraform to modernize infrastructure governance. This new framework allows teams to author policies using native HCL syntax, removing the dependency on external tools or separate languages for policy enforcement. By embedding policy creation directly into the Terraform workflow, the tool aims to streamline compliance and reduce operational friction for infrastructure teams.

• Tfpolicy is now in public beta within HCP Terraform for early adoption.
• Policies are authored in native HCL, eliminating the need for Rego or other languages.
• Enforcement is integrated directly into Terraform workflows, reducing tool sprawl.
• The framework simplifies governance by unifying policy creation and execution.
• No separate external policy engines are required for basic governance tasks.

JDK 24 eliminates the monitor-related carrier-thread pinning that previously hindered virtual thread adoption in production environments like Netflix. With this issue resolved, JDK 25 LTS shifts the primary bottleneck to downstream resource saturation, requiring explicit bounding in application code. The article details specific failure modes and provides a practical benchmark sequence to manage these new constraints.

• JDK 24 removes carrier-thread pinning, unblocking virtual thread scaling.
• JDK 25 LTS bottlenecks shift to downstream resource saturation.
• Applications must implement explicit bounding to prevent saturation.
• Review failure modes specific to post-virtual-thread adoption.
• Use provided benchmark sequence to validate resource limits.

Hugging Face highlights the critical inefficiency of idle GPU resources in modern AI infrastructure, comparing them to grounded aircraft. The blog post emphasizes that unused compute capacity represents a significant waste of capital and energy. It advocates for aggressive scheduling and utilization strategies to maximize return on investment for expensive hardware.

• Idle GPUs waste capital and energy, acting as a major operational inefficiency
• Aggressive scheduling is required to keep expensive AI hardware fully utilized
• Treating compute as a scarce resource improves overall ROI for AI deployments

AWS Lambda now supports deployment packages stored directly in customer-owned S3 buckets, which removes the per-Region code storage quota and raises the managed default limit from 75 GB to 300 GB. The per-function package size limits remain unchanged, and engineers must still use UpdateFunctionCode after replacing an object in S3. Terraform provider support for this feature is currently listed as an open enhancement request.

• Per-Region code storage quota is removed; managed default raised to 300 GB
• Per-function package size limits remain unchanged despite storage changes
• UpdateFunctionCode API call is still required after S3 object replacement
• Terraform provider support is not yet available; track as open request
• Self-managed S3 storage offers more control but requires manual code updates