December 2025 Summaries

A Thanksgiving-day routing degradation at an upstream ISP caused 7–10% of GitHub-bound requests from a datacenter to stall for five to 20 seconds, disrupting GitHub Actions checkouts and other services such as APIs, packages, and ghcr.io for hundreds of customers. Because the affected ISP could only be identified by disabling providers individually and external routing problems can take days to resolve, the team developed a disaster-recovery mechanism that transparently redirects only GitHub traffic through a proxy network with direct GitHub peering, requiring no customer changes. Linux iptables and an ipset populated from GitHub’s published CIDR ranges efficiently identify and redirect relevant traffic, while ProxyManager recovers each packet’s original destination and sends it through HTTP CONNECT tunnels to Squid proxies. Tailscale Services provide the proxy pool with stable addressing, load balancing, health checks, WireGuard encryption, and tailnet-only access without a separate load balancer or publicly exposed proxy IPs. The system is being load tested and could later be adapted for other critical services with known IP ranges, including container registries and package repositories.

GitHub will introduce a $0.002-per-minute platform fee for all GitHub Actions usage beginning March 1, 2026, ending the previous model in which workflows run on self-hosted or third-party runners incurred no GitHub charges beyond underlying compute costs. CI expenses will therefore combine runner compute pricing with GitHub’s flat Actions fee, while GitHub-hosted runner prices are being reduced. The announcement’s analysis argues that the change allows GitHub to monetize its scheduling and workflow control plane regardless of where jobs execute, potentially positioning third-party runners as ecosystem partners rather than ways to avoid GitHub fees. Because self-hosting will retain infrastructure-management responsibilities alongside the new platform charge, reducing total workflow runtime becomes increasingly important; Blacksmith highlights faster CPUs, persistent Docker layer caching, and beta container caching as methods intended to lower CI durations and associated per-minute costs.