LinkedIn Replaces Argo CD with Custom GitOps Engine to Overcome Hyperscale Kubernetes Challenges

LinkedIn has recently announced that it has replaced Argo CD with its own custom-built GitOps engine for deploying applications across its extensive Kubernetes infrastructure. This new system is responsible for managing over 4,000 services across more than 250 Kubernetes clusters, handling over 100,000 deployments weekly. The migration, which took place between July 2025 and April 2026, resulted in a significant reduction in deployment infrastructure failures, dropping from 925 to 63, despite a 3.7x increase in deployment volume. The median time to detect and resolve failed deployments also saw a dramatic improvement, falling from 16.1 minutes to 2.45 minutes. This development is highly significant for DevOps and platform engineering practitioners, particularly those operating at or aspiring to large-scale Kubernetes environments. While Argo CD has become a de-facto standard for GitOps, LinkedIn's experience reveals that its architecture presents challenges when pushed to the extreme. The core issues identified include Argo CD's single watch per group-version-resource across applications, which can lead to performance degradation (etcd churn) in clusters with low resource-type diversity and high pod counts. Additionally, limitations with Kubernetes' server-side apply, specifically the truncation of `managedFields` after 10 distinct field managers, caused issues with orphaned containers. The absence of an `observedGeneration` in Argo CD also made it difficult to precisely link sync and health status to specific deployments. This move by LinkedIn fits within a broader trend where large enterprises, facing unique scaling and operational challenges, often opt to build or heavily customize their infrastructure tooling. Companies like Uber and Airbnb have also developed their own internal deployment platforms to gain finer-grained control and optimize for their specific workloads. This isn't necessarily a condemnation of open-source tools like Argo CD, but rather an acknowledgment that at a certain scale, the trade-offs shift. The cost of maintaining hundreds of Argo CD instances and the multi-week projects for upgrades became a significant burden for LinkedIn. For practitioners, this means several things. First, it's a reminder that no single tool is a silver bullet, especially at hyperscale. While Argo CD excels for many use cases, understanding its architectural underpinnings and potential limitations is crucial for long-term planning. Second, teams operating large Kubernetes fleets should pay close attention to the specific issues LinkedIn highlighted: the number of field managers on long-lived objects and the potential for sync latency to track total pod count on shared clusters. These can be early indicators of scaling challenges. Finally, it reinforces the idea that for extreme scale, a "build vs. buy" decision often leans towards building custom solutions or extending existing ones, focusing on per-deployment isolation rather than controller tuning for scalability.