Upgrading Elasticsearch without downtime was a challenging task at Intercom, but the company's focus on shipping great product at high velocity and empowering product teams to utilize their infrastructure services drove the decision to centralize this effort. With 54 releases between versions 2.3.3 and 6.3.0, the upgrade required careful planning and execution. The team used a two-step process, first upgrading from 2.3.3 to 5.6.9 and then from 5.6.9 to 6.3.0, which involved setting up dual writing and reading, taking snapshots, and restoring them into new temporary indexes. After verifying the stability of the new cluster, the team switched over for real and turned off dual writing, deleting all documents with the "Deleted" field set to true. The upgrade resulted in significant performance improvements, including a 50% reduction in average indexing and search latency, a 40% reduction in average CPU usage, faster restarts and recoveries with sequence IDs, and massive reductions in disk usage and costs. Today, Intercom has 10 Elasticsearch clusters running the latest version of Elasticsearch, still owned by individual teams but with a single team for cross-cutting concerns like major version upgrades.