blog.confluent.io
Using logs to build a solid data infrastructure (or: why dual writes are a bad idea)
Drawing from the experience of building scalable systems at LinkedIn and other startups, this talk explores why logs are such a fine idea: making it easier to maintain search indexes and caches, making your applications more scalable and more robust in the face of failures, and opening up your data for richer analysis, while avoiding race conditions, inconsistencies and other ugly problems.