The invention discloses a big data platform auto scaling method based on service discovery and a container technology. The method comprises the following steps of (1) componentizing a big data platform by using the container technology; (2) starting a big data cluster, and registering metadata information of the cluster; (3) reporting heartbeat data to a service agent and updating related information; (4) reading big data cluster management information for the service agent by an agency service, determining whether a node fails or requirements are changed, if yes, executing a fifth step, if not, executing a sixth step; (5) if the node fails, trying to recover a failing node container; and if requirements are changed, adding or deleting the node container for the cluster according to metadata needed to be changed; and (6) repeating the third to fifth steps till the cluster service stops running. The internal state of the big data platform can be sensed, so that effective auto scaling isperformed and the cluster resources are used more efficiently.