As more and more application providers transition to the cloud and deliver their services on a Software as a Service (SaaS) basis, cloud providers need to make their provisioning systems agile enough to deliver on Service Level Agreements. At the same time they should guard against over-provisioning which limits their capacity to accommodate more tenants. To this end we propose SQLR, a dynamic provisioning system employing a customized model-free reinforcement learning algorithm that is capable of reusing contextual knowledge learned from one workload to optimize resource provisioning for different workload patterns. SQLR achieves results comparable to those where resources are unconstrained, with minimal overhead. Our experiments show that we can reduce the amount of resources that need to be provisioned by almost 25% with less than 1% overall service unavailability due to blocking, and still deliver similar response times as an over-provisioned system.