Caffeine vs Varnish Cache

Caffeine is a free, open-source in-memory caching library for Java, distributed under the Apache License, Version 2.0. It provides an API inspired by Google Guava and can load entries automatically, including asynchronously. Size-based eviction uses configured limits with frequency and recency, while entries can expire according to time since their last access or write. When a stale entry is first requested, Caffeine can refresh it asynchronously. Other functions include weak keys, weak or soft values, removal notifications, writes to external resources, and cache statistics. The project offers JSR-107 JCache, Guava adapters, and a simulation extension, with community integrations listed for Spring Cache, Play Framework, Micronaut, Quarkus, and Apache Camel. The project directs users to Maven Central and provides Gradle coordinates for the core library and optional extensions. Version 3.x is intended for Java 11 or newer; earlier Java versions use 2.x. Maven Central artifacts are accompanied by GPG and Sigstore signature files according to the security policy.

Varnish Cache is a reverse caching proxy for Linux and self-hosted deployments. It serves cached responses from memory and sends other requests to origin servers. Its caching policy is defined in Varnish Configuration Language (VCL), which compiles to native code. Cached objects follow standard HTTP cache headers and leave the cache when they expire. Varnish supports TLS termination within the process and TLS connections to dynamic origin backends. It includes varnishlog for structured logging and varnishstat for metrics, and offers official container images and a Kubernetes Helm chart. Installation guidance covers Debian, Ubuntu, Red Hat Enterprise Linux, and compatible distributions. Varnish modules add functions callable from VCL. Its security guide describes controls at the command-line, CLI, VCL, and HTTP request levels. The free plan costs nothing. One limit to account for: Varnish avoids caching data when it cannot determine that doing so is safe, and applications may need configuration changes to reach a high cache hit rate.