Apache Kafka vs Memphis
This section describes the differences between Apache Kafka and Memphis
Apache Kafka is an open-source distributed event streaming platform. Based on the abstraction of a distributed commit log, Kafka can handle a great number of events with functionality comprising pub/sub.
Memphis is a next-generation message broker.
A simple, robust, and durable cloud-native message broker wrapped with an entire ecosystem that enables fast and reliable development of next-generation event-driven use cases.
Memphis.dev enables building next-generation applications that require large volumes of streamed and enriched data, modern protocols, zero ops, rapid development, extreme cost reduction, and a significantly lower amount of dev time for data-oriented developers and data engineers.
Memphis + PostgreSQL
Kafka + Zookeeper(ZK is being removed)
Message consumption model
Both technologies are available under fully open-source licenses. Memphis also has a commercial distribution with added security, tiered storage, and more.
Kafka uses Apache Zookeeper™ for consensus and message storage. Memphis uses PostgreSQL for GUI state management only and will be removed soon, making Memphis without any external dependency. Memphis achieves consensus by using RAFT.
Both Kafka and Memphis use a pull-based architecture where consumers pull messages from the server, and long-polling is used to ensure new messages are made available instantaneously.
Pull-based architectures are often preferable for high throughput workloads as they allow consumers to manage their flow control, fetching only what they need.
Kafka uses a distributed commit log as its storage layer. Writes are appended to the end of the log. Reads are sequential, starting from an offset, and data is zero-copied from the disk buffer to the network buffer. This works well for event streaming use cases.
Memphis also uses a distributed commit log called streams (made by NATS Jetstream) as its storage layer, which can be written entirely on the broker's (server) memory or disk. Memphis also uses offsets but abstracts them completely, so the heavy lifting of saving a record of the used offsets resides on Memphis and not on the client. Memphis also offers storage tiering for offloading messages to S3-compatible storage for an infinite storage time and more cost-effective storage. Reads are sequential.
Requires deep understanding and design
3rd parties like Confluent, AWS MSK
Managed cloud offerings
3rd parties like Confluent, AWS MSK
Message tracing (aka Stream lineage)
Automatic based on policy
Kafka is a cluster-based technology with a medium-weight architecture requiring two distributed components: Kafka's own servers (brokers) plus ZooKeeper™ servers. Zookeeper adds an additional level of complexity but the community is in the process of removing the ZooKeeper component from Kafka. Kafka "Vanilla" deployment requires a manual binary installation and text-based configuration, as well as config OS daemons and internal parameters.
Memphis has a light-weight yet robust cloud-native architecture and packed as a container from day one. It can be deployed using any docker engine, docker swarm, and for production environment using helm for Kubernetes (soon with operator). Memphis initial config is already sufficient for production, and optimizations can take place on-the-fly without downtime. That approach enables Memphis to be completely isolated and apart from the infrastructure it deployed upon.
Enterprise-grade support and managed cloud offerings for Kafka are available from several prominent vendors, including Confluent, AWS (MSK), Cloudera, and more.
Memphis provides enterprise support and managed cloud offering that includes features like enhanced security, stream research abilities, an ML-based resource scheduler for better cost savings, and more.
Kafka is a robust distributed system and requires constant tune-ups, client-made wrappers, management, and tight monitoring. The user or operator is responsible for ensuring it's alive and works as required. This approach has pros and cons, as the user can tune almost every parameter, which is often revealed as a significant burden.
One of Memphis' core features is to remove frictions of management and autonomously make sure it's alive and performing well using periodic self-checks and proactive rebalancing tasks, as well as fencing the users from misusing the system. In parallel, every aspect of the system can be configured on-the-fly without downtime.
Memphis has a built-in notification center that can push real-time alerts based on defined triggers like client disconnections, resource depletion, schema violation, and more.
Apache Kafka does not offer an embedded solution for notifications. Can be achieved via commercial offerings.
Tracking stream lineage is the ability to understand the full path of a message from the very first producer through the final consumer, including the trail and evolvement of a message between topics. This ability is extremely handy in a troubleshooting process.
Apache Kafka does not provide a native ability for stream lineage, but it can be achieved using OpenTelemetry or OpenLineage frameworks, as well as integrating 3rd party applications such as datadog, epsagon, or using Confluent's cloud offering.
Memphis provides stream lineage per message with out-of-the-box visualization for each stamped message using a generated header by the Memphis SDK.
Consumer group level
No. In progress (KIP-405)
At least once, Exactly once
At least once, Exactly once
Kafka Replication means having multiple copies of the data spread across multiple servers/brokers. This helps maintain high availability if one of the brokers goes down and is unavailable to serve the requests.
Memphis station replication works similarly. During station (=topic) creation, the user can choose the number of replicas derived from the number of available brokers. Messages will be replicated in a RAID-1 manner across the chosen number of brokers.
Multi-tenancy refers to the mode of operation of software where multiple independent instances of one or multiple applications operate in a shared environment. The instances (tenants) are logically isolated and often physically integrated. The most famous users are SaaS-type applications.
Apache Kafka does not natively support multi-tenancy. It can be achieved via complex client logic, different topics, and ACL.
As Memphis pushes to enable the next generation of applications and especially SaaS-type architectures, Memphis supports multi-tenancy across all the layers from stations (=topics) to security, consumers, and producers, all the way to node selection for complete hardware isolation in case of need. It is enabled using namespaces and can be managed in a unified console.
Memphis offers a multi-tier storage strategy in its open-source version. Memphis will write messages that reached their end of 1st retention policy to a 2nd retention policy on object storage like S3 for longer retention time, potentially infinite, and post-streaming analysis. This feature can significantly help with cost reduction and stream auditing.
Both Kafka and Memphis store data durably and reliably, much like a normal database. Data retention is user configurable per Memphis station or Kafka topic.
Both Kafka and Memphis provide default support in idempotent producers. On the consumer side, in Kafka, it׳s the client's responsibility to build a retry mechanism that will retransmit a batch of messages exactly once, while in Memphis, it is provided natively within the SDK with a parameter called
Common scenarios for a geo-replication include:
- Disaster recovery
- Feeding edge clusters into a central, aggregate cluster
- Physical isolation of clusters (such as production vs. testing)
- Cloud migration or hybrid cloud deployments
- Legal and compliance requirements
Kafka users can set up such inter-cluster data flows with Kafka's MirrorMaker (version 2), a tool to replicate data between different Kafka environments in a streaming manner.
Memphis cloud users can create more Memphis clusters and form a supercluster that replicates data in an async manner between the clusters of streamed data, security, consumer groups, unified management, and more.
Yes. Using Kafka connect and KStreams
Message replay, time travel
SQL and Serverless functions
SQL-based using KStreams
Pull retry mechanism
Multiple open-source GUIs have been developed for Kafka over the years, for example, Kafka-UI. Usually, it cannot sustain heavy traffic and visualization and requires separate computing and maintenance. There are different commercial versions of Kafka that, among the rest, provide robust GUI, like Confluent, Conduktor, and more.
Memphis provides a native state-of-the-art GUI, hosted inside the broker, built to act as a management layer of all Memphis aspects, including cluster config, resources, data observability, notifications, processing, and more.
Dead-letter queue is both a concept and a solution that is useful for debugging clients because it lets you isolate and "recycle" instead of drop unconsumed messages to determine why their processing doesn't succeed.
The Kafka architecture does not support DLQ within the broker; it is the client or consumer's responsibility to implement such behavior for good and bad.
One of Memphis' core building blocks is avoiding unexpected data loss, enabling rapid development, and shortening troubleshooting cycles. Therefore, Memphis provides a native solution for dead-letter that acts as the station recycle bin for various failures such as unacknowledged messages, schema violations, and custom exceptions.
The very basic building block to control and ensure the quality of data that flows through your organization between the different owners is by defining well-written schemas and data models.
Confluent offers "Schema Registry" which is a standalone component and provides a centralized repository for schemas and metadata, allowing services to flexibly interact and exchange data with each other without the challenge of managing and sharing schemas between them. It requires dedicated management, maintenance, scale, and monitoring.
As part of its open-source version, Memphis presents Schemaverse, which is also embedded within the broker. Schemaverse provides a robust schema store and schema management layer on top of memphis broker without a standalone compute or dedicated resources. With a unique and modern UI and programmatic approach, technical and non-technical users can create and define different schemas, attach the schema to multiple stations and choose if the schema should be enforced or not. In counter to Schema Registry, the client does not need to implement serialization functions, and every schema update takes place during producers' runtime.
Kafka provides routing capabilities through Kafka Connect and Kafka Streams, including content-based routing, message transformation, and message enrichment.
Memphis message routing is similar to the implementation of RabbitMQ using routing keys, wildcards, content-based routing, and more. Similar to RabbitMQ, it is also embedded within the broker and does not require external libraries or tools.
Compaction has been created to support a long-term, potentially infinite record store based on specific keys.
Kafka supports native topic compaction, which runs on all brokers. This runs automatically for compacted topics, condensing the log down to the latest version of messages sharing the same key.
At the moment, Memphis does not support compaction, but it will in the future.
The ability to re-consume committed messages.
Kafka does support replay by seeking specific offsets as the consumers have control over resetting the offset.
Memphis does not support replay yet but will in the near future (2023).
Kafka, with its Kafka Streams library, allows developers to implement elastic and scalable client applications that can leverage essential stream processing features such as tables, joins, and aggregations of several topics, and export to multiple sources via Kafka connect.
Memphis provides a similar behavior and more. Embedded inside the broker, Memphis users can create serverless-type functions or complete containerized applications that aggregate several stations and streams, decorate and enrich messages from different sources, write complex functions that cannot be achieved via SQL, and manipulate the schema. Memphis embedded connectors frameworks will help to push the results directly to a defined sink.
In case of a failure or lack of ability to acknowledge consumed messages, there should be a retry mechanism that will retry to pull the same offset or batch of messages.
In Kafka, it is the client's responsibility to implement one. Some key factors must be considered to implement such a mechanism, like blocking vs non-blocking, offset tracking, idempotency, and more.
In Memphis, the retry mechanism is built-in and turned on by default within the SDK and broker. During consumer creation, the parameter
maxMsgDeliverieswill determine the number of retries the station will deliver a message if an acknowledgment does not arrive till
maxAckTimeMs. The broker itself records the offsets given and will expose only the unacknowledged ones to the retry request.
Apache Kafka is a robust and mature system with extensive community support and a proven record for high-throughput use cases and users. Still, it also requires micro-management, troubleshooting can take precious time, complex client implementations, wrappers, and often tunings that take the user's focus and resources from the "main event," and the most important thing - it does not scale well as the organization grows, and more use cases join in. Wix Greyhound library is an excellent proof for the needed work on top of Kafka.
We call Memphis "A next-generation message broker" because it leans towards the user and adapts to its scale and requirements, not the opposite. Most of the wrappers, tunings, management-overhead, and implementations needed from the client in Kafka, are abstract to the users in Memphis, which provides an excellent solution for both the smaller workload use cases and the more robust ones under the same system and with full ecosystem to support it. It has a milage to pass, but the immediate benefits already exist and will continue to evolve.