Architecture & scope
Architecture
Cloudera
Google
Cloudera edge
Cloudera Kafka
Where it runs
VMs in your own VPC on AWS, Azure or GCP, provisioned as a Data Hub cluster. You keep access to the nodes.
Kafka flavor
Cloudera-patched open source Kafka, shipped on the Cloudera Runtime release train. You decide when to upgrade.
Cluster shapes
Light Duty and Heavy Duty definitions, plus a Streams Messaging High Availability template for multi-AZ deployments.
Bundled components
Streams Messaging Manager (SMM), Streams Replication Manager (SRM), Schema Registry and Cruise Control.
Portability
The same stack runs on-prem (Cloudera Base), in public cloud, or on Kubernetes through the Streaming Operator.
Google Managed Kafka
Where it runs
Google-owned infrastructure, reached from your VPC through Private Service Connect. No broker VMs are visible to you.
Kafka flavor
Open source Apache Kafka 3.7.1 in KRaft mode, so application code migrates without changes. Security patches apply automatically; new minor versions do not.
Cluster shapes
One shape: three-zone high availability is the default, and every cluster must have equal resources in each zone.
Bundled components
Kafka brokers, Schema Registry (Preview; Avro and Protobuf only, no JSON) and Kafka Connect clusters.
Portability
GCP only. There is no hybrid or on-prem option.
Verdict
Cloudera has the edge on portability and bundled tooling. Google wins when you never need to leave GCP and don't want to think about brokers at all.
Cost model
Cost
Cloudera
Google
Google edge
Cloudera Kafka
| Component | Price |
|---|---|
| Data Hub | $0.04 / CCU-hour A CCU combines cores and memory |
| Cloud infrastructure | Not included Compute, storage, networking and egress are billed by your cloud provider |
| Streams Messaging (standalone) | Contact sales |
| Streaming Kubernetes Operator | Contact sales |
| Idle clusters | Cost the same as busy ones; there is no serverless mode |
| Bills you receive | Two: Cloudera and your cloud provider |
Google Managed Kafka (us-central1)
| Component | Price |
|---|---|
| Compute | $0.09 / DCU-hour 1 DCU is about 1 vCPU + 4 GiB RAM |
| Kafka Connect compute | $0.12 / DCU-hour Committed use discounts do not apply |
| Local storage | ~$0.000233 / GiB-hour 100 GB billed per vCPU regardless of use |
| Tiered storage | ~$0.000137 / GiB-hour |
| Inter-zone replication | $0.01 / GiB |
| Private Service Connect transfer | $0.004–$0.01 / GiB |
| Committed use discounts | 20% for 1 year, 40% for 3 years |
| Google's 100 MiB/s example | ~$11K/month managed vs ~$9.1K self-managed |
Verdict
Google has the edge on predictability. Cloudera's $0.04/CCU-hour sits on top of a full infrastructure bill, so total landed cost is usually comparable to or higher than Google's, and you also carry the operations effort. Cloudera wins on cost mainly when the cluster amortizes an existing CDP footprint.
Management & operations
Management
Cloudera
Google
Google edge
Cloudera Kafka
Operations
Cloudera Manager plus SMM. You own upgrade scheduling, runtime version pinning, node scaling and OS-level exposure.
Monitoring
SMM visualizes flow from producer to topic to consumer, filters across brokers, topics, producers and consumer groups, and exposes REST endpoints for APM or ticketing integration.
Rebalancing
Cruise Control.
Cross-cluster replication
Streams Replication Manager (SRM) is built in.
Version control
You can pin a version when a certified downstream application requires it.
Google Managed Kafka
Operations
Automatic broker sizing and rebalancing, with security patches applied for you. Cloud Monitoring, Cloud Logging and IAM work out of the box. Local disk size and many broker configs can't be changed.
Tooling
Terraform resources for clusters, topics and Connect (google_managed_kafka_cluster, _topic).
Recent additions
Kafka Connect GA (Oct 2025), remote MCP server for LLM-driven cluster management GA (Apr 2026), Debezium-based Cloud SQL and PostgreSQL CDC source connectors (Jul 2026).
Cross-cluster replication
Built-in MirrorMaker 2.0 Source, Checkpoint and Heartbeat connectors run on managed Connect clusters, billed per DCU.
Version control
Brokers run Kafka 3.7.1 and minor upgrades aren't automatic, but you don't choose the version or when patches restart brokers.
Verdict
Google has the edge for teams that want zero Kafka operations. Cloudera has the edge when you need to pin versions, tune brokers and disks, or want replication management through SRM. Google also offers MirrorMaker 2.0 connectors, so replication is not a gap.
Lineage & governance
Lineage
Cloudera
Google
Cloudera dominant
Cloudera Kafka
Catalog
Apache Atlas ships as part of SDX. Services publish activity metadata to the ATLAS_HOOK Kafka topic, and Atlas creates kafka_topic entities and resolves lineage between them.
Multi-hop view
SMM integrates with Atlas to show lineage across multiple Kafka hops, producers and consumers.
End-to-end
NiFi, Flink, Spark and Hive emit to the same catalog, giving one graph from edge to warehouse.
Extra
Cloudera Data Lineage (formerly Octopai) is a separate, contact-sales offering.
Google Managed Kafka
Kafka lineage
No first-class support. Dataplex Universal Catalog tracks lineage for BigQuery, Dataflow and Composer, but Kafka topics are not catalog entities.
Practical options
Stop the lineage graph at the Kafka boundary, reconstruct it from Connect and Dataflow job metadata, or add a third-party catalog such as DataHub, OpenMetadata or Collibra.
Schema governance
Schema Registry is in Preview, supports Avro and Protobuf only, and implements the Confluent Schema Registry REST API for client compatibility.
Verdict
Cloudera dominates here. If regulated lineage across streaming hops is a hard requirement, this is usually the deciding factor.
Authentication & authorization
Authentication
Cloudera
Google
Split
Cloudera Kafka
Authentication
Kerberos via FreeIPA in the CDP environment, workload username and password over SASL_SSL, and mTLS. LDAP and SAML single sign-on at the control plane, with Knox as the perimeter gateway.
Authorization
Apache Ranger policies on topics, consumer groups and transactional IDs.
Tag-based policies
Atlas classifications drive Ranger policies. A topic classified as PII can be restricted to one group without editing ACLs.
Client setup
Heavier: clients add the FreeIPA certificate to their truststore and set a separate workload password.
Google Managed Kafka
Authentication
Google IAM with service account credentials and OAuth tokens (SASL_SSL, OAUTHBEARER). mTLS certificate auth went GA in Sep 2025 for clusters created after June 24, 2025.
Encryption in transit
TLS is mandatory on every connection, with broker certificates signed by a public CA. Unauthenticated access is not allowed, and Kerberos (GSSAPI) and SCRAM are not supported.
Authorization
IAM roles plus standard Kafka ACLs, configurable through REST, gRPC and gcloud (GA May 2025).
Perimeter
VPC Service Controls (GA Sep 2025), Organization Policy custom constraints, and HIPAA compliance support.
Client setup
Simple: IAM integrates with the rest of GCP.
Verdict
Google wins on simplicity and native integration with other GCP services. Cloudera wins on expressiveness: Ranger's tag-driven policies handle rules like "any PII-classified topic is readable only by group X", which IAM plus ACLs cannot express natively.
Support & SLA
Support
Cloudera
Google
Cloudera edge
Cloudera Kafka
Model
Enterprise technical support, plus version, maintenance and security updates, included in the subscription.
Expertise
Engineers who know Kafka and the Cloudera stack specifically, with named-account escalation and professional services available.
SLA
Set by contract. Confirm uptime terms with Cloudera.
Catch
Support is tied to the Runtime version lifecycle. Falling behind on upgrades erodes your support position.
Google Managed Kafka
Model
Standard Google Cloud support tiers (Standard, Enhanced, Premium), priced separately as a percentage of spend.
Expertise
Generalist cloud support rather than Kafka specialists.
SLA
Managed Kafka API uptime SLO of at least 99.95%
Catch
The covered-service definition excludes topic and consumer group operations in a specific cluster. It is narrower than it first appears.
Verdict
Cloudera has the edge on Kafka-specific depth. Read Google's SLA carve-out carefully before relying on the 99.95% figure.
Decision guide
Which one fits?
Choose Cloudera Kafka if
→
You need hybrid or multi-cloud with one stack across on-prem, AWS, Azure and GCP.
→
You have regulatory lineage requirements across NiFi, Kafka, Flink and the warehouse.
→
You need Ranger's tag-based authorization.
→
You already run SDX, so another Data Hub is a small marginal cost.
→
You need version pinning, broker and disk tuning, or Kerberos authentication.
Choose Google Managed Kafka if
→
Your consumers are BigQuery, Dataflow or Pub/Sub.
→
You want minimal operations headcount for Kafka.
→
You're comfortable with Google managing patching and a fixed Kafka version.
→
Lineage past the topic boundary isn't a compliance requirement, or you'll add a third-party catalog.
→
You want IAM-based security without Kerberos or truststore management.
Watch out for
!
Google: Schema Registry was still in Preview when this was researched, and it supports Avro and Protobuf but not JSON.
!
Google: the 99.95% SLO excludes topic and consumer group operations in a specific cluster.
!
Cloudera: $0.04/CCU-hour excludes infrastructure, and idle clusters cost the same as busy ones.
!
Cloudera: falling behind on Runtime upgrades erodes support eligibility.