Kafka Versions and Configurations
Versions
Kafka versions 3.5 and 4.1 are supported.
Node configurations
When creating a Kafka managed database cluster, you can select the node configuration — the number of vCPUs, RAM, and disk size.
Available node configurations are divided into lines. Depending on the line, configuration, and pool, different processors and volumes.
When choosing a configuration, keep in mind that a portion of the disk space is reserved for service needs and is unavailable for database placement. More details can be found in the instructions Using disk space in a Kafka cluster.
After creating a cluster, you can change the node configuration — scale the cluster.
Configuration lines
The resource ratio in configuration lines can be:
To check the availability of configuration lines across locations, see the Managed databases availability in locations. instructions.
Standard
A line of fixed configurations with a local disk and a balanced vCPU:RAM ratio, suitable for most DBMS. We recommend using this line if you do not know the workload profile.
Intel® Xeon® Scalable or AMD EPYC™ processors are used. To check processor clock speeds across different lines, see the table Processors.
To check disk performance and throughput across different lines, see the table Performance and throughput.
CPU
A line of fixed configurations with a local disk and a balanced vCPU:RAM ratio. Suitable for workload profiles that are computationally demanding. For example, if the database runs analytical queries, multiple nested queries, or data encryption. One way to determine such a workload profile is to monitor the Load Average metric, which shows the average system load over one, five, or 15 minutes.
Intel® Xeon® Scalable or AMD EPYC™ processors are used. To check processor clock speeds across different lines, see the table Processors.
To check disk performance and throughput across different lines, see the table Performance and throughput.
Memory
A line of fixed configurations with a local disk and a balanced vCPU:RAM ratio. Suitable for workload profiles that are demanding on caching. For example, if the database runs multiple, rarely repeated queries to different parts of tables. One way to determine such a workload profile is to monitor the Попадание in кэш (Cache_hit_ratio) metric, which shows the percentage of data in a query read from the cache.
Intel® Xeon® Scalable or AMD EPYC™ processors are used. To check processor clock speeds across different lines, see the table Processors.
To check disk performance and throughput across different lines, see the table Performance and throughput.
HighFreq
A line of fixed configurations with a balanced vCPU:RAM ratio.
High-performance Enterprise-grade equipment is used:
- Intel® Xeon® Gold 6354 processors (in pools ru-9, ru-2, ru-8, ke-1) or AMD EPYC 9474F (in pools ru-3, ru-7, uz-1, and uz-2). You can check processor clock speeds across different lines in the table Processors;
- RAM ECC Reg 3.2 GHz (in pools ru-9, ru-2, ru-8, ke-1) or RAM DDR5 ECC Reg 4.8 GHz (in pools ru-3, ru-7, uz-1 and uz-2);
- High-performance NVMe SSDs. To check disk performance and throughput across different lines, see the table Performance and throughput.
Dedicated
A line of fixed configurations with cluster nodes on separate cloud servers. Each cloud server occupies an entire dedicated host (physical server). Suitable for users who need physical isolation of databases from other clients, maximum performance, and maximum available resource sizes.
High-performance Enterprise-grade equipment is used:
- one Intel® Xeon® Gold 6336Y, Intel® Xeon® Gold 6240, Intel® Xeon® W-2255, or Intel® Xeon® E-2488 processor depending on the configuration. To check processor types and clock speeds across different lines, see the table Processors;
- RAM 128 GB DDR4 ECC Reg, 64 GB DDR4 ECC Reg or 32 GB DDR5 ECC Reg depending on the configuration;
- NVMe SSDs in RAID 1. The number of disks in a cluster depends on the configuration. To check disk performance and throughput across different lines, see the table Performance and throughput;
- two 2 × 25 GE network cards for the main network + MC-LAG with 25 Gbps connection speed for the service network (for backup, monitoring, and data replication in the cluster).
* DBMS clusters use Hyper-Threading Technology. Each physical core can process multiple threads in parallel — thanks to this, the total number of virtual cores (vCPU) exceeds the number of physical cores. This allows for more efficient workload distribution in multithreaded or analytical tasks.
** Part of the RAM is reserved for the physical server services.
* ** Disks are combined into RAID 1 — a mirrored disk array that provides additional fault tolerance. When using mirroring, 50% of the total disk space is available for databases. Part of the disk space is also reserved for physical server background services.
Flex
A line of custom configurations with local or network volume, in which you can choose the resource ratio.
To view disk information, see the Disks. instructions.
Intel® Xeon® Scalable or AMD EPYC™ processors are used. To check processor clock speeds across different lines, see the table Processors.
Configuration limits depend on the pool.
Local disk
Network volume
* The vCPU:RAM ratio must be at least 1:2:
- for clusters in the ru-6 pool;
- for clusters in other pools, if the configuration has more than 8 vCPUs.
For example, for 10 vCPUs, you need at least 20 GB of RAM.
** The vCPU:Local disk ratio must be:
- at least 1:32 for clusters in the ru-1 pool, if the configuration has more than 8 vCPUs. For example, for 10 vCPUs, you need a disk of at least 320 GB;
- at least 1:16 for clusters in the ru-6 pool. For example, for 10 vCPUs, you need a disk of at least 160 GB;
- at least 1:16 for clusters in all pools, except ru-1 and ru-6, if the configuration has more than 8 vCPUs.
If the configurations are not suitable, you can order a custom configuration. Create a ticket and specify the resource ratio.
Processors
Available processors differ across configuration lineups. Processor frequency affects the speed of processing user requests, executing complex algorithms, and data operations. At 100% cloud server load, the processor runs with Turbo Boost technology. Since the processor is emulated, a lower frequency will be displayed during testing.