Amounts net, without term discount and without region factor. Locations: 5.
An analysis is never a product, but a pipeline of four steps. A part of the bill is generated at each one, and a separate clock runs at each one. If you know both, you know where your money is going before choosing a product.
The event originates in your application and must be received before anyone analyses it. Kafka, the managed service for event streams, receives it and stores it three times — once on each of the three brokers, the machines of the cluster.
What it costs
from €353.03$409.51 / month
Stream S with 3 brokers. The amount depends on the broker class, not the number of events.
How long it takes
5.7 days
This is how long an event remains retrievable in the stream: 480 GB usable storage, divided by 84.4 GB per day at 1 MB/s continuous load.
The raw data goes unchanged into the Object Storage and remains there. This zone is never overwritten — it is the only place from which an incorrect transformation rule can be corrected later.
What it costs
€0.0070$0.0081 / GB per month
€7.17$8.32 per terabyte a month, €86.04$99.81 a year. NVMe block storage costs 12 times as much for the same amount.
How long it takes
indefinite
Storage does not end by itself, but only when a cleanup rule applies. That is exactly its purpose — the rest of the chain can be rolled out again from here.
Check, transform, write column by column. What does not fit the stored schema goes into a separate storage with an error reason, instead of aborting the entire run.
What it costs
from €229.70$266.45 / month
€65.70$76.21 control plane per environment, plus €82.00$95.12 per processing node. Platform S runs 2 nodes.
How long it takes
Hourly schedule
The entry-level class is designed for scheduled runs at hourly intervals. If you compute more frequently or in parallel, you add nodes — each additional one costs €82.00$95.12 a month.
Only here is the metric created. ClickHouse, the columnar database, computes across columns. OpenSearch, the full-text search, searches in the text. The dashboards show both in one view.
What it costs
€38.71$44.90 / TB raw data
Analytics M costs €189.00$219.24 a month and holds around 4.9 TB of raw data at 10:1. It is not the disk size that determines the price, but the compression.
How long it takes
12.5 days
This is how far back the entry-level full-text search goes: 300 GB usable index with 20 GB of raw logs a day and an index overhead of 1.2.
Ingestion, validation, transformation and storage run as a managed pipeline. The raw data remains unchanged in Object Storage — the object storage billed by occupied space. Any subsequent error can be rolled back from there without querying the source again.
The pipeline reads directly from a managed Kafka cluster and remembers the offset per task — the reading position in the stream. Upon restart, reading continues from the last confirmed offset, not from the beginning.
Kafka protocol, SASL/SCRAM
New objects in a bucket — an Object Storage container — trigger a run. The trigger comes from the storage notification, not from a polling interval; this prevents empty runs.
S3 API, event notification
Changes are captured from managed PostgreSQL and MySQL via the transaction log. The source database receives a read connection for this, not additional query load.
logical replication, binlog
For events from applications that neither speak Kafka nor write to a database. The endpoint accepts JSON lines and stores them in a buffer.
HTTPS, token per pipeline
What arrives is read. Writing is done column by column — anything else makes the subsequent query more expensive than the one-time transformation costs.
The class only determines the number of processing nodes. Control plane and storage are the same in every class; a change alters neither pipelines nor data.
| Class | Nodes | total vCPU | Price / month |
|---|---|---|---|
| Platform S | 2 | 16 | €229.70$266.45 |
| Platform M | 6 | 48 | €557.70$646.93 |
| Platform L | 16 | 128 | €1,377.70$1,598.13 |
Net amounts plus 19% VAT. Each row is the sum of control plane and node count: €65.70$76.21 plus nodes times €82.00$95.12. Storage is billed by consumption and is not included.
One terabyte of raw data is stored at ENTRONYX CLOUD for €7.17$8.32 per month in Object Storage. On NVMe block storage, the same amount costs 12 times as much — which is why the raw zone is located there and not next to the database.
Raw zone · €0.0070$0.0081 per GB and month
Three brokers — the machines that hold the message stream. The cluster manages itself in KRaft mode, without the previously required ZooKeeper service. Schema Registry (the repository of message formats) and Connect runtime (the connection to sources and destinations) run in the same cluster. Consumers read at their own pace; a slow consumer slows down neither the producer nor the others.
| Class | Broker | vCPU / RAM per broker | Storage per broker | Continuous loadMB/s | PeakMB/s | Price / month |
|---|---|---|---|---|---|---|
| Stream S | 3 | 2 / 4 GB | 480 GB | 18 | 45 | €353.03$409.51 |
| Stream M | 3 | 4 / 15 GB | 1.9 TB | 74 | 160 | €1,329.99$1,542.79 |
Measurement condition: kafka-producer-perf-test, 1 KB per message, acks=all, replication factor 3, compression lz4, one topic with 24 partitions. net amounts plus 19% VAT.
With a replication factor of 3 — each message is kept 3 times — it is stored on every broker. This means the storage of a single broker is usable. How long it lasts depends on the continuous load: a data stream of 1 MB/s writes 84.4 GB a day.
| Class | Usable | at 1 MB/s84.4 GB/day | at 10 MB/s843.8 GB/day |
|---|---|---|---|
| Stream S | 480 GB | 5.7 days | 13.7 hours |
| Stream M | 1.9 TB | 22.8 days | 2.3 days |
Without offloading. If you want to retain data longer, attach Block Storage NVMe (€0.0860$0.0998 per GB a month) or enable offloading to Object Storage (€0.0070$0.0081 per GB a month). The difference between the two rates is the reason why long-term retention ends up in object storage.
Avro, Protobuf and JSON Schema are versioned per topic. The registry rejects a new version if it violates the configured compatibility rule — backward compatibility is the default.
The registry runs on the same nodes as the brokers and is accessible via the same private address. A separate external endpoint is not opened.
Confluent-compatible, included in the price
Older segments move to Object Storage and remain queryable there. The broker only keeps the hot part on NVMe; retention is therefore no longer tied to disk size.
The object storage price applies to the offloaded segments, not the block storage price. This difference is the reason why long-term retention remains affordable at all.
Tiered storage to Object Storage
Source and target connectors run as separate tasks. If one fails, it is restarted on another node without stopping the others.
Connectors for Object Storage, PostgreSQL, MySQL, ClickHouse and OpenSearch are included. Custom connectors can be uploaded as a JAR file (Java package).
distributed, with restart per task
A sum over a column reads exactly that column. A row-oriented database reads every row completely and discards the superfluous part. With forty columns, this is not fine-tuning, but an order of magnitude.
A sum over one column reads only that one column. A row-oriented database reads every row completely and discards the rest — with 40 columns, this is the factor, not the fine-tuning.
The data is stored in blocks sorted by time. A one-hour window touches the blocks of that hour; the rest of the table is not touched.
Approximation methods for quantiles — the threshold below which a certain proportion of all values lies — and for unique values are built in and run across the column blocks. An exact count over a billion rows takes factors longer than the approximation.
Modifying individual rows, checking foreign keys, many short writes per second. PostgreSQL remains the right choice for this — the section on differentiation explains this.
The compression rate depends on the data, not the service. The range observed in operations is given; calculations on this page consistently use the lower value. The decisive factor here is the cardinality of a column — the number of its distinct values. The smaller it is, the more the column can be packed: a column with twelve status values shrinks significantly more than one with free text.
| Data type | Method | Range | Comment |
|---|---|---|---|
| Time series with key metrics | Delta + ZSTD | 12:1 to 24:1 | timestamps and counters at regular intervals |
| Application events | ZSTD | 8:1 to 14:1 | recurring field names and identifiers |
| Log texts | ZSTD | 4:1 to 6:1 | free text with low repetition |
| Identifiers with few values | LowCardinality + LZ4 | 20:1 to 60:1 | status values, country codes, device types |
Measured as the ratio of uncompressed to compressed size on disk, across tables from one billion rows.
| Class | Nodes | vCPU / RAM | Storage | Raw dataat 10:1 | per TB of raw data | Price / month |
|---|---|---|---|---|---|---|
| Analytics M | 1 | 8 / 32 GB | 500 GB | 4.9 TB | €38.71$44.90 | €189.00$219.24 |
| Analytics L | 3 | 16 / 64 GB | 2.0 TB | 19.5 TB | €28.11$32.61 | €549.00$636.84 |
Net amounts plus 19% VAT. The column “per TB of raw data” is the monthly price divided by the amount of raw data that fits into the maintained storage at 10:1. The classes themselves are taken unchanged from the product catalogue and appear with the same amount under Databases.
Logs, documents and events are parsed and indexed as they are written. A search across a hundred million lines therefore answers in milliseconds the same question that a scan through files takes minutes to answer.
| Class | Nodes | vCPU / RAM per node | Index per node | Usable2 copies | Retentionat 20 GB/day | Price / month |
|---|---|---|---|---|---|---|
| Search S | 3 | 2 / 8 GB | 200 GB | 300 GB | 12.5 days | €169.00$196.04 |
| Search M | 3 | 4 / 16 GB | 600 GB | 900 GB | 37.5 days | €409.00$474.44 |
| Search L | 3 | 8 / 32 GB | 2.0 TB | 2.9 TB | 125 days | €1,019.00$1,182.04 |
Net amounts plus 19% VAT. Each class runs with 3 nodes because the cluster management needs an odd number for its quorum. Half of the gross storage is usable because each document has a replica. The index overhead of 1.2 compared to the raw data is already included in the “Retention” column: 20 GB of raw logs take up 24 GB of index per day.
OpenSearch is not yet maintained as a separate plan in the product catalogue. The amounts above are therefore calculated, not set: instance price plus block storage for the index, times management markup, times number of nodes. The markup of 1.17 is not a free value itself. It is calculated backwards from the maintained ClickHouse plan: the same raw configuration there, against the maintained price there.
Indices are rolled daily: a new index starts for each day. After the set period, they move to slower nodes, then to Object Storage, and finally they are deleted. Each step is stored as a rule, not as a task in the calendar.
The retention period applies per index pattern. Access logs with 14 days and audit logs with 400 days are therefore in the same cluster.
hot, warm, cold, deleted
Analysed text fields and exact keyword fields lie next to each other. A search for an error message can thus be restricted to a tenant and a time window in the same expression.
German stemming and compound decomposition are preconfigured. Custom dictionaries and synonym lists can be uploaded.
one query across both
The cluster comes with its own interface for search and log analysis — discover, saved queries, visualisations across index patterns.
The service in the next section is intended for metrics from multiple sources. OpenSearch Dashboards only sees its own indices.
OpenSearch Dashboards, included in the price
Grafana — the popular open interface for metric views — as a managed service, with the platform's data sources already configured. We handle updates, backup of the views and recovery; the views themselves remain exportable as JSON and thus versionable.
Configured and provided with access credentials as soon as the respective service runs in the same account. A view may show panels from multiple sources.
Four ways to share a view. Each has a different limit — the limit is stated there, not just the way.
Folders carry the permissions, not the individual view. Anyone with folder permissions sees all views within it — this keeps permission management manageable even with three-digit view numbers.
A view can be shared without logging in. The link carries an expiry date and a random component, and the underlying data source remains hidden. Shared views are listed in a separate overview.
Individual panels can be embedded into your own application. Sharing is restricted to specified origins; without an entry in this list, the panel will not load.
Views can be generated as PDFs on a schedule and sent via email. For reports to parties that do not have access to the service.
| Tier | Nodes | vCPU / RAM per node | Scope | Price / month |
|---|---|---|---|---|
| Dashboards Team | 1 | 4 / 8 GB | One node, up to 25 registered users, alerts via email. | €61.00$70.76 |
| Dashboards Operations | 2 | 8 / 16 GB | Two nodes behind a Load Balancer, on-call schedules. | €183.00$212.28 |
| Dashboards Enterprise | 3 | 4 / 16 GB | Three nodes, tenant separation, tracing data source. | €332.00$385.12 |
Net amounts plus 19% VAT. Each tier is the sum of the nodes and the data endpoint it reads: €19.00$22.04 for metrics and logs, €59.00$68.44 for the tier with tracing. Both amounts come unchanged from the monitoring options of the product catalogue — the dashboard service reads from the same endpoint and does not bill it a second time.
13 classes across five services. The last column states for each row where the amount comes from — from the maintained product catalogue or from a disclosed calculation based on catalogue values.
| Service and class | Nodes | vCPU / RAM per node | Storage per node | Price / month | Origin of the amount |
|---|---|---|---|---|---|
| Data Platform · Platform S | 2 | 8 / 16 GB | zustandslos | €229.70$266.45 | Control plane €65.70$76.21 + 2 × €82$95.12 per node |
| Data Platform · Platform M | 6 | 8 / 16 GB | zustandslos | €557.70$646.93 | Control plane €65.70$76.21 + 6 × €82$95.12 per node |
| Data Platform · Platform L | 16 | 8 / 16 GB | zustandslos | €1,377.70$1,598.13 | Control plane €65.70$76.21 + 16 × €82$95.12 per node |
| Kafka · Stream S | 3 | 2 / 4 GB | 480 GB | €353.03$409.51 | unchanged from the product catalogue |
| Kafka · Stream M | 3 | 4 / 15 GB | 1.9 TB | €1,329.99$1,542.79 | unchanged from the product catalogue |
| ClickHouse · Analytics M | 1 | 8 / 32 GB | 500 GB | €189.00$219.24 | unchanged from the product catalogue |
| ClickHouse · Analytics L | 3 | 16 / 64 GB | 2.0 TB | €549.00$636.84 | unchanged from the product catalogue |
| OpenSearch · Search S | 3 | 2 / 8 GB | 200 GB | €169.00$196.04 | ND2 + 120 GB Block NVMe, × 3 nodes, × management surcharge |
| OpenSearch · Search M | 3 | 4 / 16 GB | 600 GB | €409.00$474.44 | ND3 + 440 GB Block NVMe, × 3 nodes, × management surcharge |
| OpenSearch · Search L | 3 | 8 / 32 GB | 2.0 TB | €1,019.00$1,182.04 | ND4 + 1760 GB Block NVMe, × 3 nodes, × management surcharge |
| Dashboards · Dashboards Team | 1 | 4 / 8 GB | zustandslos | €61.00$70.76 | 1 × €42$48.72 per node + €19$22.04 data endpoint |
| Dashboards · Dashboards operation | 2 | 8 / 16 GB | zustandslos | €183.00$212.28 | 2 × €82$95.12 per node + €19$22.04 data endpoint |
| Dashboards · Dashboards group | 3 | 4 / 16 GB | zustandslos | €332.00$385.12 | 3 × €91$105.56 per node + €59$68.44 data endpoint |
Net amounts plus 19% VAT., without term discount and without region factor. Storage billed by consumption — Object Storage in the Data Platform, offloaded Kafka segments — is not included. The management surcharge of 1.1714 in the calculated rows is reverse-calculated from the maintained “ClickHouse Analytics M” plan: the price maintained there against a raw configuration of €161.35$187.17 from instance and block storage.
Both categories sit side by side in the menu and share two services. The dividing line is not between products, but between questions: analytics answer questions about many events, databases answer the question about a single one.
The rows are decision criteria, not features. If you land in the right column for more than half of them, you belong in Managed databases.
Both appear under databases because they are listed there as an engine with a plan, and under analytics because they have their role in the processing pipeline there. It is one service with one price, not two offerings.
If you only need ClickHouse and neither a processing pipeline nor dashboards, you book it under databases. The scope is the same.
the same service, the same price
A small table that is read a thousand times per second via the primary key belongs in PostgreSQL — even if it is only a few megabytes in size. A table that is fully calculated once an hour belongs in ClickHouse, even if it is small.
The data volume only determines the size class, not the category.
not in the data volume
The rule is not either-or, but side by side. Three setups that have proven themselves in operations.
The application writes to PostgreSQL. The changes are captured via the transaction log and mirrored to the columnar database. Analytics no longer burden the application's database.
Kafka carries the stream of events and decouples producers from consumers. The derived state remains in the business database; Kafka does not replace it, but feeds it.
Recurring analyses are stored in Valkey, an in-memory key-value store. The response time drops to milliseconds without the analysis itself being duplicated.
ClickHouse, Kafka and the other managed services are available in the database configurator with the same amounts as in the tables above. If you want to see the total bill first, you will find all items in the price overview — including storage, network and term discount.
