Skip to content
My accountDeutsch
Identical database nodes in open racks in dim light, each with a green light bar. In the video, the bar of one node changes to amber, goes dark, and then lights up green again; the other nodes remain lit throughout.

Managed databases from data centres in Germany: Nine engines, 26 plans, from โ‚ฌ47.30$54.87 per month net. Billed hourly without a term, or at a discounted monthly rate with a term. ENTRONYX CLOUD handles provisioning, backups, version upgrades and failover, i.e. switching to the running copy.

Engines
9
relational, document, cache, analytics, search, streaming, dashboards
Entry
โ‚ฌ47.30$54.87
Grafana Essential S, monthly net
Failover
< 30 s
from two nodes, a replica takes over
Delivery
< 10 min
private endpoint, TLS enforced

You choose the project or service, size, deployment, backup and location. Calculations use the same catalogue values as in the configurator, and the database is added directly to the cart.

What are you planning?optional

Without a project, select the service directly below. Each is calculated using the same catalogue values as in the configurator.

Service

The relational default. Managed PostgreSQL with automatic failover, point-in-time recovery and pgvector for embeddings.

Major versions
  • 17 (default)
  • 16
  • 15

The default is PostgreSQL 17; you can set a different major version in the configurator.

Size and deployment

โ‚ฌ54.46$63.17net / month

Node profile

Configuration per node; the amount applies to one node at the selected location.

Breakdown

A single node is the most cost-effective but has no fallback. High availability keeps replicas ready โ€” running copies that are switched to in the event of a failure (failover); the cluster adds further nodes.

Backup

+ โ‚ฌ9.00$10.44net / month

Daily snapshots, 7 recovery points, stored in the ENTRONYX network.

Location

Term

The discount applies to the monthly net amount; you pay for a term in advance with the first invoice. Hourly billing is only available without a term.

Net per monthโ‚ฌ63.46$73.61โ‚ฌ75.52$87.60 incl. VATโ‚ฌ0.0869$0.1008/hSummary

PostgreSQL 17, Essential S, one node: โ‚ฌ63.46 net per month.

First the data model, then the plan

The wrong engine cannot be cured with more memory. Here are the nine services, ordered by what they store and how they are accessed.

Still undecided?

In the finder, you start with your project, such as a web application, sessions, analytics or search, and the appropriate service is preselected.

Start with the project

What happens when something fails

Two things go wrong: a node fails, or a human deletes something. The replica, a running copy on a second node, helps against the former, while only the backup helps against the latter โ€” the distance between the two is the most important number on this page.

Status

During normal operations

The primary node writes; the replica, a running copy on a second node, follows the write stream. The journal, the log of every write operation, runs continuously.

Service

How much is lost in the worst case

The recovery point is the distance between the last backup state and the failure. It does not depend on the plan, but on the engine.

Three of the nine services keep a journal between two full backups, i.e. the log of every change, and recover to a freely chosen second: PostgreSQL, MySQL and MongoDB. Only what the journal had not yet recorded at the time of the failure is lost there.

The remaining six do not recover to the second; where they back up in snapshots, the worst case is a full day, 1,440 minutes. If you cannot afford this, choose an engine with a journal or a second write path in the application.

How far back: the retention tiers

Retention does not determine how accurately you hit the mark, but how far back you can go. The accuracy comes from the journal, the range from the tier.

Snapshots (7 days)
โ‚ฌ9.00$10.44
Daily snapshots, 7 recovery points, stored in the ENTRONYX network.
Managed backup (30 days)
โ‚ฌ29.00$33.64
Daily incremental backup, 30 points, geo-redundant in a second region.
Vault backup (365 days)
โ‚ฌ79.00$91.64
Immutable storage with WORM lock for compliance requirements.
Two identical, graphite-coloured nodes one above the other in an open light grey rack, each with a row of drive bays and a green light bar.

Primary and replica ยท failover under 30 s

A replica keeps the same state as the primary node โ€” including the accidentally deleted one. ENTRONYX CLOUD therefore keeps the backup separate from it in Object Storage, a dedicated storage service; from tier Managed Backup, it is located in a second region.

What we take over, what stays with you

A managed database takes operations off your hands, not data modelling โ€” the two columns show which part lies on which side.

Operations

ENTRONYX CLOUD takes over

Without additional agreement. The only separate items are retention and the backup journal.

Delivery
Node, operating system, engine version and network connection. You get a connection string, not a machine to set up.
Backup
We run and verify them: full daily backups, with the journal in between, encrypted in Object Storage. You choose how far back it goes as the retention tier.
Failover
From two nodes upwards, a sentinel monitors the primary role. If it fails, a replica takes over, typically in under 30 seconds.
Encryption
AES-256 for stored data, TLS 1.3 for every transmission. A connection attempt without TLS is rejected, not silently permitted.
Version upgrades
Minor versions in the maintenance window, major versions after pre-checking. We announce the dates.
Metrics
Connections, buffer hit ratio, replication lag and lock wait times are available at a Prometheus endpoint โ€” the common query format for monitoring tools.
Application

You keep

Everything that depends on your application โ€” and therefore nobody can decide for you.

Schema and queries
Tables, indexes, execution plans. We see the slow query in the log โ€” but we do not rewrite it.
The right size
Which plan, how many nodes, how many read replicas. There is no automatic scaling that silently increases the invoice.
Reconnection
After a failover, the connection string has the same name, but a different address. Your application must be able to reconnect.
Connection management
How many connections your application pool keeps open simultaneously. The limit of the instance is a limit, not a target.
Access rights
Roles, passwords and the allowlist of the public endpoint. An empty list means blocked, not open.
Archiving
Retention protects against errors from the last few weeks, not against retention obligations over years.
A light grey node in an open rack with a row of vertical drive bays; one is pulled out halfway, revealing the dark drive, with a green light bar illuminated next to it.

Nodes ยท ECC memory, local NVMe drives

Every database node runs with registered ECC memory, which detects and corrects flipped bits before they end up in a table, and on local NVMe drives โ€” fast, directly attached SSDs. Replicas are located in separate fire zones.

Four words this page assumes

Replica
a continuous second copy of the same database on another node
Failover
the switch of the write role to a replica when the primary node fails
Point-in-time recovery
the rollback to a freely chosen second instead of just to the last backup
Connection limit
how many clients may be connected simultaneously; in English Connection Limit

From data model to machine

26 plans across nine engines, from โ‚ฌ47.30$54.87 to โ‚ฌ1,666.59$1,933.24 per month. The choice depends on three factors: the working set, the dataset and whether an outage can last minutes or seconds.

All 26 plans of the nine engines with vCPU, memory, storage, number of nodes and monthly price, net
EnginePlanvCPURAMStorageNodesConfigureMonth
PostgreSQLEssential S24 GB80 GB1no failoverIn the finder โ€” PostgreSQL Essential Sโ‚ฌ54.46$63.17
PostgreSQLBusiness M27 GB160 GB2In the finder โ€” PostgreSQL Business Mโ‚ฌ277.25$321.61
PostgreSQLBusiness L415 GB320 GB2In the finder โ€” PostgreSQL Business Lโ‚ฌ554.36$643.06
PostgreSQLEnterprise XL830 GB640 GB3In the finder โ€” PostgreSQL Enterprise XLโ‚ฌ1,666.59$1,933.24
MySQLEssential S24 GB80 GB1no failoverIn the finder โ€” MySQL Essential Sโ‚ฌ54.46$63.17
MySQLBusiness M415 GB320 GB2In the finder โ€” MySQL Business Mโ‚ฌ554.36$643.06
MySQLEnterprise L830 GB640 GB3In the finder โ€” MySQL Enterprise Lโ‚ฌ1,666.59$1,933.24
ValkeyCache S28 GBRAM only1no failoverIn the finder โ€” Valkey Cache Sโ‚ฌ107.75$124.99
ValkeyCache M416 GBRAM only2In the finder โ€” Valkey Cache Mโ‚ฌ481.36$558.38
ValkeyCache L832 GBRAM only2In the finder โ€” Valkey Cache Lโ‚ฌ962.87$1,116.93
MongoDBBusiness S28 GB40 GB3In the finder โ€” MongoDB Business Sโ‚ฌ243.97$283.01
MongoDBBusiness M416 GB80 GB3In the finder โ€” MongoDB Business Mโ‚ฌ483.77$561.17
MongoDBBusiness L832 GB160 GB3In the finder โ€” MongoDB Business Lโ‚ฌ941.92$1,092.63
ClickHouseAnalytics M832 GB500 GB1no failoverIn the finder โ€” ClickHouse Analytics Mโ‚ฌ189.00$219.24
ClickHouseAnalytics L1664 GB2.0 TB3In the finder โ€” ClickHouse Analytics Lโ‚ฌ549.00$636.84
Apache KafkaStream S24 GB480 GB3In the finder โ€” Apache Kafka Stream Sโ‚ฌ353.03$409.51
Apache KafkaStream M415 GB1.9 TB3In the finder โ€” Apache Kafka Stream Mโ‚ฌ1,329.99$1,542.79
OpenSearchEssential S24 GB40 GB1no failoverโ‚ฌ59.42$68.93
OpenSearchBusiness M27 GB240 GB3โ‚ฌ386.10$447.88
OpenSearchBusiness L415 GB480 GB3โ‚ฌ772.19$895.74
OpenSearchEnterprise XL415 GB960 GB6โ‚ฌ1,564.10$1,814.36
Apache CassandraEssential S24 GB40 GB1no failoverIn the finder โ€” Apache Cassandra Essential Sโ‚ฌ96.80$112.29
Apache CassandraBusiness M415 GB480 GB3In the finder โ€” Apache Cassandra Business Mโ‚ฌ867.90$1,006.76
Apache CassandraBusiness L830 GB960 GB3In the finder โ€” Apache Cassandra Business Lโ‚ฌ1,597.17$1,852.72
GrafanaEssential S24 GBData in the sources1no failoverIn the finder โ€” Grafana Essential Sโ‚ฌ47.30$54.87
GrafanaEssential M27 GBData in the sources1no failoverIn the finder โ€” Grafana Essential Mโ‚ฌ95.70$111.01

The node count is not a performance metric: a second node does not double the throughput, but shortens the outage. The largest plan features 64 GB memory โ€” enough to keep the working set of most applications entirely in memory.

If there's no reason not to: PostgreSQL

Strict schemas, true transactions, JSONB for everything that doesn't fit into columns โ€” and extensions that make a second database obsolete.

The relational default

PostgreSQL

Managed PostgreSQL with automatic failover, point-in-time recovery and pgvector for embeddings.

Versions171615

fromโ‚ฌ54.46$63.17per month, net

Choose it for

  • Transactional core systems with foreign keys, check constraints and real transactions
  • Mixed workloads: relational tables alongside JSONB documents in the same query
  • Similarity search over embeddings with pgvector, without a second database alongside
  • Geospatial data with PostGIS, time series with native range partitioning

Do not choose it for

  • Analytical queries over billions of rows โ€” ClickHouse is orders of magnitude faster for this
  • As a high-throughput queue; SKIP LOCKED goes a long way, but not as far as Kafka
  • Pure session or page cache, for which memory is sufficient
Extensions
pgvector, PostGIS, pg_cron, pg_partman
Connection pooler
upstream, transaction mode
Read replicas
up to 5, also cross-region
Recovery point
to the second (PITR)
Plans for PostgreSQL: vCPU, memory, storage, number of nodes and monthly price
PlanvCPURAMStorageNodesMonth
Essential S24 GB80 GB1no failoverโ‚ฌ54.46$63.17
Business M27 GB160 GB2โ‚ฌ277.25$321.61
Business L415 GB320 GB2โ‚ฌ554.36$643.06
Enterprise XL830 GB640 GB3โ‚ฌ1,666.59$1,933.24

MySQL, MongoDB, ClickHouse โ€” and three more in detail

Six services without their own section, but with the same data sheet: major versions, plans with features and price, and an honest differentiation โ€” what they were built for and what expressly not.

Relational. Fixed schema, foreign keys and transactions โ€” here in the version for legacy applications that speak MySQL and should continue to do so.

MySQLFor established applicationsVersions8.48.0fromโ‚ฌ54.46$63.17Data sheet

For established applications

MySQL

Managed MySQL with InnoDB cluster, read replicas and automatic backups.

Versions8.48.0

fromโ‚ฌ54.46$63.17per month, net

Choose it for

  • Existing applications from the PHP, WordPress and Magento ecosystem without porting effort
  • Read-heavy workloads that scale horizontally via replicas
  • InnoDB cluster with group replication and automatic primary election

Do not choose it for

  • Complex analytical queries with window functions over large datasets
  • Applications built on PostgreSQL extensions like pgvector or PostGIS
  • New projects without MySQL dependency โ€” PostgreSQL offers more there for the same price
Character set
utf8mb4, collation utf8mb4_0900_ai_ci
Replication
GTID-based, semi-synchronous
Journal
Binlog, ROW format, 7 days
Recovery point
to the second (PITR)
Plans for MySQL: vCPU, memory, storage, number of nodes and monthly price
PlanvCPURAMStorageNodesMonth
Essential S24 GB80 GB1no failoverโ‚ฌ54.46$63.17
Business M415 GB320 GB2โ‚ฌ554.36$643.06
Enterprise L830 GB640 GB3โ‚ฌ1,666.59$1,933.24

Schemaless. Structure per record instead of per table. For data where fields can differ between two entries.

MongoDBWhen the schema changesVersions8.07.0fromโ‚ฌ243.97$283.01Data sheet

When the schema changes

MongoDB

Document database with replica set, aggregation pipeline and Atlas-compatible driver API.

Versions8.07.0

fromโ‚ฌ243.97$283.01per month, net

Choose it for

  • Catalogue and product data with highly variable fields per entry
  • Event and telemetry data whose structure changes over time
  • Applications that write aggregation pipelines instead of SQL

Do not choose it for

  • Highly connected models with many relationships โ€” $lookup is not a JOIN
  • Reports over the entire dataset; the column-oriented storage is missing for this
  • Accounting consistency across many documents; transactions exist, but they are expensive
Topology
Replica Set with three nodes
Driver API
compatible with official drivers
Recovery
Oplog-based, accurate to the second
Indexes
compound, text, geo, TTL
Plans for MongoDB: vCPU, memory, storage, number of nodes and monthly price
PlanvCPURAMStorageNodesMonth
Business S28 GB40 GB3โ‚ฌ243.97$283.01
Business M416 GB80 GB3โ‚ฌ483.77$561.17
Business L832 GB160 GB3โ‚ฌ941.92$1,092.63

Analytics. Not built for the individual record, but for throughput across billions of them.

ClickHouseSub-second analyticsVersions25.324.8fromโ‚ฌ189.00$219.24Data sheet

Sub-second analytics

ClickHouse

Column-oriented analytics database for billions of rows with sub-second response times.

Versions25.324.8

fromโ‚ฌ189.00$219.24per month, net

Choose it for

  • Queries across billions of rows with sub-second response times
  • Log, metric and clickstream data with high compression
  • Materialised views that pre-aggregate on write

Do not choose it for

  • Transactional workloads with individual updates and deletions per row
  • Very high number of concurrent point queries โ€” the optimiser is built for throughput, not concurrency
  • Referential integrity; foreign keys do not exist
Table engine
MergeTree family, default ReplicatedMergeTree
Compression
ZSTD, typically factor 8 to 12
Interfaces
HTTP, native protocol, MySQL port
Ingestion
Kafka engine for direct ingestion
Plans for ClickHouse: vCPU, memory, storage, number of nodes and monthly price
PlanvCPURAMStorageNodesMonth
Analytics M832 GB500 GB1no failoverโ‚ฌ189.00$219.24
Analytics L1664 GB2.0 TB3โ‚ฌ549.00$636.84

Search. An inverted index instead of tables: matches are found via words and relevance, not via keys.

OpenSearchSearch and analyse logsVersions3.12.19fromโ‚ฌ59.42$68.93Data sheet

Search and analyse logs

OpenSearch

Distributed search and analytics engine with full-text index, aggregations and dashboards.

Versions3.12.19

fromโ‚ฌ59.42$68.93per month, net

Choose it for

  • Full-text search across products, documents and articles, sorted by relevance and with facets
  • Evaluate logs and events centrally, with index lifecycle by age
  • Aggregations over search results, such as counts per category or time period

Do not choose it for

  • Primary storage of data whose loss is unacceptable โ€” restored to the last snapshot
  • Relational queries with joins across many tables; this is what PostgreSQL is built for
  • Transactions across multiple documents โ€” they do not exist
Nodes per plan
1 to 6
Read replicas
up to 6
Recovery point
last snapshot
Backup
snapshots of entire indices
Plans for OpenSearch: vCPU, memory, storage, number of nodes and monthly price
PlanvCPURAMStorageNodesMonth
Essential S24 GB40 GB1no failoverโ‚ฌ59.42$68.93
Business M27 GB240 GB3โ‚ฌ386.10$447.88
Business L415 GB480 GB3โ‚ฌ772.19$895.74
Enterprise XL415 GB960 GB6โ‚ฌ1,564.10$1,814.36

Distributed. No primary node, but a ring: every node accepts writes, the copies are distributed across all of them.

Apache CassandraA ring without a primary nodeVersions5.04.1fromโ‚ฌ96.80$112.29Data sheet

A ring without a primary node

Apache Cassandra

Wide-column storage without primary nodes. Every node accepts writes, and copies are distributed across the ring.

Versions5.04.1

fromโ‚ฌ96.80$112.29per month, net

Choose it for

  • Write-intensive workloads where every node accepts writes
  • Time series, events and metrics with a very high number of writes per second
  • Throughput that scales with the number of nodes instead of the size of a single one

Do not choose it for

  • Queries that the data model did not anticipate โ€” reads are done via the partition key
  • Recovery to a ring-wide consistent point in time; the nodes are not on the exact same second
  • Transactions across multiple tables and foreign keys โ€” neither exist
Nodes per plan
1 to 3
Additional ring nodes
up to 6
Query language
CQL
Recovery point
last snapshot per node
Plans for Apache Cassandra: vCPU, memory, storage, number of nodes and monthly price
PlanvCPURAMStorageNodesMonth
Essential S24 GB40 GB1no failoverโ‚ฌ96.80$112.29
Business M415 GB480 GB3โ‚ฌ867.90$1,006.76
Business L830 GB960 GB3โ‚ฌ1,597.17$1,852.72

Monitoring. No storage, but the interface on top โ€” dashboards across the sources you connect.

GrafanaDashboards for everything that provides numbersVersions1211fromโ‚ฌ47.30$54.87Data sheet

Dashboards for everything that provides numbers

Grafana

Managed Grafana instance for dashboards across metrics, logs and databases.

Versions1211

fromโ‚ฌ47.30$54.87per month, net

Choose it for

  • Dashboards across metrics, logs and databases in one place
  • Display the metrics of your managed databases via their Prometheus endpoint
  • Alerting rules on thresholds without operating your own monitoring instance

Do not choose it for

  • Storage of metrics โ€” they remain in the connected sources
  • High availability via additional nodes; Grafana runs as a single instance
  • Evaluations across billions of rows โ€” ClickHouse calculates them, Grafana displays them
Nodes per plan
1
Storage
no storage, data in the sources
Backed up
Dashboards, data sources, users
Recovery
from the last export
Plans for Grafana: vCPU, memory, storage, number of nodes and monthly price
PlanvCPURAMStorageNodesMonth
Essential S24 GBData in the sources1no failoverโ‚ฌ47.30$54.87
Essential M27 GBData in the sources1no failoverโ‚ฌ95.70$111.01

Valkey: memory with a network connection

Valkey is the open continuation of Redis: the same protocol, the same drivers, the same data structures โ€” under a licence that explicitly allows it to be run as a managed service. Existing Redis clients connect without modification.

Choose it for

Versions8.07.2
  • Session management, object cache, rate limiting and distributed locks
  • Publish and subscribe as well as lightweight queues via streams
  • Leaderboards and counters with sorted sets at very high request rates

Do not choose it for

  • Primary storage of data whose loss is unacceptable
  • Data volumes larger than the memory of the plan
  • Queries over relationships โ€” there are no joins and no indexes over values
Persistence
AOF every second, RDB hourly
No block storage
Dataset must fit into RAM
Cluster mode
from three nodes, 16,384 slots
Protocol
RESP3, Redis driver compatible

Storage classes

There is no block storage to add: the memory of the plan is the capacity. If the dataset does not fit, the plan is not too small, but the wrong engine was chosen.

Valkey storage classes: the memory is also the usable capacity โ€” there is no โ€œStorageโ€ column here.
PlanvCPUMemoryNodesTypical useMonth
Cache S28 GB1no failoverSessions and object cache of a single applicationโ‚ฌ107.75$124.99
Cache M416 GB2Shared cache of multiple services, rate limiting, locksโ‚ฌ481.36$558.38
Cache L832 GB2Leaderboards and counters with very high request ratesโ‚ฌ962.87$1,116.93

Persistence mode

Two processes run simultaneously: the append-only file (AOF) logs every modifying command and is forced to the drive once per second. The snapshot (RDB) captures a complete image of the memory every hour.

When a node restarts, Valkey reads the append-only file again; at most, the last second is lost. A recovery from the backup, on the other hand, relies on the last complete snapshot โ€” Valkey does not keep a continuous journal.

The practical rule from this: everything Valkey contains must be rebuildable from another source. If this is not the case, the data belongs in PostgreSQL and only its copy here.

Kafka: what happened, not what currently applies

A distributed log instead of a table: every entry remains until the retention period ends, and every consumer reads it at their own pace. From โ‚ฌ353.03$409.51 per month, always with three nodes, because a single broker does not secure a log.

Events instead of states

Apache Kafka

Managed event streaming with KRaft mode, schema registry and Connect runtime.

Versions3.93.8

fromโ‚ฌ353.03$409.51per month, net

Choose it for

  • Decoupling of services that do not need to be available at the same time
  • Change data capture from databases via Debezium and the Connect runtime
  • Repeatable processing: consumers rewind without loading the source

Do not choose it for

  • Request-and-response patterns; Kafka has no return channels
  • Queries on content โ€” there is no index, only offsets and partitions
  • Small systems with few events per second; the operational effort is not worth it there
Operating mode
KRaft, without ZooKeeper
Schema registry
included, Avro, Protobuf, JSON Schema
Retention
by time or size, per topic
Encryption
TLS and SASL/SCRAM enforced
Plans for Apache Kafka: vCPU, memory, storage, number of nodes and monthly price
PlanvCPURAMStorageNodesMonth
Stream S24 GB480 GB3โ‚ฌ353.03$409.51
Stream M415 GB1.9 TB3โ‚ฌ1,329.99$1,542.79

What applies to every engine

Operational characteristics differ less between services than the data models do. Where there are differences, they are in the table โ€” not in the small print.

Automatic failover

For multi-node plans, a sentinel monitors the primary role. If it fails, a replica takes over; the failover typically takes less than 30 seconds. The connection string points to a name, not an address โ€” your application only needs reconnection logic.

Encryption

Data at rest is AES-256 encrypted on NVMe, backups are additionally encrypted in Object Storage. In transit, TLS 1.3 with certificate validation applies; a connection attempt without TLS is rejected, not silently permitted.

Private network connection

Each instance gets an endpoint in the vRack, the private network between your servers, and no public address by default. If you need external access, you can enable the public endpoint and provide an allowlist โ€” empty means blocked, not open.

Metrics and logs

A Prometheus-compatible endpoint provides connections, buffer hit ratio, replication lag and lock wait times. Slow queries end up in the log and in the console, with an execution plan.

Operational characteristics per engine โ€” where they differ
FeaturePostgreSQLMySQLValkeyMongoDBClickHouseApache KafkaOpenSearchApache CassandraGrafana
Failoverfrom two nodesPatroni, < 30 sGroup replication, < 30 sSentinel or cluster modeReplica set, election in < 15 sReplica takes over shardPartition leader switches automaticallyCluster manager re-elects, < 30 sno primary node โ€” ring continues runningStorage recovery
Distribute read loadacross additional nodesup to 5, cross-regionup to 5in cluster modein replica setvia shard replicasvia consumer groupsadditional data nodesRing nodeโ€”
Storage in planup to 640 GBup to 640 GBRAM onlyup to 160 GBup to 2.0 TBup to 1.9 TBup to 960 GBup to 960 GBData in the sources
Major version upgradein maintenance window, with pre-checkin maintenance window, with pre-checkrolling, without downtimerolling, without downtimerolling, without downtimerolling, without downtimerolling, without downtimerolling, without downtimein the maintenance window
Connection limitGuideline 4 ร— vCPU + 100Pooler upstreamPooler upstream10,000 concurrentDriver pool per applicationdesigned for throughput, not concurrencyunlimited, partition count is the limitDriver pool per applicationdesigned for throughputWeb interface
Three open light grey racks in a bright room, each containing three identical nodes with a green light bar.

Primary nodes and replicas are located in separate fire zones: if an entire room fails, the failover finds the node in the next one. From tier Managed Backup, the backup is stored in a second region.

Nodes ยท separate fire zones

Daily full backup, journal in between

A backup that no one has restored is an assumption. That is why recovery here runs not only in an emergency, but monthly as a random sample โ€” and always into a new instance, never over the running one.

  1. 01

    Full backup

    daily

    Once a day, the complete dataset is backed up and stored encrypted โ€” in tier Snapshots within the ENTRONYX network, from Managed Backup in a second region.

  2. 02

    Journal

    continuous

    With PostgreSQL, MySQL and MongoDB, the journal runs between two full backups, i.e. the log of every change. It closes the gap between the last full image and the point in time you select later.

  3. 03

    Retention

    7, 30 or 365 days

    You select the tier in the finder or in the configurator. It determines how far back you can go โ€” not how precisely.

  4. 04

    Recovery

    into a new instance

    You choose a point in time, we create a second instance from it. The running one remains untouched โ€” comparing both states is thus part of the process and not its risk.

  5. 05

    Validation

    monthly

    Every month, a sample is actually restored and checked against the source. Anything noticed during this is an operational procedure and not a discovery in an emergency.

Recovery point and process per engine
EngineRecovery pointMethod
PostgreSQLper secondBase backup and continuous WAL journal
MySQLper secondFull backup and binlog in ROW format, 7 days
Valkeylast full dumpfrom RDB and AOF, without continuous journal
MongoDBper secondFull backup and oplog
ClickHouselast backupFull backup per partition
Apache KafkaRetention instead of recoveryEvents remain readable in the topic
OpenSearchlast snapshotsnapshots of entire indices
Apache Cassandralast snapshotsnapshot and commit log per node
Grafanalast exportDashboards, data sources and users as export
A storage enclosure in a light grey rack, densely packed with drive bays; a small green status light is illuminated on each one.

Backup ยท daily, testing monthly

Daily full, in between the journal. The copy in a second region is included from Managed Backup; for the Snapshots it costs โ‚ฌ20.00$23.20 net per month. Every month, a sample is actually restored and checked against the source.

How an instance becomes accessible

A managed database is only managed if the route to it is also defined. The default is therefore: private, encrypted, with separate roles โ€” and none of this can be switched off accidentally.

Each instance receives a name in the internal DNS and an endpoint in the vRack, the private network between your servers โ€” no public A record, no accessible address on the internet. Anyone who needs to access it from the outside must explicitly enable the public endpoint and provide an allowlist; an empty list means blocked, not open.

Included is vRack 1 Gbit/s. More bandwidth costs per month, net: vRack 10 Gbit/s โ‚ฌ39.00$45.24, vRack 25 Gbit/s โ‚ฌ119.00$138.04, vRack 50 Gbit/s โ‚ฌ219.83$255.00 and vRack 100 Gbit/s โ‚ฌ421.49$488.93.

Three roles per instance

admin
Creates databases, schemas and other roles. Belongs in the deployment run, not in the application.
app
Read and write in exactly one database. This is the role that is in your connection string.
readonly
Read only, against a replica by default. For reports, exports and analysis tools.

Each role has its own password, which can be renewed individually. An application that connects with admin is a finding, not an operating state.

Back of a node in a light grey rack: on the left two power supplies with black power cables, on the right fibre optic ports with orange cables and next to them a single copper port with a black cable, all cables routed into a side cable duct.

Endpoint ยท private in the vRack, TLS 1.3

The data path runs via the private endpoint in the vRack, TLS 1.3 is enforced. Management has its own connection โ€” it does not share the line with your queries.

Connect, recover, replicate
# Get connection string โ€” TLS is mandatory, not an option
entronyx db uri shop-pg --role app
# postgresql://app:โ€ฆ@shop-pg.fra1.db.entronyx.internal:5432/app?sslmode=verify-full
 
# Recovery to a point in time within the retention
entronyx db restore shop-pg \
  --point-in-time "2026-08-27T21:14:05Z" \
  --target shop-pg-restore
 
# Attach read replica in a second region
entronyx db replica create shop-pg --region hel1 --plan pg-m
sslmode=verify-full is the default. The root certificate is included in every image under /etc/ssl/certs/entronyx-db.pem and is updated via the package repository.

Migrate existing databases

The path is the same for every engine: transfer completely once, then continuously catch up until the lag is zero, and only then switch over. A midnight dump is not a migration, it's a gamble.

  1. 01

    Inventory

    1 to 2 hours

    Size per table, used extensions, character set and collation, roles and permissions, number of concurrent connections, longest running query. Everything that is special on the source becomes visible here โ€” not just during import.

  2. 02

    Choose target plan

    Working set counts

    Memory at least as large as the frequently read part of the data, storage about one and a half times the inventory, number of nodes according to failure requirements. You can upgrade later, but downgrading is not always possible.

  3. 03

    Initial transfer

    depending on data volume

    Dump or base backup, imported in parallel. Owners and permissions of the source are deliberately omitted and reassigned in the target โ€” otherwise you drag along roles that do not exist here.

  4. 04

    Continuously catch up

    Hours to days

    Logical replication, GTID replication or change streams keep the target up to date while the source continues to work. In this phase, you already test the application against the target โ€” read-only.

  5. 05

    Switchover

    Maintenance window, a few minutes

    Stop write access on the source, wait until the lag is zero, swap connection string, check sequences, enable write access. If you do not measure the lag, you lose records.

  6. 06

    Keep the way back open

    7 days

    The source remains unchanged for a week. This costs little and is the only reliable rollback plan there is for a migration.

PostgreSQL migration with logical replication
# 1. Create target instance, private endpoint in the vRack
entronyx db instance create shop-pg \
  --engine postgresql --version 17 --plan pg-m \
  --region fra1 --vrack prod --no-public-endpoint
 
# 2. Transfer schema and data, without owner and permissions of the source
pg_dump --no-owner --no-privileges --format=directory --jobs=4 \
  --dbname="$SOURCE_URL" --file=/tmp/dump
 
pg_restore --no-owner --jobs=4 \
  --dbname="$(entronyx db uri shop-pg --role admin)" /tmp/dump
 
# 3. Continuously catch up until the lag is zero
psql "$SOURCE_URL" -c "CREATE PUBLICATION migration FOR ALL TABLES;"
psql "$(entronyx db uri shop-pg)" -c \
  "CREATE SUBSCRIPTION migration CONNECTION '$SOURCE_URL' PUBLICATION migration;"
 
# 4. Measure lag โ€” only switch over at 0 bytes
psql "$(entronyx db uri shop-pg)" -c \
  SELECT application_name,
          pg_wal_lsn_diff(sent_lsn, replay_lsn) AS lag_bytes
     FROM pg_stat_replication;"
 
# 5. Clean up after the switchover
psql "$(entronyx db uri shop-pg)" -c "DROP SUBSCRIPTION migration;"
For MySQL, step 3 is replaced by GTID-based replication, for MongoDB mongosync with change streams. The pattern remains identical.
Tools and realistic downtime per engine
EngineInitial transferContinuousOutage
PostgreSQLpg_dump / pg_basebackuplogical replication< 5 min
MySQLMySQL Shell, util.copyInstanceGTID replication< 5 min
ValkeyImport RDB fileREPLICAOF against the source< 1 min
MongoDBmongodump / mongorestoremongosync, Change Streams< 10 min
ClickHouseINSERT โ€ฆ SELECT via remoteSecure()lag per partitionminutes to hours
Apache KafkaMirrorMaker 2MirrorMaker 2 with offset translationwithout downtime

There is no standard path in the table for OpenSearch, Apache Cassandra and Grafana; we will coordinate the migration with you. Discuss migration

What people ask before their first database

Eleven questions that regularly come up before ordering. The answers use the same catalogue values as the sections above.

Can't find your question? Contact us.

How much does Managed PostgreSQL in Germany cost per month?

Managed PostgreSQL runs in data centres in Germany and costs per plan and month: The smallest plan Essential S runs on one node with 4 GB memory, the largest Enterprise XL on three nodes with 30 GB. The amounts below this answer are net plus 19% VAT and without a contract term; for 12 or 24 months in advance, they decrease by 5% and 8% respectively.

PostgreSQL Essential S, per month netโ‚ฌ54.46$63.17

PostgreSQL Enterprise XL, per month netโ‚ฌ1,666.59$1,933.24

PostgreSQL in detail

What does โ€œmanagedโ€ mean?

We provision the node, operating system and engine, perform the backup, monitor operations, fail over in the event of an outage and apply version updates. Schema, queries, size and access rights remain with you.

What we take over, what stays with you

How fast is an instance ready?

Usually in under 10 minutes after the order, with a private endpoint and forced TLS. You get a connection string, not a machine to set up.

Configure database

What happens if a node fails?

From two nodes, a sentinel detects the failure, and a replica takes over the write role, typically in under 30 seconds. The connection name remains, the address changes โ€” your application must be able to reconnect.

Failure and user error illustrated

How far back does a recovery go?

As far back as the selected retention period goes: 7 to 365 days. A journal is kept by 3 of the 9 engines; they restore to the second within this time, the others to the last backup โ€” the table under Backup shows which ones.

Backup and recovery

Can I scale up later?

Yes. vCPU, memory and storage can be increased during operations, on a rolling basis for multi-node plans; the connection string remains the same.

All plans compared

Is the database accessible from the internet?

Not without your action: each instance gets an endpoint in the vRack, the private network between your servers, and no public address. You explicitly enable the public endpoint and provide an allowlist; an empty list means blocked.

Access and network

How much does the smallest database cost?

The smallest plan is Grafana Essential S with one node, i.e. without failover. With failover, the list starts at MongoDB Business S with three nodes.

Grafana Essential S, net per monthโ‚ฌ47.30$54.87

MongoDB Business S, net per monthโ‚ฌ243.97$283.01

Where do the databases run, and where are the backups stored?

The databases run at two locations in Germany; you choose the location in the finder or in the configurator. Primary nodes and replicas are located in separate fire zones of the same location, the backup is additionally stored in a second region from level Managed Backup.

Where the backup is stored

How fast does support respond, and how do I reach them?

Via ticket in the account: The price includes the basic support of the SLA with an initial response in 8 hours for critical incidents (P1), available Monday to Friday from 08:00 to 18:00. Initial response means: A human has accepted the report, classified it and reported back the processing status โ€” the SLA does not guarantee a resolution time. In addition to basic support, it offers three further support levels; the highest responds to P1 in 15 minutes.

Is there a minimum term, and how do I cancel?

Not without a term commitment: You can delete services billed by the hour at any time without notice in the console or via the API, and billing ends at the time of deletion (Section 11 paragraph 3 of the Terms and Conditions). You can cancel monthly contracts with one day's notice to the end of the term, contracts for 12 or 24 months with 30 calendar days โ€” otherwise they renew for twelve months (Section 11 paragraph 2 of the Terms and Conditions). You pay the term in advance and receive a 5% or 8% term discount in return.

Choose term in the finder

from, per month, netโ‚ฌ47.30$54.87

Configure database

Configure database or migrate existing

In the finder, the price of your database is fixed after a few decisions, and the cart takes it directly. For the migration of existing systems, there is a guide and, upon request, support from our technicians.

Engines
9
from
โ‚ฌ47.30$54.87
Plans
26
Failover
< 30 s