Managed database hosting
Managed databases run on the Colchester cluster. You choose MySQL, MariaDB, PostgreSQL, MongoDB or SQL Server 2022 Standard when you customise. Other engines can be discussed for an enterprise order. Connection details arrive by email. A single instance has one dedicated IP. High availability uses a shared virtual IP so the address stays the same during failover. The order page is database hosting. The long article is MariaDB and PostgreSQL. The choice guide is Choosing MariaDB or PostgreSQL.
Pick the engine the application already speaks. MariaDB and MySQL are separate options even though they are close. PostgreSQL is not either of them. MongoDB and SQL Server 2022 Standard are on the same plans and are not interchangeable with a SQL dump. Test a restore on the engine you ordered.
Plans
Starter is £29.99 a month for development, staging and smaller production databases: 20 GB of storage, 2 GB of RAM, a multi-core CPU, a dedicated IP, daily backups kept 7 days, optional high availability and 24/7 technical support. Standard is £39.99: 50 GB and 4 GB of RAM, for production that needs more headroom. Business is £59.99: 100 GB and 8 GB. Enterprise is £149.99: 250 GB, a 16-core dedicated CPU, 32 GB of dedicated RAM, the multi-node cluster with a shared virtual IP included, and hourly backups kept 28 days. Yearly prices are on the same cards. Options change the checkout total.
On Starter, Standard and Business, extra backup retention is £5 a month for each extra week, in 7-day steps, up to 28 days. High availability on those plans is either a pair, replica plus automatic failover, or a three-node synchronous cluster. Enterprise already includes the multi-node cluster and 28 days. A replica copies data, including a mistake. A backup is what you restore. Order the one that matches the failure you care about, or both.
MariaDB and PostgreSQL
MariaDB is the engine to name when the application or the import is MariaDB, and when a Kubernetes scaler should be allowed to use the hosted MariaDB. PostgreSQL is the engine to name when the client, the extensions and the dump are PostgreSQL. They do not share a data directory. Two engines means two instances and two connection strings. A small site database can stay inside Plesk shared hosting. Move here when you want the address, the retention and the optional replica written on a plan.
Adminer in the Compute app library can browse MariaDB, MySQL, PostgreSQL or SQLite. It is a client you run. It is not this product and it does not take these backups. Use the managed plan for the data that has to stay up.
Security
Every plan can sit behind no extra firewall, the managed cluster firewall where you add rules, or a dedicated self-managed Cisco appliance. SSL, automatic security patches and IP restrictions apply. Enterprise adds VPN access and encryption at rest. Those are not included merely because the engine is PostgreSQL. Restrict the database to the application and the people who administer it. The Cisco ASA and Firepower plans are the rack firewalls. They are a separate order from the database firewall option.
Most storage, RAM and CPU changes can be done with little or no downtime, as the FAQ describes. A restore is not one of those changes. Migration from another provider is available and should include a rehearsal. Cut over when staging against the restored copy behaves. Keep the old instance until you have read production on the new address, not only written to it.
Beside Kubernetes and hosting
A pod can call the hosted MariaDB, a service in its own namespace, or a public host. It cannot use a private address in another namespace as a scaler target. The database password belongs in the application secret, not in the image. More replicas do not add database RAM. Change the database plan when the database is the limit. SSD storage is what the plans run on. The exact performance depends on the plan and the workload, which is as specific as the product page gets, and it is the right level of promise.
Order the engine, then the disk, then the retention, then the replica if you need the address to survive a node, then the firewall tier. Send this page to someone who needs the service in one place, and the plan page to someone who is ready to configure it. If the engine name is still unknown, wait. The wrong dialect wastes the migration, and the backup you take afterwards is a backup of a database the application cannot open.
Which failure you are buying for
A single instance on Starter, Standard or Business has one dedicated IP and daily backups for 7 days. That covers a database you can restore, inside a week, onto an address that is allowed to change if the instance is rebuilt. Add weeks at £5 a month, up to 28 days, when a bad change might be noticed later than a week. Add the high-availability pair when a node failure must fail over by itself and the application must keep the same address. Add the three-node synchronous cluster when that is the option you actually selected at checkout. Use Enterprise when you want 250 GB, 32 GB of dedicated RAM, a 16-core dedicated CPU, hourly backups for 28 days and the multi-node cluster included, plus VPN access and encryption at rest.
Do not buy Enterprise only because the engine is PostgreSQL, and do not skip the replica only because the engine is MariaDB. The engine is the dialect. The plan is the disk, the memory, the backup and the address. Starter at £29.99 with 20 GB and 2 GB of RAM is for development, staging and smaller production. Standard at £39.99 with 50 GB and 4 GB is the next production step. Business at £59.99 with 100 GB and 8 GB is the step after that. Those are the card prices. Checkout adds the options.
MariaDB next to Kubernetes, PostgreSQL next to the application
If a namespace scales from a table or a queue, the allowed hosts are a service in that namespace, the hosted MariaDB, or a public host. Order MariaDB on this product when that sentence is your architecture. Order MySQL when the vendor said MySQL. Order PostgreSQL when the dump and the extensions are PostgreSQL. Order MongoDB or SQL Server 2022 Standard when those are the drivers. Two dialects means two instances. Adminer can look at several of them from a browser and still is not the backup.
Connection details come by email. Use SSL. Restrict the IP. Put the password in the application's secret store. A Kubernetes image that contains the password is a mistake the database plan cannot fix. More pods do not add RAM to the database. The FAQ says most RAM, CPU and storage changes on the database can be done with little or no downtime. Plan them before the disk is full. 20 GB includes more than the raw table size once indexes and engine logs are counted.
A restore you have tried
High availability copies the drop of a table to the replica. The backup, inside its retention, is the way back. Know whether you bought 7 days or longer before you need the eighth day. Enterprise's hourly copies are a tighter loop than a daily copy. They are still a loop you should have restored once, onto a staging address, before you trust them. Migration help from another provider is the same idea: rehearse, then cut over, then keep the old instance until reads on the new one look right.
The firewall tier is part of the same order. No extra firewall leaves you with SSL and IP restrictions. The managed cluster firewall is rules you add. A dedicated Cisco appliance is one you manage. The ASA and Firepower plans are the hardware in a rack, and they are not silently included in Starter. Colchester is where the cluster sits. Performance is the performance of the plan and the workload on SSD, which is the claim the product makes and the claim to repeat.
Order from the database page when the engine name is known. Use the guide if you are choosing between MariaDB and PostgreSQL. Use the article if you want the plans written out as a narrative. Use this page as the link that says what managed database hosting is: an engine you choose, an address we send, backups with a stated retention, and an optional replica that keeps that address.
From the first connection to a plan you can grow
When the instance is provisioned, the email with the connection details is the start, not the end. Put the address in the application, confirm SSL, and allow only the IPs that should connect: the site, the virtual server, or the egress of the namespace. Run the application's own schema step against a restored or empty staging database before production points at the new IP. If the address is a shared virtual IP, label it as such in the runbook so nobody goes looking for a new hostname during failover. If it is a dedicated IP, label that too, so nobody expects failover you did not buy.
Watch the disk against the plan: 20 GB, 50 GB, 100 GB or 250 GB. The raw data file is not the whole use. Indexes and the logs the engine keeps are part of the same storage. Moving up is the change the FAQ says can usually happen with little or no downtime. An emergency restore from a 7-day window that has already closed is the change that cannot. Buy the extra weeks at £5 a month when the business would need day eight, or use Enterprise's 28 days of hourly copies. Rehearse one restore. A backup you have never restored is a file you hope is complete.
Keep the application and the database on the products that match them. Plesk can hold a small site database. A self-managed virtual server can run an engine you patch yourself, with root. This service is the one where we run MySQL, MariaDB, PostgreSQL, MongoDB or SQL Server 2022 Standard on the Colchester cluster, send you an address, and keep the retention you selected. Kubernetes can call the hosted MariaDB. It should not be given a private address in some other namespace and expected to pass the scaler rules. The password stays out of the image either way.
The firewall choice closes the order. None extra, rules on the managed cluster firewall, or a Cisco appliance you manage. Enterprise adds VPN and encryption at rest on top of the SSL and the patches every plan has. ASA and Firepower on the firewall page are for a rack, not a default attachment to a £29.99 database. When all of that is written down, the checkout is mechanical. When the engine is still a guess, stop. A MariaDB backup of a PostgreSQL application is not a migration. It is a second database the application will not open.
What to have ready at checkout
Bring the engine name, the current size of the data, and how many days back you would restore. Bring the list of IPs that will connect, including a Kubernetes namespace if that is the client, and say whether a node failure may change the address. Those answers select MariaDB, MySQL, PostgreSQL, MongoDB or SQL Server 2022 Standard, then Starter, Standard, Business or Enterprise, then extra weeks of retention if you are not on Enterprise, then the pair or the three-node cluster if you are not already on the included Enterprise cluster. They also select the firewall tier. Without them the checkout is a guess, and a guess is how a PostgreSQL application is pointed at MariaDB. With them, the email that follows provisioning has an address you already know how to use, and the first backup has a retention you already chose.
The Colchester cluster, the SSD plans, and the address we email you are the service. The engine is the choice you bring.