Rowsafe Cloud
A database server Rowsafe runs for you in its own cloud account, billed by the hour, with every Rowsafe feature included.
No cloud account? Rowsafe Cloud creates a database server for you in Rowsafe's own cloud account and bills you by the hour, never more than the monthly price.
- Clouds: Hetzner, DigitalOcean, Vultr, AWS, OVHcloud, Azure or Google Cloud.
- Databases: PostgreSQL, MySQL, MariaDB, Valkey, ClickHouse, OpenSearch, Qdrant or Meilisearch.
- Included: continuous backups, Proof, Pulse and Guard, from the first minute.
It's the same server as Create a server for me, except Rowsafe holds the cloud account and bills you.
What Rowsafe can and can't reach
We don't log in to your server, and backups are encrypted on it with a passphrase only you have. Because we hold the cloud account, we could reach its disk. To rule that out, create the server in your own cloud account.
Create a Rowsafe Cloud server
- Open Servers, click Add server and choose Create a server for me.
- Under Where should it run?, choose Rowsafe Cloud.
- Pick the Database, a region (it decides the cloud), a size, a name and the version. For PostgreSQL, tick any extensions.
- Add the addresses that may connect.
- Click Create server. The first time, an owner adds a card.
It's ready in about 5 minutes. Then save the backup passphrase and connect your app.
From your coding agent
Claude Code, Codex or another AI agent can create a server when your app needs a database. It tells you the size and price and gets your OK, then creates it as you, right away (an owner's or admin's agent). It puts the app's connection string in .env; Rowsafe never sees the password. The first time, an owner adds a card at a checkout link.
See Need a database? and How AI agents act. An owner can set a monthly budget for agents.
From the terminal
Install the CLI with curl -fsSL https://rowsafe.sh/cli | sh, then rowsafe login (or use a read-write API key). Paid actions and deletions ask first; --yes confirms in scripts.
rowsafe cloud sizes # clouds, regions, sizes, prices, what's sold out now
rowsafe cloud create shop-db --wait # the cheapest size free now; only this computer may connect
rowsafe cloud create shop-cache --engine valkey # or mysql, mariadb, clickhouse, opensearch, qdrant, meilisearch (PostgreSQL is the default)
rowsafe cloud create search-db --extensions vector,postgis # pgvector and PostGIS on from the start
rowsafe env --on shop-db # the app's database and login: DATABASE_URL in .env
rowsafe connect shop-db # psql, as a temporary user removed when you quit
rowsafe cloud allow shop-db 203.0.113.4 # let the app's server connect
rowsafe cloud resize shop-db medium # a Mark first; asks before the restart
rowsafe cloud clone shop-db shop-test --at "10m ago" --delete-after 1d
rowsafe cloud passphrase shop-db --file shop-db-passphrase.txt
rowsafe cloud delete shop-db # asks you to type the name- Who can connect: only this computer, unless you pass
--allow ADDRESS(or--allow none). rowsafe envandrowsafe connectare PostgreSQL only. Elsewhere, make the login in the dashboard (orrowsafe db user add);rowsafe cloud showgives the host and port.- Passwords never pass through Rowsafe.
rowsafe envmakes the password on your computer, writesDATABASE_URLto.env(a new file is readable only by you), adds.envto.gitignorein a git repository, and prints the password only with--print. rowsafe cloud deleterefuses a server whose passphrase was never saved.
All flags: CLI reference.
Databases
Every database costs the same.
| Database | Versions | Apps connect on |
|---|---|---|
| PostgreSQL | 15, 16, 17 or 18 (17 by default), with pgvector, PostGIS and TimescaleDB if you want them | port 5432 |
| MySQL | 8.4 LTS | port 3306 |
| MariaDB | 11.8 LTS (supported until 2030) | port 3306 |
| Valkey | 8, the open-source Redis: Redis clients and redis-cli work with it | port 6380, TLS only (nothing listens on 6379 for apps) |
| ClickHouse | 26.8 LTS (default) or 26.3 LTS | port 9440 (native protocol) and 8443 (HTTPS), TLS only (nothing listens on 9000 or 8123 for apps) |
| OpenSearch | 3, one node | port 9200 (HTTPS), with a user and password |
| Qdrant | 1.19, the open-source vector database | port 6333 (REST) and 6334 (gRPC), with TLS and an API key |
| Meilisearch | 1.54, Community Edition | port 7700 (HTTPS), with an API key |
Network logins always use TLS (1.2 or newer on MySQL and MariaDB) and a password or API key.
Restores: PostgreSQL, MySQL, MariaDB, Valkey and ClickHouse restore to any second. OpenSearch, Qdrant and Meilisearch restore to a snapshot, taken every 5 minutes (OpenSearch) or 15 minutes (Qdrant, Meilisearch) and at every Mark.
- MySQL needs an Intel or AMD processor, so no AWS sizes (all Arm).
- ClickHouse needs 4 GB of memory. Users made in Databases & users are capped at 40% of memory and one thread per processor.
- OpenSearch needs 4 GB of memory. One node, security plugin on, no demo users (details).
- Qdrant needs 2 GB of memory. Apps use an API key; admin keys always expire (details).
- Meilisearch is the Community Edition (MIT); Enterprise features stay off. Apps use an API key (details).
- Valkey comes from Debian's own packages. The other databases come from their projects' own packages or releases, checked before install.
- PostgreSQL, MySQL and MariaDB only, for now: a standby and Clone to a new server.
- PostgreSQL only, for now: private connections from AWS,
rowsafe envandrowsafe connect. - No MongoDB or Redis: their licenses don't allow Rowsafe to host them. To protect them (or SQLite), install Rowsafe on your own server (Quickstart).
Extensions
On PostgreSQL 15 to 18, three extensions are one switch each:
| Extension | For | Name in SQL |
|---|---|---|
| pgvector | AI search: store embeddings and find the nearest ones | vector |
| PostGIS | Maps and places: locations, distances, areas, spatial indexes | postgis |
| TimescaleDB | Time series: tables split by time automatically, time_bucket | timescaledb |
- At create time: tick them under Extensions (CLI
--extensions vector,postgis,timescaledb). They're on in every database, noCREATE EXTENSIONneeded. - Later: turn one on in Databases & users. TimescaleDB then restarts PostgreSQL once (a few seconds, after a Mark).
- TimescaleDB is the Apache-2.0 edition: hypertables and
time_bucket, but no compression or continuous aggregates. - All three come from the PostgreSQL project's repository (apt.postgresql.org). Rowsafe keeps them working: Pulse offers to put TimescaleDB back if it goes missing, and Update when a database runs an older version.
- Backups, Proof, copies, clones and standbys all work with them.
- Other extensions that ship with PostgreSQL (
pg_trgm,pgcrypto, ...) are a switch away.
Limits:
- No PostgreSQL 15 to 16 upgrade with TimescaleDB through Rowsafe yet. The upgrade check tells you first.
- With a standby, only extensions already installed (and loaded, for TimescaleDB) can be turned on later. Pick them at create time.
Sizes and prices
In USD, Rowsafe included. Billed by the hour (the monthly price ÷ 730, rounded up to a tenth of a cent), and never more than the monthly price in a month.
Hetzner
The lowest prices.
| Size | Server | Per hour | At most a month |
|---|---|---|---|
| Small | 2 vCPU, 4 GB memory, 40 GB disk | $0.014 | $10 |
| Medium | 4 vCPU, 8 GB memory, 80 GB disk | $0.027 | $19 |
| Large | 8 vCPU, 16 GB memory, 160 GB disk | $0.048 | $35 |
| XL | 16 vCPU, 32 GB memory, 320 GB disk | $0.09 | $65 |
Regions: Falkenstein and Nuremberg (Germany), Helsinki (Finland).
Hetzner often runs out of a size for a while (Large and XL often everywhere). The dashboard then shows None free right now until it's back (checked every 15 minutes).
DigitalOcean
| Size | Server | Per hour | At most a month |
|---|---|---|---|
| XS | 1 vCPU, 2 GB memory, 50 GB disk | $0.027 | $19 |
| Small | 2 vCPU, 4 GB memory, 80 GB disk | $0.048 | $35 |
| Medium | 4 vCPU, 8 GB memory, 160 GB disk | $0.09 | $65 |
| Large | 8 vCPU, 16 GB memory, 320 GB disk | $0.172 | $125 |
| XL | 8 vCPU, 32 GB memory, 400 GB disk | $0.328 | $239 |
| 2XL | 16 dedicated vCPU, 64 GB memory, 400 GB disk | $1.04 | $759 |
Regions: New York, San Francisco, Toronto, London, Amsterdam, Frankfurt, Singapore, Bangalore, Sydney.
Vultr
22 cities on six continents. High Performance servers (AMD EPYC, NVMe disks).
| Size | Server | Per hour | At most a month |
|---|---|---|---|
| Small | 2 vCPU, 4 GB memory, 100 GB disk | $0.044 | $32 |
| Medium | 4 vCPU, 8 GB memory, 180 GB disk | $0.085 | $62 |
| Large | 8 vCPU, 16 GB memory, 350 GB disk | $0.164 | $119 |
| XL | 12 vCPU, 24 GB memory, 500 GB disk | $0.254 | $185 |
Regions: New Jersey, Chicago, Dallas, Los Angeles, Silicon Valley, Toronto, Mexico City, Santiago, London, Amsterdam, Frankfurt, Paris, Madrid, Stockholm, Warsaw, Tel Aviv, Johannesburg, Mumbai, Singapore, Tokyo, Seoul, Sydney.
Sold-out sizes show as None free right now (checked every 15 minutes).
LightNode
38 cities, many where other clouds cost more: the Middle East, Latin America, Africa and Asia. Standard servers (shared CPU, 50 GB local SSD).
| Size | Server | Per hour | At most a month |
|---|---|---|---|
| Small | 2 vCPU, 4 GB memory, 50 GB disk | $0.028 | $20 |
| Medium | 4 vCPU, 8 GB memory, 50 GB disk | $0.053 | $38 |
| Large | 8 vCPU, 16 GB memory, 50 GB disk | $0.099 | $72 |
Regions: Dubai, Buenos Aires, Dhaka, Sofia, Bahrain, São Paulo, Toronto, Santiago, Bogotá, Frankfurt, Quito, Marseille, London, Athens, Jakarta, Baghdad, Tokyo, Seoul, Yangon, Mexico City, Kuala Lumpur, Muscat, Lima, Manila, Karachi, Doha, Bucharest, Moscow, Riyadh, Singapore, Bangkok, Istanbul, Silicon Valley (two sites), Washington, Hanoi, Ho Chi Minh City, Johannesburg.
- No size changes: create a server of the size you want and move the database.
- Traffic: 1 TB a month on Small, 2 TB from Medium. Past it LightNode slows the server down; it is never billed.
AWS
| Size | Server | Per hour | At most a month |
|---|---|---|---|
| Small | 2 vCPU, 4 GB memory, 40 GB disk | $0.062 | $45 |
| Medium | 2 vCPU, 8 GB memory, 80 GB disk | $0.117 | $85 |
Large, XL and 2XL are coming back soon.
Regions: N. Virginia, Oregon, Frankfurt, Ireland, Singapore.
Azure
| Size | Server | Per hour | At most a month |
|---|---|---|---|
| Small | 2 vCPU, 4 GB memory, 32 GB disk | $0.076 | $55 |
| Medium | 2 vCPU, 8 GB memory, 64 GB disk | $0.136 | $99 |
| Large | 4 vCPU, 16 GB memory, 128 GB disk | $0.273 | $199 |
| XL | 8 vCPU, 32 GB memory, 256 GB disk | $0.547 | $399 |
Regions: Ireland, Sweden, Frankfurt, London, Paris, Virginia, Washington, Singapore.
Small to Large are burstable: they slow down under hours of constant load. For a database busy all day, pick XL or another cloud.
Google Cloud
| Size | Server | Per hour | At most a month |
|---|---|---|---|
| Small | 2 vCPU, 4 GB memory, 40 GB disk | $0.068 | $49 |
| Medium | 2 vCPU, 8 GB memory, 80 GB disk | $0.122 | $89 |
| Large | 4 vCPU, 16 GB memory, 160 GB disk | $0.24 | $175 |
| XL | 8 vCPU, 32 GB memory, 320 GB disk | $0.465 | $339 |
Regions: Belgium, Frankfurt, London, N. Virginia, Oregon, Singapore.
OVHcloud Flex
OVHcloud's Public Cloud, billed by the hour.
| Size | Server | Per hour | At most a month |
|---|---|---|---|
| Small | 2 vCPU, 4 GB memory, 50 GB disk | $0.035 | $25 |
| Medium | 4 vCPU, 8 GB memory, 50 GB disk | $0.062 | $45 |
| Large | 4 vCPU, 16 GB memory, 100 GB disk | $0.172 | $125 |
| XL | 8 vCPU, 32 GB memory, 200 GB disk | $0.342 | $249 |
| 2XL | 16 vCPU, 64 GB memory, 400 GB disk | $0.684 | $499 |
Regions: Gravelines, Strasbourg, Frankfurt, London, Warsaw, Beauharnois.
OVHcloud Value
A fixed price, billed monthly. Turn on Value in the OVHcloud tab.
| Size | Server | A month |
|---|---|---|
| Small | 2 vCPU, 4 GB memory, 40 GB disk | $9 |
| Medium | 4 vCPU, 8 GB memory, 75 GB disk | $15 |
| Large | 8 vCPU, 24 GB memory, 200 GB disk | $45 |
Regions: Gravelines, Strasbourg, Roubaix, Limburg, London, Warsaw, Milan, Beauharnois, Singapore, Sydney, Mumbai (while OVHcloud has servers free there; the dashboard shows which have some right now).
- Paid a month ahead, one subscription per server, cancel any time. Deleting the server ends it at the end of the paid month.
- Ready in about 10 to 30 minutes.
- Shared CPU. Network up to 500 Mbps (Small), 1 Gbps (Medium) or 3 Gbps (Large).
- No clones, standby or size changes.
Every size includes backups in Rowsafe Storage, Rewind, Proof, Pulse, Guard and a firewall. A Rowsafe Cloud server doesn't count against your plan.
How billing works
- Add a card once. The first server asks an owner for a card: one pay-as-you-go subscription for the organization, through Polar, our merchant of record (it handles sales tax and VAT). An owner adds it under Settings → Billing.
- Every hour a server exists is billed, never more than its monthly price in a month. Polar invoices at the end of each month. OVHcloud Value is billed monthly instead.
- Delete a server and its billing stops.
- If a payment fails or you cancel, nothing is deleted right away. After emails and a grace period, servers are powered off (not billed), then deleted after another notice. Backups stay in Rowsafe Storage while the database stays in Rowsafe.
Copies for a test or an AI agent
Clone to a new server copies a database (now, at any second, or at a Mark) onto a new Rowsafe Cloud server in one click, billed by the hour. Rowsafe can delete it after 4 hours, 1 day or 7 days, with an email an hour before. Three hours on the smallest OVHcloud Flex size cost about $0.09.
Bandwidth
Outbound traffic included each month. Inbound is free, and backups never count.
| Cloud | XS | Small | Medium | Large | XL | 2XL | Above that |
|---|---|---|---|---|---|---|---|
| Hetzner | Unlimited | Unlimited | Unlimited | Unlimited | |||
| DigitalOcean | 2 TB | 4 TB | 5 TB | 6 TB | 10 TB | 7 TB | 1.4¢ per GB |
| Vultr | Unlimited | Unlimited | Unlimited | Unlimited | |||
| OVHcloud Flex | Unlimited | Unlimited | Unlimited | Unlimited | Unlimited | ||
| OVHcloud Value | Unlimited | Unlimited | Unlimited | ||||
| LightNode | 1 TB | 2 TB | 2 TB | Slowed, never billed | |||
| AWS | None | None | None | None | None | 11¢ per GB (14¢ in Singapore) | |
| Azure | None | None | None | None | 11¢ per GB | ||
| Google Cloud | None | None | None | None | 14¢ per GB |
A standby has its own allowance, and what it receives from the primary never counts. The dashboard shows each server's traffic. Extra traffic is invoiced with the hours.
Connect your app
Each server gets a name that stays the same, like x7kq2mfa3pzd.cloud.rowsafe.sh. Use it as your app's host, with the port from Databases. The server's page shows both under Connect your app.
- Click Create database and user (Valkey: Create a login for your app; Qdrant: Make an API key for your app). The password is made on the server and shown to you once. See Databases & users.
- Make sure your app's address is under Who can connect.
About the name:
- It follows the primary within about a minute after a failover or switchover. To switch at once, use the connection strings that follow the primary.
- For reads:
x7kq2mfa3pzd-ro.cloud.rowsafe.shpoints at the standby, if there is one. - IPv6 only when Who can connect includes an IPv6 address.
- It's random, so it doesn't reveal your server's name.
- It works a minute or so after the server is ready. Until then, use the IP address on the page.
Check the server's certificate
Each server gets a Let's Encrypt certificate for both names, renewed by Rowsafe without a restart. Its private key never leaves the server. Use sslmode=verify-full to check that you reach your server and nothing else:
postgresql://USER:[email protected]:5432/DATABASE?sslmode=verify-full&sslrootcert=system| Driver | What to set |
|---|---|
| psql, Python (psycopg), Ruby (pg), PHP and other clients built on libpq | sslmode=verify-full&sslrootcert=system (needs libpq 16 or newer). |
| Node.js (node-postgres) | sslmode=verify-full in the URL (without sslrootcert), or ssl: { rejectUnauthorized: true }. |
| Go (pgx), Java (JDBC) | sslmode=verify-full. |
Until the certificate is installed (a minute or two after the server is ready), or with an older libpq, use sslmode=require. It's still encrypted.
MySQL, MariaDB, Valkey and ClickHouse
Same name, TLS required:
mysql://USER:[email protected]:3306/DATABASE?ssl-mode=REQUIRED
rediss://USER:[email protected]:6380/
clickhouse://USER:[email protected]:9440/DATABASE?secure=true
https://x7kq2mfa3pzd.cloud.rowsafe.sh:8443mysql -h x7kq2mfa3pzd.cloud.rowsafe.sh -P 3306 -u USER -p --ssl-mode=REQUIRED DATABASE # MySQL
mariadb -h x7kq2mfa3pzd.cloud.rowsafe.sh -P 3306 -u USER -p --ssl DATABASE # MariaDB
redis-cli --tls -h x7kq2mfa3pzd.cloud.rowsafe.sh -p 6380 --user USER --askpass # Valkey (valkey-cli too)
clickhouse-client --host x7kq2mfa3pzd.cloud.rowsafe.sh --port 9440 --secure --user USER --ask-password # ClickHouse
curl --user USER https://x7kq2mfa3pzd.cloud.rowsafe.sh:8443/ --data-binary 'SELECT 1' # ClickHouse over HTTPSClickHouse: 9440 is the native protocol (clickhouse-client, Go, clickhouse:// URLs); 8443 is HTTPS (curl, clickhouse-connect, Node.js, JDBC, DBeaver).
OpenSearch, Qdrant and Meilisearch
HTTPS on the same name:
https://USER:[email protected]:9200 # OpenSearch
https://x7kq2mfa3pzd.cloud.rowsafe.sh:6333 # Qdrant REST (gRPC: x7kq2mfa3pzd.cloud.rowsafe.sh:6334, TLS)
https://x7kq2mfa3pzd.cloud.rowsafe.sh:7700 # Meilisearchcurl --user USER https://x7kq2mfa3pzd.cloud.rowsafe.sh:9200/ # OpenSearch
curl -H "api-key: $QDRANT_API_KEY" https://x7kq2mfa3pzd.cloud.rowsafe.sh:6333/collections # Qdrant
curl -H "Authorization: Bearer $MEILISEARCH_KEY" https://x7kq2mfa3pzd.cloud.rowsafe.sh:7700/indexes # Meilisearch- OpenSearch: a user and password from Databases & users.
- Qdrant: an API key in the
api-keyheader, never in the URL. - Meilisearch: a search-only key in browsers, a key that can add documents in your backend.
When the name goes away
Deleting the server removes its name for good. After Take it with you, apps connect to your own server's address. Servers in your own cloud account get no Rowsafe name.
Your passphrase, only yours
Backups are encrypted with a passphrase made on the server. You save a copy through your browser, encrypted end to end. Rowsafe never has it, so without it nobody can restore your backups. The dashboard keeps asking until you confirm you saved it.
Change the size
- On the server's page, click Change size.
- Pick a size. Each shows its price per hour and per month.
- Click Change size now.
The database is offline for a few minutes while the server restarts. Rowsafe saves a Mark first.
- The disk can grow, never shrink. To go smaller, create a smaller server and move the database.
- On Vultr, the restart is a power cut. Committed data is safe; transactions in flight are rolled back.
- If the cloud can't make the change, the server comes back at its old size and the page says why.
- Billing: the new size counts from the next full hour. A month never costs more than the monthly price of the biggest size the server had that month.
Owners and admins can change the size (or rowsafe cloud resize), and so can AI agents acting as them.
A server with a standby
With a standby, Change size resizes both servers, one at a time:
- Rowsafe saves a Mark and resizes the standby. The primary keeps serving.
- Rowsafe switches over. Writes pause for a few seconds (usually under 15). Nothing is lost.
- The old primary becomes the standby and gets the new size.
The servers swap roles and stay that way. The standby must be in sync first, or the change is refused. If something fails before the switchover, nothing changes. If the cloud refuses the second server, click Change size again to finish it, without another switchover.
A pair in your own cloud account can't change size yet: remove the standby, resize, add it again.
With a standby
A standby is a second server of the same size with a live, read-only copy of the database. If the primary goes down, the standby takes over. PostgreSQL, MySQL and MariaDB.
- New server: turn on With a standby. Fail over automatically is ticked.
- Existing server: click Add a standby. Nothing on the primary restarts.
Where the standby runs
| Cloud | Where the standby runs |
|---|---|
| AWS | Another availability zone of the same region. |
| DigitalOcean | In New York and San Francisco, the city's other datacenter. Elsewhere, a separate server in the same datacenter. |
| Azure | A separate server in the same region. |
| Google Cloud | A separate server in the same zone. |
| Hetzner, Vultr, OVHcloud Flex | A separate server in the same datacenter. |
The same datacenter covers a server failing, not a whole-datacenter outage.
Price
Twice the price: the standby is billed like any server of its size. A Small DigitalOcean pair is at most $70 a month. The pair counts as one server against your Rowsafe Cloud limit.
Automatic failover, and what it can lose
Rowsafe promotes the standby only when all of these hold:
- the primary is unreachable for about 3 minutes, for Rowsafe and for the standby;
- the standby was at most about a minute behind;
- nothing planned (a restart, an update, a resize) is going on.
Up to about a minute of the newest changes can be lost. The standby is usually under a second behind. If the primary is cut off from Rowsafe but not from your apps, both can take writes for a while: see Split brain.
Turn it on or off on the database's Standby page. Without it, click Promote when the primary is down.
Connect so your app follows a failover
The server's page shows connection strings that follow the primary: both addresses in one string, and the driver uses the one that accepts writes.
postgresql://USER@PRIMARY_ADDRESS:5432,STANDBY_ADDRESS:5432/DBNAME?target_session_attrs=read-writeJava uses targetServerType=primary. Drivers that take one address (node-postgres) can use the server's name.
MySQL and MariaDB drivers take one address: use the server's name, which follows the primary within about a minute. Reports can read from the standby.
After a failover
On the old primary's page, click Make it the standby again, or Remove the standby to delete it. After a planned switchover, this happens by itself.
Updates and size changes with a standby
One server at a time, with a switchover of a few seconds in between. See Change the size and Maintenance with a standby.
Remove the standby
On either server's page, click Remove the standby and type the server's name. The standby is deleted and its billing stops; the primary keeps running.
Updates and maintenance
Rowsafe keeps Rowsafe Cloud servers patched. What needs no restart installs by itself. What needs a restart waits for your weekly window, with an email the day before.
Security updates install by themselves
Every day, the server installs Debian's security updates:
- The database is never restarted, and its own packages wait for the window.
- The server never reboots. A new kernel waits for the window.
The server's page shows Security updates: installed automatically.
The maintenance window
Weekly, two hours, Sunday 03:00 in the region's local time by default. Change the day, hour or time zone when you create the server or later.
| What | Downtime |
|---|---|
| A PostgreSQL update of the version you run, e.g. 17.5 to 17.6 (never a new major version) | A few seconds |
| PostgreSQL using an updated library (OpenSSL, the C library) | A few seconds (a restart) |
| A new kernel | A minute or two (the server restarts) |
| Security updates still waiting, if any | None |
- A Mark first, then each step, logged as maintenance window. Nothing starts outside the window.
- Emails about 24 hours before (only if there's something to apply) and after.
- Postpone a week skips the next window, once. Apply now runs it right away. Both are in the email and in Updates and maintenance.
Maintenance with a standby
The standby is updated first, then Rowsafe switches over (writes pause for a few seconds), then the old primary is updated. The servers swap roles. Automatic failover waits throughout. If the standby isn't in sync, the window updates in place instead.
Critical security fixes
For a flaw that is being exploited, or that exposes data, Rowsafe staff can mark an update critical:
- Right away: an email and a red banner say what it is and when.
- 7 days later: it's applied in your window (Sunday 03:00 if the window is off), with an email 48 hours before and a Mark first.
This happens even if the window is off or postponed: the one exception. Click Apply now to apply it sooner.
Change or turn off the window
In Updates and maintenance, an owner or admin clicks Change window. With the window off, only critical fixes apply by themselves; the rest waits for Apply now. AI agents can see the window but can't change it, postpone it or apply anything.
Your own servers are not affected
On servers in your own cloud account and your own servers, nothing patches, restarts or reboots on its own. See Update and upgrade your database.
Take it with you
Move the database to your own server any time, with about a minute of downtime:
- Install Rowsafe on your server (Quickstart).
- On the Rowsafe Cloud server's page, click Take it with you and pick your server.
- Your server copies the database and stays in sync, then takes over. Nothing is lost.
- Point your app at your server. The Rowsafe Cloud server is stopped, and deleted when you confirm.
It's the plain open-source database: no lock-in. Remove a standby first. See Move to a new server.
Not yet for OpenSearch, Qdrant or Meilisearch. Their snapshots open without Rowsafe (OpenSearch, Qdrant, Meilisearch).
Private connection from your AWS network
Coming soon
Private connections are built and being switched on. Until then the server's page doesn't offer them.
Apps in your AWS network (VPC) can reach a Rowsafe Cloud server on AWS without going over the internet, through AWS PrivateLink. Only the AWS accounts you name can connect.
- Price: $0.028 an hour while on, at most $20 a month.
- AWS, same region, PostgreSQL only.
- It follows the database after a failover.
Turn it on
- On the server's page, under Private connection from your AWS network, click Turn on.
- Enter your apps' AWS account IDs (12 digits), tick the price and confirm.
- After a few minutes, Rowsafe shows the service name (
com.amazonaws.vpce.…) and the zones (AWS zone IDs likeeuc1-az2).
Connect from your VPC
If your AWS account is connected to Rowsafe, click Create the endpoint for me and pick the VPC and subnets. Older connections need the updated policy.
Connect your app
postgresql://USER:PASSWORD@<your endpoint's DNS name>:5432/postgres?sslmode=requireUse sslmode=require, not verify-full: the certificate doesn't carry your endpoint's name. Once apps connect privately, one switch closes the database's port to the internet.
Only the accounts you named can connect. Removing an account disconnects its endpoints at once.
Turn it off
Click Turn off and confirm. Billing stops, and the endpoints Rowsafe created are deleted. Delete the ones you made yourself in your VPC console.
Limits
- PostgreSQL, MySQL and MariaDB only: a standby and Clone to a new server. PostgreSQL only: private connections.
- MySQL runs on Intel and AMD only (not AWS). ClickHouse and OpenSearch need 4 GB of memory, Qdrant 2 GB.
- OpenSearch, Qdrant and Meilisearch restore to a snapshot, not any second, and can't use Take it with you yet.
- TimescaleDB is the Apache-2.0 edition, with no 15 to 16 upgrade through Rowsafe yet.
- Failover waits about 3 minutes and can lose up to about a minute of changes.
- Changing the size takes the database offline for a few minutes (a few seconds with a standby). Disks never shrink.
- Maintenance restarts the database for seconds, or a minute or two for a new kernel.
- No SSH access. For a server you can log in to, create it in your own cloud account.