The short answer

Managed Redis, and managed Valkey, are for cache, sessions, rate limits, and queues. They are not a stand-in for Postgres. On Unbroker you either run an Apps Redis node, which is an image and a volume you operate, or a managed Redis or Valkey cluster, which the provider operates.

Free includes neither. Managed databases and Apps database templates both require Pro.

What Redis is good at

Redis keeps the working set in memory and can evict keys when that memory is full. That fits a cache of rows you can load again, a counter, a session that can expire, and a queue of jobs a worker can retry.

It is a poor primary store for records that need joins, constraints, and a restore you have already run. Those stay in Postgres. How to connect an API is API with PostgreSQL. How to operate that database is Postgres in production. The smallest published managed Postgres plan is a separate page, managed Postgres, and its price is not a Redis price.

Redis can persist. The official persistence guide describes RDB snapshots, an append-only file, both, or no persistence at all. No persistence is a cache setting. It is not a rule that Redis data can never survive a restart. The source is Redis persistence.

Cache, sessions, and queues

These are three jobs.

A cache is a copy. Give keys a TTL. A miss reads Postgres, or whatever the source is, and writes the copy back. Eviction is acceptable when the copy can be built again.

A session is state the next request must see. One process-local map breaks as soon as a second replica handles the request. A shared Redis lets any replica read the session. If the eviction policy can drop keys without a TTL, or if the process stops before a write reaches disk, that session is gone. Put the production credential in Production, not in All. The scope rules are environment variables.

A queue is work to do later. The consumer is a different service from the web process. Acknowledgement, retry, and a dead-letter path belong to the queue library you run. Unbroker does not add those. The worker guide says to use the mechanisms of that library, and that a loop around an in-memory array is not durable delivery. See background workers.

Persistence: RDB, AOF, and a restart

Redis documents two disk formats, and you can use either, both, or neither.

RDB writes a point-in-time snapshot. Redis can save every N seconds when at least M keys changed, or you can call SAVE or BGSAVE. A snapshot is a compact backup file. If the process stops between snapshots, the writes since the last snapshot are not in that file. Redis says that gap is often minutes, depending on the save points you set.

AOF logs every write and replays the log on startup. appendfsync always syncs on each append. appendfsync everysec is the suggested default and can lose about one second of writes. appendfsync no leaves flushing to the operating system. Redis says that using both RDB and AOF is the general indication when you want data safety comparable to PostgreSQL, and that RDB alone is enough when a few minutes of loss is acceptable.

Eviction is a different loss. A key removed because memory is full is gone even if the last snapshot was perfect. Persistence answers “what is on disk after a restart.” Eviction answers “what is still in memory while the process is up.”

Unbroker does not publish one persistence default for every managed Redis or Valkey plan. Confirm the engine and plan before you treat a restart as safe. The choose-a-model page says managed does not mean every recovery feature is included.

The Apps Redis template is specific. It runs redis:7, port 6379, a volume at /data, a generated password, and this start command: redis-server with --requirepass, --appendonly yes, and --appendfsync always. That is AOF on that volume, synced on each append. It is not an RDB schedule, and the template does not set maxmemory. You still upgrade the image, test a backup, and restore it. High availability is not part of the template.

An Apps node or a managed cluster

Managed Redis or ValkeyApps Redis node
What runsProvider-operated clusterPrivate service from the Redis image
Who operates the hostThe providerUnbroker runs the container. You own more of the engine
ScalePlan and node countService size and the volume
UsersThrough UnbrokerThrough Redis itself
MetricsProvider cluster metricsCPU, memory, health, and logs
BackupsWhatever that engine and plan includeVolume backups to an S3-compatible destination you control
ConnectionProvider credentialsPrivate host and port, or a local tunnel

The practical default in the docs: use a managed database for important production data unless you have a reason to operate the engine. Use an Apps node for development, a smaller workload, or when you want the image, the volume, and the backup destination in the project. The table is Choose a database model.

Supported managed engines at the API layer include PostgreSQL, MySQL, Redis, Valkey, MongoDB, Kafka, and OpenSearch. The create flow can refuse a combination the API lists in general. Apps templates are PostgreSQL, Redis, MongoDB, and MySQL. There is a Redis template. There is not a separate Valkey template. Details are managed databases and database nodes.

Memory and eviction

maxmemory is the cap Redis compares to the dataset. On 64-bit Redis the default is 0, which means no cap. On 32-bit builds there is an implicit 3 GB limit. When usage passes the cap, maxmemory-policy decides what happens.

noeviction refuses commands that would add data and still serves reads. allkeys-lru, allkeys-lfu, allkeys-lrm, and allkeys-random can evict any key. The volatile-* policies only evict keys that have a TTL, and they behave like noeviction when no key has one. Redis suggests allkeys-lru as a common cache default when you have no other reason. A full memory limit is not the same thing as a slow command. Eviction, a fork for a snapshot, and a large write are different costs. The policy list is Redis key eviction.

This page does not pick a byte threshold for your app. Set a cap below the memory the service is allowed to use, and pick a policy that matches the keys you can lose.

Connecting from the app

Keep the password out of the repository. For an Apps database, ubctl apps databases connection <serviceId> prints the connection details. ubctl apps databases tunnel <serviceId> --port 5432 opens a local tunnel. The docs use 5432 as the local port in that example. For a managed cluster, ubctl db connection <databaseId> prints provider credentials as JSON. That command requires infrastructure-management permission, because the payload contains secrets. The reference is Connect to a database.

Variables are All, Production, or Preview. A new preview copies All and Preview and never copies Production-only values. A production Redis URL in All is a production Redis URL in every new preview. The October 6 note is environment variables for production and previews.

Workers and queues

A worker is a process with no inbound HTTP. It requires Pro and runs one replica. The queue, or another durable job store, is a separate part of the design. Add the queue credentials as runtime variables. Do not add a public web service only so a process with no listener has an HTTP health check.

Enqueue a test job that is not production, and follow it to a durable result. A running process does not prove it could authenticate or finish the job. Route repeated failures to the queue’s dead-letter path. Unbroker workers stay at one replica, so concurrency is inside the consumer.

Beacon can tell a browser that the job finished. It is not the queue. The guide is Run a background worker. When the work is a consumer, a schedule, or a single command, the split is Background workers and cron jobs. An n8n install that already uses Redis for jobs is a different product: self-host n8n.

How this works on Unbroker

Create a managed cluster when the provider should operate the host. Choose Redis or Valkey only if the create flow offers it for that region and plan, and read the quoted price before you confirm. This site publishes a starting price for managed PostgreSQL and MySQL. It does not publish a Redis or Valkey dollar figure. Do not copy the Postgres number across.

Create an Apps Redis node when you want redis:7, the /data volume, and AOF with appendfsync always inside the project. The volume can grow up to 50 GB and cannot shrink. Deleting the service deletes the volume. A volume backup to S3-compatible storage is a separate copy, and you should restore one. That limit is persistent volumes.

Both paths require Pro. Pro is $25 per month with $25 of usage credit. The credit does not roll over. If usage stays under the credit, the charge is still $25. Free does not include managed databases or database templates. Current plan rows are on pricing and in Plans.

A Compose file that already has a Redis service is the same split: an Apps node you operate, or a managed cluster, decided per database. The map is Docker Compose in production.

Checklist

  1. Put primary records in Postgres. Put copies, sessions, and jobs in Redis or Valkey.
  2. Decide which keys can be evicted and which must survive a restart.
  3. Set maxmemory and a maxmemory-policy on a cache. Do not assume the 64-bit default is a cap.
  4. For an Apps node, know that the Redis template enables AOF with appendfsync always on /data, and that you still test a restore.
  5. For a managed cluster, read the recovery the create flow actually offers. Do not assume every plan has the same backup.
  6. Read the quoted Redis or Valkey price in that flow. Do not reuse the Postgres price.
  7. Scope the production URL to Production. Point previews at a store you can flush.
  8. Run the consumer as a worker. Keep acknowledgement, retry, and the dead-letter path in the queue.
  9. Use Pro. Free has neither managed databases nor Apps database templates.

FAQ

What is managed Redis used for?

Cache, sessions, rate limits, and queues: data that is hot, temporary, or safe to rebuild. Redis can also persist writes, with RDB snapshots, an AOF, or both. It is still not the place for joins, constraints, and the recovery plan you keep for primary records.

Is Redis a replacement for Postgres?

No. Use Redis when a miss can be rebuilt from somewhere else, or when a queue can retry. Use Postgres when the row has to stay, with constraints and a restore you have tested. Redis's own persistence guide says that using both RDB and AOF is the path toward data safety comparable to PostgreSQL. That is a Redis configuration choice, not a second primary database.

What is the difference between an Apps Redis node and a managed Redis cluster?

An Apps node is a private Redis image, a volume, and a generated password. You upgrade the image, test backups, and restore. A managed cluster is operated by the provider. Unbroker shows status, connection details, resizing, and the metrics that provider exposes. Backups on a managed cluster depend on that engine and plan. Managed does not mean every recovery feature is included.

Does Unbroker support Valkey?

Yes, as a managed engine, next to Redis, at the API layer. The create flow shows the combinations that provider, engine, plan, and region actually offer. An Apps database template exists for Redis, not as a separate Valkey template.

Do you need Pro for managed Redis?

Yes. Managed databases are not on Free. Apps database templates, including the Redis template, are also Pro. Pro is $25 per month with $25 of usage credit. The cluster or the service usage is separate from that fee.

How much does managed Redis cost on Unbroker?

The public plan pages publish a starting price for managed PostgreSQL and MySQL. They do not publish a Redis or Valkey price. The create flow shows the quoted price for the engine, region, and plan it offers. Read that quote before you confirm. Do not reuse the Postgres figure for Redis.

Should a pull request preview use the production Redis?

No. A new preview copies All and Preview variables and does not receive Production-only values. Put the production Redis credential in Production. Point Preview at a Redis you can flush.

Can Redis be a durable job queue with a background worker?

The worker and the queue are separate. A worker on Unbroker runs one replica and requires Pro. Durability comes from the queue's own acknowledgement, retry, and dead-letter behavior, plus whatever persistence that Redis is configured to use. A loop over an in-memory array is not a durable queue.

Is Beacon a job queue?

No. Beacon is realtime messaging. The worker guide says it is not a durable job queue. Use it to notify clients after durable work has finished.