Infrastructure
Kindo connects to these services, which run as managed services, on your own servers, or in the cluster. Requirements and sizing gives the size of each, and each environment page covers what its provider needs.
Kubernetes cluster
Section titled “Kubernetes cluster”| Attribute | Requirement |
|---|---|
| Version | 1.32 or higher |
| Networking | Any conformant CNI |
| Storage | Dynamic volume provisioning, with a default StorageClass that provisions ReadWriteOnce volumes |
| Ingress | An ingress controller (NGINX, Traefik, AWS Load Balancer Controller, or equivalent) |
| RBAC | Enabled |
| Metrics | metrics-server installed |
| External Secrets Operator | Kindo installs its own. If the cluster already runs one, it must serve external-secrets.io/v1 (ESO 0.17 or later), or installing stops |
| Volume attach limit | The CSI driver allows at least 10 attached volumes per sandbox node at Medium and 20 at Large. Nodes that allow fewer need more sandbox nodes |
| Availability zones | Nodes in every zone the StorageClass provisions volumes in. A pod with a volume restarts only in its volume’s zone |
| Sandbox node group (optional) | A dedicated, tainted node group keeps agent activity off the application nodes. A matching sandbox node selector and toleration are set at install |
PostgreSQL
Section titled “PostgreSQL”| Attribute | Requirement |
|---|---|
| Version | 17.0 or higher (17.4 recommended) |
| Concurrent connections | 200 or more |
Kindo uses a main server, and can use two more:
| Server | Config key | Holds |
|---|---|---|
| Main | inputs.postgres.main | Kindo’s databases |
| Auxiliary (recommended for production) | inputs.postgres.auxiliary | The databases of Kindo’s supporting services. Without one, they go on the main server. |
| Hatchet (optional, recommended) | inputs.postgres.hatchet | The database of Hatchet, Kindo’s task scheduler. Without one, Hatchet uses the auxiliary server, or the main one. |
Each server takes its host, port, TLS mode and administrator login; see the config template. The Kindo CLI creates the databases and the extensions they need. When the provider or DBA restricts who can create extensions, as Google Cloud SQL does, someone allowed to must create Flagsmith’s database and its extension before installing or upgrading:
CREATE DATABASE flagsmith;\c flagsmithCREATE EXTENSION IF NOT EXISTS pg_trgm;Settings for Hatchet
Section titled “Settings for Hatchet”The server that hosts Hatchet’s database uses these settings, at every size. Installing, upgrading, and applying the config warn about any Hatchet setting that differs from these, and name the fix.
| Parameter | Recommended |
|---|---|
autovacuum_max_workers | 10 |
autovacuum_vacuum_scale_factor | 0.1 |
autovacuum_analyze_scale_factor | 0.05 |
autovacuum_vacuum_threshold | 25 |
autovacuum_analyze_threshold | 25 |
autovacuum_vacuum_cost_delay | 10 |
autovacuum_vacuum_cost_limit | 1000 |
| Attribute | Requirement |
|---|---|
| Version | 6.2 or higher |
| Concurrent connections | 100 or more |
| Config key | inputs.redis.url, holding the address and password as one URL |
| Mode | Standalone: one primary address, no sharding. A replica for failover is fine. |
Redis 7 or later is recommended.
On a standalone server, this command returns the line below it:
redis-cli INFO server | grep redis_moderedis_mode:standaloneRabbitMQ
Section titled “RabbitMQ”Kindo uses your own RabbitMQ, or runs one in the cluster. Set one of them:
| Option | Config key |
|---|---|
| Your own | inputs.rabbitmq.bringYourOwn, with its connection URL |
| Run by Kindo | inputs.rabbitmq.bundled: {} |
- Your own. High availability takes a cluster of 3 or more nodes with quorum queues. It needs:
| Attribute | Requirement |
|---|---|
| Version | 3.13 or higher |
| Disk | 20 GB minimum |
| Management plugin | Enabled |
- Run by Kindo. Kindo runs a single RabbitMQ broker with no failover, and creates its credentials. While it restarts, Kindo can’t queue work. It uses 4 GiB of memory, up to 1 CPU, and a 20 GiB volume on the application nodes. The infrastructure checks skip it.
Moving an instance from your own RabbitMQ to Kindo’s starts on an empty broker: work still queued on yours isn’t moved.
Object storage
Section titled “Object storage”Kindo stores uploaded files in one of two kinds of storage:
| Kind | Config key | Notes |
|---|---|---|
| S3 or S3-compatible storage | inputs.storage.s3 | Any store that serves the S3 API, managed or self-hosted. |
| Azure Blob Storage | inputs.storage.azureBlob | Kindo runs the S3 gateway in front of it for you. See Blob storage setup. |
S3 or S3-compatible storage uses a private bucket for user file uploads, with server-side encryption and public access blocked. An example bucket name is kindo-uploads. An s3proxy you run yourself in front of Azure Blob Storage, instead of using Azure Blob Storage here, needs version 4.1.1 or later and Java 25.0.5.1 or a later Java 25 security update. The upstream 4.1.1 image ships Java 25.0.4, so update its Java or use an image that has the update.
Users’ browsers can’t reach the cluster-local address of a store that runs inside the cluster, so the store is published on its own hostname:
| Attribute | Requirement |
|---|---|
| Hostname | HTTPS with a valid certificate and no path prefix, such as https://s3.<domain> |
| Requests | Path-style: https://s3.<domain>/<bucket>/<key> |
| Proxy | Any proxy in front of it forwards the Host header unchanged |
| Config key | Holds |
|---|---|
inputs.storage.s3.publicEndpoint | The published hostname |
inputs.storage.s3.endpoint | The store’s address inside the cluster, in the short form http://<service>.<namespace>:<port>, such as http://s3proxy.s3proxy:9000. Images attached in chat fail to load with the long form ending in .svc.cluster.local. |
Kindo needs a mail relay’s host and the from address to send sign-in codes and other email. The port defaults to 587. A username and password are optional:
| Relay | Config key | Entries |
|---|---|---|
| Authenticated | inputs.smtp.auth.login, with its username and password | SMTP_USER, SMTP_PASSWORD |
| IP-allowlisted | inputs.smtp.auth.none | None |
An IP-allowlisted relay needs the cluster’s outbound addresses allowed on it. Kindo doesn’t remove the entries from your external secrets manager, and a value in the secrets manager takes precedence; remove them there if the relay no longer needs them.
Installing Kindo and every upgrade send a test email from the sender address to the superadmin email address before they deploy Kindo.
If the relay offers STARTTLS, Kindo verifies its certificate against the SMTP host. If the relay is reachable only under a name its certificate doesn’t cover, such as an internal name, set a name the certificate does cover in inputs.smtp.tlsServerName.
Sending limits
Section titled “Sending limits”Kindo refuses email over these limits. A relay with a lower daily quota needs its own limit as well.
| Limit | |
|---|---|
| All email for one organization | 2,500 a day |
| Email to one recipient | 20 a minute and 200 a day |
Each limit resets at the start of the next minute or day, in UTC. Sign-in codes and superadmin dashboard confirmation codes are exempt from these limits. A refused email is dropped, not sent later. When a limit starts refusing email, task-worker-ts logs this event:
email_safety_limit_reachedSyslog receiver
Section titled “Syslog receiver”Optional. Kindo can forward each audit event to a syslog receiver such as a SIEM collector. The receiver supports RFC 5424 and is reachable from the cluster, with outbound traffic to it allowed as listed in Outbound. Observability and log forwarding covers setup.
