Help

Installation Guide

Kubernetes Requirements - Latest

Updated 13 Nov 2026 3 min read

Product : Monosign

Version : v2026.10 and later

Monofor products operate on a platform-independent container architecture. Consequently, Monofor products support platforms such as Docker, Docker Swarm, Kubernetes, Azure Kubernetes Service (AKS), OpenShift, Amazon Elastic Container Service (ECS), and Google Kubernetes Engine (GKE). The resource requirements necessary for services to be created on these platforms are specified in the tables below.

Up to 1 million users, no pod needs more than 8Gi of memory. A larger installation adds copies of a service, not bigger pods.

Keep the CPU limit of every Monosign application pod (all pods except mq, cache and cache-sentinel) at 2000m or more. With less than 2 cores the .NET runtime turns Server GC off and the health check reports ServerGC as an error. The CPU request can be lower.

Platform services

Name

Instance

Request CPU

Request Memory

Limit CPU

Limit Memory

mq

3 (1 without high availability)

500m

1Gi

4000m

2Gi

cache

3 (1 without high availability)

250m

1Gi

1000m

2Gi

cache-sentinel

3 (only with high availability)

100m

128Mi

500m

256Mi

seed

1

500m

1Gi

2000m

2Gi

api

2 or more

1000m

2Gi

4000m

4Gi

account

2 or more

1000m

2Gi

4000m

4Gi

management

1 (2 for high availability)

1000m

2Gi

4000m

4Gi

access-gateway

1

500m

512Mi

2000m

1Gi

ldap-gateway

1

500m

512Mi

2000m

1Gi

radius-server

1 (2 for high availability)

500m

512Mi

2000m

2Gi

api and account run at least 2 copies. For more sign-in load, add copies; do not make the pods bigger.

Services

The Service runs in three tiers. Each tier is its own deployment of the same image.

Name

Variable

Work

service-critical

MONOSIGN_SERVICE_EXECUTOR_MODES=Critical

Rules, notifications, webhooks

service-standard

MONOSIGN_SERVICE_EXECUTOR_MODES=Standard

User sync, workflow, provisioning

service-batch

MONOSIGN_SERVICE_EXECUTOR_MODES=Batch

Certification, reports, long jobs

Every tier runs 2 copies, so the work continues when one copy fails.

Option 1 - Up to 100K Users

Name

Instance

Request CPU

Request Memory

Limit CPU

Limit Memory

service-critical

2

1000m

2Gi

4000m

4Gi

service-standard

2

1000m

2Gi

4000m

4Gi

service-batch

2

1000m

2Gi

4000m

4Gi

Option 2 - Up to 1M Users

Name

Instance

Request CPU

Request Memory

Limit CPU

Limit Memory

service-critical

2

2000m

4Gi

8000m

8Gi

service-standard

2

2000m

4Gi

8000m

8Gi

service-batch

2

2000m

4Gi

8000m

8Gi

Both options assume up to 25 group memberships per user.

Option 3 - Above 1M Users

Name

Instance

CPU and memory

service-critical

3, with user sharding

Defined per project

service-standard

2

Defined per project

service-batch

2

Defined per project

User sharding splits the users between the Critical copies. Each copy is its own deployment with 1 replica and its own range:

Deployment

MONOSIGN_SERVICE_USER_SHARDING_START_PREFIX

MONOSIGN_SERVICE_USER_SHARDING_END_PREFIX

service-critical-1

0

5

service-critical-2

6

A

service-critical-3

B

F

  • The ranges must cover 0 to F with no gap and no overlap.

  • Each range runs on one copy. The platform restarts a failed copy.

  • Do not set the sharding variables on service-standard or service-batch.

Verify

GET /ping/hc?detail=1 on the API shows:

  • Monosign-ServiceMode: OK … Critical: 2, Standard: 2, Batch: 2

  • Monosign-ServiceSharding: OK … Sharding not enabled. or, with sharding, OK … 3 Shard(s). with the ranges and the user count of each.

Permanent link