Resource Requests and Limits

Overview

This page details how to set resource requests and limits for Appian on Kubernetes for self-managed customers. It provides per-component sizing guidelines for different classes of sites and details how to configure resource requests and limits—and related configurations, if applicable.

Zookeeper

Zookeeper resource requests and limits are defined via the .spec.zookeeper.resources field on Appian custom resources. Zookeeper's start scripts set the Xms and Xmx flags when starting the Zookeeper JVM to the values of the ZK_HEAP_MIN and ZK_HEAP_MAX environment variables, respectively. If the environment variables are omitted, the flags default to 128m and 256m. If set, .spec.zookeeper.resources.requests.memory and .spec.zookeeper.resources.limits.memory should be assigned values equivalent to those of the aforementioned environment variables—which can be defined via the .spec.zookeeper.env field—plus a small buffer.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
apiVersion: crd.k8s.appian.com/v1beta1
kind: Appian
metadata:
  name: appian
spec:
  zookeeper:
    env:
      - name: ZK_HEAP_MIN
        value: 128m
      - name: ZK_HEAP_MAX
        value: 256m

    resources:
      requests:
        cpu: 100m
        memory: 307Mi # ZK_HEAP_MAX * 1.2 to account for non-heap memory
      limits:
        cpu: 200m
        memory: 448Mi # ZK_HEAP_MAX * 1.75 to account for non-heap memory

Kafka

Kafka resource requests and limits are defined via the .spec.kafka.resources field on Appian custom resources. Kafka's start scripts set the Xms and Xmx flags when starting the Kafka JVM to the values of the KAFKA_HEAP_MIN and KAFKA_HEAP_MAX environment variables, respectively. If the environment variables are omitted, the flags default to 512m and 1024m. If set, .spec.kafka.resources.requests.memory and .spec.kafka.resources.limits.memory should be assigned values equivalent to those of the aforementioned environment variables—which can be defined via the .spec.kafka.env field—plus a small buffer.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
apiVersion: crd.k8s.appian.com/v1beta1
kind: Appian
metadata:
  name: appian
spec:
  kafka:
    env:
      - name: KAFKA_HEAP_MIN
        value: 512m
      - name: KAFKA_HEAP_MAX
        value: 1024m

    resources:
      requests:
        cpu: 500m
        memory: 1229Mi # KAFKA_HEAP_MAX * 1.2 to account for non-heap memory
      limits:
        cpu: 2000m
        memory: 1792Mi # KAFKA_HEAP_MAX * 1.75 to account for non-heap memory

Search Server

Search Server resource requests and limits are defined via the .spec.searchServer.resources field on Appian custom resources. Search Server's start scripts set the Xms and Xmx flags when starting the Search Server JVM to the value of the SS_HEAP environment variable. If the environment variable is omitted, the flags default to 1g. If set, .spec.searchServer.resources.requests.memory should be assigned a value equivalent to that of the aforementioned environment variable—which can be defined via the .spec.searchServer.env field—plus a small buffer. If set, .spec.searchServer.resources.limits.memory should be assigned a value at least double that of the aforementioned environment variable and .spec.searchServer.resources.requests.memory.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
apiVersion: crd.k8s.appian.com/v1beta1
kind: Appian
metadata:
  name: appian
spec:
  searchServer:
    env:
      - name: SS_HEAP
        value: 1280m

    resources:
      requests:
        cpu: 100m
        memory: 1.5Gi # SS_HEAP + 256Mi to account for non-heap memory
      limits:
        cpu: 200m
        memory: 3Gi # Double requests.memory

Data Server

Data server resource requests and limits are defined via the .spec.dataServer.resources field on Appian custom resources.

Additionally, the real-time store count and memory circuit breaker configurations should be set. If left unset, the real-time store count will be set to 2 and the memory circuit breaker will not be enforced.

The real-time store count is configured by setting the rtscount property in the Appian YAML file. The memory circuit breaker, which is used to limit the amount of memory that each real-time store component can use during query execution, is configured using the .spec.customProperties.rts.queryMemoryLimits.circuitBreaker.threshold.bytes field.

For more sizing guidance regarding data server memory, see Memory.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
apiVersion: crd.k8s.appian.com/v1beta1
kind: Appian
metadata:
  name: appian
spec:
  customProperties:
    rts.queryMemoryLimits.circuitBreaker.threshold.bytes: 500000000
  dataServer:
    resources:
      requests:
        cpu: 1000m
        memory: 4Gi
      limits:
        cpu: 2000m
        memory: 10Gi

Service Manager

JVM Memory

Service Manager's start scripts set the Xmx and XX:MaxDirectMemorySize flags when starting the Service Manager JVM to the values of the KOMODO_HEAP_MAX and KOMODO_DIRECT_MEM_MAX environment variables, respectively. If the environment variables are omitted, the flags both default to 8g.

Kubernetes Resources

Service Manager resource requests and limits are defined via the .spec.serviceManager.resources field on Appian custom resources. Resource requests and limits can be overridden on a per-engine basis via the .spec.serviceManager.engineOverrides.[ENGINE].resources fields.

CPU

Memory

Each Service Manager pod contains one of the Appian Engines, which have unpredictable and unbounded memory usage. As such, the recommended values for .spec.serviceManager.resources.limits.memory are set very high to prevent the engine from being killed during a spike. Optionally the memory limit for the Service Manager pods can be omitted entirely. While it is not necessary to have available memory to cover the limits for each pod, providing worker nodes without enough memory headroom could lead to availability issues. Configuring .spec.serviceManager.resources.requests.memory will depend on a number of factors, but as a general guideline for new sites refer to the table below.

The formula for calculating the memory requests size of an existing engine is KOMODO_DIRECT_MEM_MAX + KOMODO_HEAP_MAX + (size of kdb on disk * 2.5). The sizing table above can still be used to set the memory limits.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
apiVersion: crd.k8s.appian.com/v1beta1
kind: Appian
metadata:
  name: appian
spec:
  serviceManager:
    env:
      - name: KOMODO_DIRECT_MEM_MAX
        value: 512m
      - name: KOMODO_HEAP_MAX
        value: 1g
    resources:
      requests:
        cpu: 200m
        memory: 2.5Gi
      limits:
        cpu: 2000m
        memory: 32Gi

In addition to setting .spec.serviceManager.resources.requests.memory, it is recommended that memory requests be set on a per-engine basis to further optimize your deployment. This can be done using each individual .spec.serviceManager.engineOverrides.[ENGINE].resources.requests.memory setting.

For example, if you generally have smaller engines but a big execution engine, you could optimize using a configuration similar to the following:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
apiVersion: crd.k8s.appian.com/v1beta1
kind: Appian
metadata:
  name: appian
spec:
  serviceManager:
    env:
      - name: KOMODO_DIRECT_MEM_MAX
        value: 512m
      - name: KOMODO_HEAP_MAX
        value: 1g
    resources:
      requests:
        cpu: 200m
        memory: 2.5Gi
      limits:
        cpu: 2000m
        memory: 32Gi
    engineOverrides:
      execution:
        requests:
          cpu: 1600m
          memory: 12.5Gi
        limits:
          cpu: 3000m
          memory: 128Gi

Webapp

Webapp resource requests and limits are defined via the .spec.webapp.resources field on Appian custom resources. Webapp's start scripts set the Xmx flags when starting the Webapp JVM to the value of the conf.appserver.maxHeapSize custom property. If the custom property is omitted, the flag defaults to 4096m. If set, .spec.webapp.resources.requests.memory and .spec.webapp.resources.limits.memory should be assigned values equivalent to those of the aforementioned custom properties—which can be defined via .spec.customProperties—plus a small buffer.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
apiVersion: crd.k8s.appian.com/v1beta1
kind: Appian
metadata:
  name: appian
spec:
  customProperties:
    conf.appserver.maxHeapSize: 3072m
  webapp:
    resources:
      requests:
        cpu: 100m
        memory: 4.5Gi
      limits:
        cpu: 1000m
        memory: 6Gi

RPA

CPU requests and limits

RPA resource requests and limits are defined via the .spec.rpa.resources field on Appian custom resources. CPU requests should be set to 1000m. CPU limits should be set to match the number of CPUs the site node has.

Memory allocation

RPA's start scripts set the Xmx flags when starting the RPA JVM to the value of the rpa.max.memory RPA property. If the RPA property is omitted, the flag defaults to 2048m.

.spec.rpa.resources.requests.memory should be assigned a value proportional to rpa.max.memory and can be specified via .spec.rpa.properties.

We recommend applying a small buffer to .spec.rpa.resources.requests.memory to account for non-heap memory (rpa.max.memory * 1.2)

We do not recommend putting explicit limits on RPA memory in order to leverage swap.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
apiVersion: crd.k8s.appian.com/v1beta1
kind: Appian
metadata:
  name: appian
spec:
  rpa:
    resources:
      requests:
        cpu: 1000m
        memory: 5.0Gi
      limits:
        cpu: 4000m
    properties:
      rpa.max.memory: 4096

Feedback