feat(default-memory-limits): set default memory limits

[ryanhristovski]

Fixes #4882

• sets default memory limits for shutdown manager to 32Mi and default containers to 512Mi (== to limit)
• applied calculated max heap size envoy config to test data for new limit (512Mi)

[zirain]

why these changes happen?

[ryanhristovski]

It's set based on the memory limit by default, since memory limits weren't defined before they weren't being applied

Code here

gateway/internal/infrastructure/kubernetes/proxy/resource.go

Lines 356 to 369 in 2a10d47

	// calculateMaxHeapSizeBytes calculates the maximum heap size in bytes as 80% of Envoy container memory limits.
	// In case no limits are defined '0' is returned, which means no heap size limit is set.
	func calculateMaxHeapSizeBytes(envoyResourceRequirements *corev1.ResourceRequirements) uint64 {
	if envoyResourceRequirements == nil || envoyResourceRequirements.Limits == nil {
	return 0
	}
	if memLimit, ok := envoyResourceRequirements.Limits[corev1.ResourceMemory]; ok {
	memLimitBytes := memLimit.Value()
	return uint64(float64(memLimitBytes) * 0.8)
	}
	return 0
	}

[zirain]

thanks for clarify

[cnvergence]

LGTM

[zirain]

looks like 32 cause OOM on master.

shutdown-manager:
    Container ID:    containerd://e43b953745ffdc5c2e86cf97c061d9e3560b7d800b15b625ca27d4ba315f9ece
    Image:           docker.io/envoyproxy/gateway-dev:latest
    Image ID:        docker.io/envoyproxy/gateway-dev@sha256:cb9737c83359588d3c10683a90ad835ae62696fb3ef8df20383425048aba1133
    Port:            <none>
    Host Port:       <none>
    SeccompProfile:  RuntimeDefault
    Command:
      envoy-gateway
    Args:
      envoy
      shutdown-manager
      --ready-timeout=15s
    State:          Waiting
      Reason:       CrashLoopBackOff
    Last State:     Terminated
      Reason:       OOMKilled
      Exit Code:    137
      Started:      Sat, 28 Dec 2024 14:15:42 +0800
      Finished:     Sat, 28 Dec 2024 14:15:42 +0800
    Ready:          False
    Restart Count:  5
    Limits:
      memory:  32Mi
    Requests:
      cpu:      10m
      memory:   32Mi
    Liveness:   http-get http://:19002/healthz delay=0s timeout=1s period=10s #success=1 #failure=3
    Readiness:  http-get http://:19002/healthz delay=0s timeout=1s period=10s #success=1 #failure=3
    Startup:    http-get http://:19002/healthz delay=0s timeout=1s period=10s #success=1 #failure=30
    Environment:
      ENVOY_GATEWAY_NAMESPACE:  envoy-gateway-system (v1:metadata.namespace)
      ENVOY_POD_NAME:           envoy-gateway-conformance-infra-all-namespaces-302def45-6brqfc7 (v1:metadata.name)
    Mounts:                     <none>

[ryanhristovski]

@zirain hmm, it's generally not best practice to have memory request & limits not equal to each other - but given that this is a shutdown manager it could make sense to have separate values.

What do you think:

1. set memory request to 32Mi and memory limit to something like 128Mi
   or
2. set both request and limit to 128Mi

I didnt have time to fully benchmark it's resource use but it seems to sit idle around 25Mi and during shutdown (in my environment) it can hit around 70Mi.

I'm leaning #1 due to the nature of the shutdown manager - but this could cause resource issues for users in the future.

[zirain]

can you take a look at #4978?

[zirain]

@ryanhristovski this block CI, I didn't figure out why(test passed on my local).

let's revert it first, please resubmit later if you still want this, sorry for this.