Understand what is using your Pod’s memory.

KubeMemLens separates charged memory into anonymous memory, filesystem cache, tmpfs/shared memory and other cgroup charges, then adds pressure, limits and event evidence.

Install the alphaView on GitHub

v0.0.1-alpha.3 · published alpha
Evaluation only on disposable or explicitly authorised clusters.

KubeMemLens terminal interface showing memory composition, Pod risk order and current cgroup signals
Qualified Kitty capture at 180 × 50 using synthetic cluster names.

01 · decomposition

One number is not a diagnosis.

kubectl top reports a useful total. KubeMemLens reads cgroup v2 signals to show what contributes to it, with confidence and caveats.

Total charged memory330.55 MiObserved Pod cgroups only
Anonymous
257.58 Mi
Private process memory, heaps, stacks and mmap.
File cache
71.15 Mi
Filesystem-backed pages that may be reclaimable under pressure.
Shmem / tmpfs
0 B
tmpfs, /dev/shm and shared-memory segments.
Other
1.82 Mi
Residual kernel memory, slab and socket charges.

02 · investigation

From number to insight. Evidence first.

A read-only path that follows the evidence already inside the node.

  1. See the composition.

    Separate anonymous memory, cache, shared memory and residual charges.

    KubeMemLens memory composition detail

  2. Keep Kubernetes context.

    Read limits, requests, QoS, ownership and node pressure beside the cgroup evidence.

    KubeMemLens Kubernetes and node context detail

  3. Follow the evidence.

    Move between nodes, workloads, Pods and containers with bounded trends and caveats.

    KubeMemLens Pod risk ordering detail

  4. Stay read-only.

    Compare, recommend and export without applying workload changes.

    KubeMemLens current cgroup signals detail

03 · qualification threshold

Evidence has boundaries.

Exact combinations only. Current unsupported status is a design and evidence boundary, not a permanent provider verdict.

Qualified evidence

  • GKE Standard
  • EKS managed nodes
  • Self-managed containerd
  • Self-managed CRI-O

Currently unsupported

  • Standard AKS
  • GKE Autopilot
  • EKS Fargate
  • AKS virtual nodes
  • Windows worker nodes
  • cgroup v1

Deep mode requires Linux, cgroup v2, DaemonSet scheduling and an enforcing NetworkPolicy-capable CNI.

Read the full compatibility matrix

04 · operator control

Cluster evidence stays under operator control.

KubeMemLens runs inside your cluster. It does not send product telemetry or copy cluster data to a hosted service.

No hosted data service.
Current snapshots and bounded history remain in the collector’s memory.
No product telemetry.
The project does not transmit usage data to a KubeMemLens service.
You control the evidence.
Incident captures stay local and are redacted, not anonymous.

Read-only investigation · no automated changes · operator-managed retention and sharing

Read the security policy

05 · exact alpha

Start with a disposable cluster.

v0.0.1-alpha.3 is for evaluation on disposable or explicitly authorised clusters. It has no production stability or support guarantee.

helm upgrade --install kube-memlens \
  oci://ghcr.io/danushkastanley/charts/kube-memlens \
  --version 0.0.1-alpha.3 \
  --namespace kube-memlens \
  --create-namespace

View installation guideView on GitHub