Storage best practices¶

Use this document to plan storage for a COS or COS Lite deployment.

Evaluate storage volume needs¶

Evaluate the telemetry volume needed for your solution and refer to the sizing guide for concrete numbers.

Set up distributed storage¶

In production, do not use hostPath storage (hostpath-storage in MicroK8s; local-storage in Canonical K8s):

PersistentVolumeClaims created by the host path storage provisioner are bound to the local node, so it is impossible to move them to a different node.
A hostpath volume can grow beyond the capacity set in the volume claim manifest.

Canonical K8s¶

Use Ceph CSI. Refer to Canonical Kubernetes snap and charm docs.

MicroK8s¶

Use the rook-ceph add-on together with Microceph. See the Microceph tutorial.

General-purpose storage recommendations¶

The following storage recommendations are based on:

retention period of 30 days;
10k log lines per minute;
400k metrics per minute.

COS¶

Kubernetes persistent volume storage¶

The default storage allocation for charmed persistent volumes is 1GB. The following mount points typically require non-default storage allocations.

Charm	Role	Storage volume	Description	Capacity	Typical unit count
loki-worker-k8s	write or ingester	loki-persisted	WAL for received logs before they are sent off to S3	100GB	3
mimir-worker-k8s	write or ingester	data	WAL for received metrics before they are sent off to S3	50GB	3
tempo-worker	ingester	data	WAL for received traces	100GB	3
grafana-k8s	-	database	Configurations, plugins, user data	10GB	3
alertmanager-k8s	-	data	`nflog` and silences snapshots	1GB	3
opentelemetry-collector-k8s	-	persisted	Self-monitoring queued telemetry	10GB	1
traefik-k8s	-	configurations	Dynamic configuration files (YAML), x509 certificates and keys	1GB	1
cos-configuration-k8s	-	content-from-git	Checked-out content from the git repository	1GB	1

The total Kubernetes persistent volume storage needed by COS depends on the scale of each application, and on the replication count. For the table above, a COS deployment would require 795 GB per replicated storage pool (e.g. MicroCeph).

S3 buckets¶

Telemetry is stored in S3 for long-term storage, decoupling storage from compute and supporting cost-effective retention policies. The S3 Integrator provides integration with S3-compatible object storage backends (e.g. Rados Gateway). The object storage should be backed by a 3x replication pool.

Bucket name	Bucket size
loki	1TB
mimir	500GB
tempo	200GB

The total object storage needed by COS depends on the replication count. For the table above, a COS deployment would require 1.7 TB per replicated storage pool.

COS Lite¶

Kubernetes persistent volume storage¶

The default storage allocation for charmed persistent volumes is 1GB. The following mount points typically require non-default storage allocations.

Charm	Storage volume	Description	Capacity	Typical unit count
loki-k8s	loki-chunks	WAL for received logs	100GB	3
prometheus-k8s	database	WAL for received metrics	50GB	3
grafana-k8s	database	Configurations, plugins, user data	10GB	1
alertmanager-k8s	data	`nflog` and silences snapshots	1GB	3
traefik-k8s	configurations	Dynamic configuration files (YAML), x509 certificates and keys	1GB	1
cos-configuration-k8s	content-from-git	Checked-out content from the git repository	1GB	1

The total Kubernetes persistent volume storage needed by COS depends on the scale of each application, and on the replication count. For the table above, a COS Lite deployment would require 465 GB per replicated storage pool (e.g. MicroCeph).