For the complete documentation index, see llms.txt. This page is also available as Markdown.

LWH storage sizing

Size /opt/wekahome/data for Local WEKA Home deployments on K3s. Capacity depends on monitored WEKA processes and statistics retention. This storage holds time-series statistics, events, and queues.

Storage sizing summary

  • Provision approximately 1 GiB per monitored WEKA process per month of statistics retention.

  • Add 40 GiB for events, queues, and databases.

  • Size for the retention window plus one month. VictoriaMetrics removes complete monthly partitions.

Each process runs on one CPU core, so the monitored process count is the number of cores used on the cluster backends for the Management, Frontend, Compute, and Drives roles, plus the cores used on the clients. This is the same count that wekaNodesMonitored takes.

Storage requirements by process count

Use these capacities for the default 30-day statistics retention. Values include VictoriaMetrics, Postgres, and queue storage.

Monitored WEKA processes
/opt/wekahome/data

100

~140 GiB

250

~290 GiB

500

~540 GiB

1,000

~1 TiB

2,000

~2 TiB

5,000

~5 TiB

10,000

~10 TiB

20,000

~20 TiB

For a different retention, multiply the per-process capacity by (retention_days + 31) / 61.

This is approximately 0.7 for 14 days and 2 for 90 days. For 3,000 processes at 45-day retention, allocate approximately 3.8 TiB.

Storage consumers

Consumer
Scales with
Share

VictoriaMetrics statistics

Processes × retention

Approximately 99%

Events DB (Postgres)

event activity, bounded by retention

small, bounded

Main and support databases

Fixed

Negligible

FSQ and NATS queues

Transient backlog

Tens of GiB

VictoriaMetrics statistics consume most of the storage.

Statistics storage calculation

Active series: A unique time-series measurement stored by VictoriaMetrics.

VictoriaMetrics capacity depends on active series, samples, and sample size. LWH uses approximately 13 MiB per monitored process each day. This estimate uses 60-second raw resolution and no downsampling.

Use this formula. It includes monthly-partition peak usage and VictoriaMetrics merge headroom:

This simplifies to approximately 1 GiB per process per month of retention.

The estimate is validated against these environments:

  • Production fleet: 124 TiB across six vmstorage shards for 336,000 processes. Usage was approximately 12.5 MiB per process daily.

  • Cluster simulator: 6,400 processes used 24.5 GB in 6.5 hours. Usage was approximately 13.4 MiB per process daily.

Storage requirements by retention period

Values include approximately 40 GiB for events, queues, and databases.

Monitored processes
14 days
30 days (default)
90 days

100

~110 GiB

~140 GiB

~240 GiB

500

~410 GiB

~540 GiB

~1.1 TiB

1,000

~780 GiB

~1.05 TiB

~2 TiB

5,000

~3.7 TiB

~5 TiB

~9.8 TiB

20,000

~14.5 TiB

~20 TiB

~39 TiB

Storage requirements by resource preset

Resource presets use the expected monitored process count:

Preset
wekaNodesMonitored (processes)
/opt/wekahome/data @ 30-day retention

small

≤ 1,000

up to ~1 TiB

medium

1,000 – 5,000

~1 – 5 TiB

large

5,000 – 10,000

~5 – 10 TiB

xlarge

10,000+

10 TiB+

VictoriaMetrics partition behavior

VictoriaMetrics stores data in monthly partitions. It removes a partition only after all data exceeds retention. A 30-day retention period can therefore use one or two months of storage. Peak use occurs immediately after a new month begins.

wekaNodesMonitored selects a resource preset. It does not change retention. Set retention with retentionDays.stats. The default is 30 days.

Events database capacity

The events database uses garbage collection. Its size eventually stabilizes:

  • Raw events: Deleted after retentionDays.events. The default is 30 days.

  • Hourly aggregates: Retained for 365 days. The database stabilizes after approximately one year.

Typical deployments use less than the default 20 GiB events volume. Do not disable events retention. Without a retention period, events storage grows without limit.

Validate capacity after deployment

Run homecli local diagnose to view VictoriaMetrics data_size and monitored process count. Calculate the observed daily rate:

Measure after several weeks. Adjust capacity using the observed rate. TimeToFull provides early warning before storage fills.

Sizing considerations

  • Role mix: I/O-intensive deployments use approximately 13 MiB per process daily. Client-heavy deployments use less. Management-heavy deployments use more.

  • Retention: LWH stores full-resolution statistics without downsampling. Doubling retention approximately doubles statistics storage.

  • Measured usage: Use homecli local diagnose to refine estimates for your deployment.

Last updated