Skip to content

perf(selfhost): profile stack resources and tune defaults #1828

Description

@JSONbored

Parent: #1819

Context

The self-host stack includes the app, Redis, optional Qdrant, optional Ollama, observability services, Grafana, Loki, Tempo, runners, backup services, and optional reverse-proxy/tailnet pieces. The default stack should be practical for maintainers without wasting resources or hiding heavyweight tradeoffs.

This issue profiles the stack and tunes defaults before official release.

Requirements

  • Measure CPU, memory, disk, image size, startup time, and steady-state behavior across minimal, observability, qdrant, ollama, runners, and backup profiles.
  • Identify services that should remain opt-in, services that need resource limits, and defaults that should change before release.
  • Review Redis maxmemory policy, Prometheus/Loki/Tempo retention, Grafana plugin install overhead, Qdrant/Ollama startup expectations, and queue concurrency defaults.
  • Audit image size and bundled tools; decide whether a smaller image variant is needed.
  • Keep beginner-friendly defaults without pretending heavy profiles are free.
  • Document expected resource envelopes and warning signs.

Deliverables

  • Resource profile matrix for common deployment modes.
  • Recommended compose defaults and optional resource limits.
  • Image-size audit and optimization recommendations.
  • Dashboard or docs updates showing how to inspect resource pressure.
  • Follow-up PRs for any safe tuning changes.

Acceptance criteria

  • Operators can choose profiles with realistic CPU/memory/disk expectations.
  • Minimal self-host installs do not pull or run heavyweight optional services.
  • Observability retention and runner profiles are documented with operational costs.
  • Any resource limit or default change preserves reliability under normal review load.

Validation

  • Capture representative docker stats / image-size / disk-use snapshots from a non-sensitive environment.
  • Run readiness and focused self-host smoke checks after tuning changes.
  • Confirm docs build and relevant tests pass.

Ownership

Maintainer-only. No gittensor:* labels.

Metadata

Metadata

Assignees

No one assigned

    Labels

    gittensor:bugGittensor-scored bug fix — scores a 0.05x multiplier.

    Projects

    Status
    Done

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions