Observability at Scale: Traces, Cardinality, and Cost

Monthly research note. Theme: DevSecOps & Resilience Engineering.

TL;DR

A focused memo on Observability at Scale: Traces, Cardinality, and Cost: define the model, state the properties, then design the system so those properties remain true under failure and adversaries.

Key takeaways

• Provenance is a cryptographic statement; ship evidence with artifacts.
• Treat CI/CD as attacker-controlled until proven otherwise; minimize secrets and privileges.
• Short-lived credentials (OIDC) beat long-lived tokens in pipelines.
• Define safety properties before performance goals.
• Write assumptions down; treat them as interfaces.

Why this matters

• Infrastructure-as-code without policy is just scripting the attack surface.
• Policy drift is the default; guardrails must be automated and enforced.
• Reproducibility is how you know what you shipped is what you built.
• Runtime security needs evidence pipelines, not just dashboards.

Key questions

• What is the minimum set of humans who can ship to production?
• Where do you enforce policy (pre-merge, build, deploy, runtime)?
• How do you rehearse incident response as code (runbooks, chaos, drills)?
• How do you do safe rollouts (canary, blast-radius, rapid rollback)?
• Which signals prove correctness (not just availability) in production?
• What is your supply-chain threat model (dependency poisoning, CI compromise)?

Assumptions

• Policy enforcement must be consistent across environments.
• Dependencies can be compromised upstream (typosquatting, maintainer takeover).
• CI runners are exposed to untrusted code (PRs, dependencies).
• Observability pipelines can be attacked (log injection, PII leaks).

Non-goals

• Manual policy enforcement or manual security review as the only control.
• Assuming deploy equals success without runtime evidence.

Model & invariants

Build provenance is a cryptographic statement:

Make provenance verifiable: “what built this” must be cryptographically bound.

Policy should be code with diffs and reviews—guardrails, not guidelines.

Security properties

• Downgrade resistance: negotiation can’t silently weaken security posture.
• Authenticity: actions are bound to identity and purpose.
• Least authority: privileges are scoped by purpose and time.
• Integrity: invalid transitions are rejected (and detectable).

Failure modes

• Recovery paths that only work when nothing is broken.
• Mixed-version behavior that violates assumptions silently.
• Timeout ambiguity causing double-apply or partial state transitions.
• Resource exhaustion (CPU/bandwidth/storage) turning into correctness failures.

Design sketch

flowchart LR
  src["Source"] --> build["Build (reproducible)"]
  build --> attest["Attestation"]
  attest --> scan["SAST/DAST/SCA"]
  scan --> deploy["Deploy (policy gates)"]
  deploy --> runtime["Runtime Policy + Observability"]

Implementation notes

Build systems that can prove what happened after an incident.

// Treat CI as untrusted: keep tokens short-lived and scoped.
type Token struct {
  Value string
  ExpiresAtUnix int64
  Scope string
}

Verification strategy

• Dependency tampering drills: lockfile changes, integrity failures.
• Runtime conformance: detect drift between desired and actual state.
• Pipeline attack simulations: compromise a runner and measure blast radius.
• Policy tests: unit tests for policy-as-code rules.
• Rollback tests as part of release (not “if needed”).

Operational notes

• Continuously scan and inventory dependencies; prioritize by exposure.
• Keep a provenance trail for every artifact deployed to production.
• Rehearse incident response for the pipeline itself.
• Treat policy changes as security-sensitive deploys (review + rollout).
• Audit who can ship and how; remove implicit paths.

What to monitor

• Retry/timeout rates by endpoint and client cohort.
• Admission-control / rate-limit rejections (by reason).
• Invariant violation rate (should be ~0).
• Error budget burn + tail latency under load.
• Rollback events and the conditions that triggered them.

Rollback plan

• Define an explicit rollback trigger (metrics + thresholds).
• Prefer backward-compatible changes; avoid “flag day” upgrades.
• Use canaries and staged rollout; stop early when signals degrade.
• Keep dual-write / dual-verify windows where appropriate.
• Preserve evidence (configs, artifacts, audit logs) to reconstruct what changed.

Evidence

• Site Reliability Engineering (Google) (1) — Error budgets, incident response, and reliability as an engineering discipline.
  • Evidence: Error budgets and incident response are correctness controls; tie monitoring and rollback triggers to SLO burn.
• Jepsen (2) — Fault injection and correctness testing for distributed systems.
  • Evidence: Turn faults into test cases; prioritize partition and clock-skew scenarios that violate user-visible guarantees.

Open questions

• What is the smallest CI compromise that becomes a prod compromise today?
• Which deploy actions are irreversible and how do you mitigate that?
• Can you answer “what code is running” with cryptographic evidence?
• How quickly can you revoke all pipeline credentials in an incident?

Checklist

• Telemetry captures correctness signals.
• Assumptions listed and reviewed.
• Safety properties stated as invariants.
• Rollback plan rehearsed and automated.
• Failure modes enumerated with mitigations.
• Costs bounded (CPU/memory/bandwidth) under adversarial inputs.

Further reading

• Sigstore — Signing and verifying artifacts at scale with transparency logs.
• SLSA v1.0 Specification — Supply-chain levels and provenance requirements.
• NIST SP 800-218 (SSDF) — Secure software development practices as an engineering framework.
• in-toto — Securing the integrity of software supply chains with attestations.
• Jepsen — Fault injection and correctness testing for distributed systems.
• Site Reliability Engineering (Google) — Error budgets, incident response, and reliability as an engineering discipline.