Transactions: 2PC, 3PC, and Coordinators You Can't Trust

Monthly research note. Theme: Distributed Systems Under Failure.

TL;DR

A focused memo on Transactions: 2PC, 3PC, and Coordinators You Can't Trust: define the model, state the properties, then design the system so those properties remain true under failure and adversaries.

Key takeaways

• Mixed-version operation is the default; upgrades must preserve invariants.
• Backpressure and admission control are correctness mechanisms under load.
• Write the safety property first; liveness is always conditional on timing assumptions.
• Automate guardrails; humans are for judgment, not for consistent enforcement.
• Define safety properties before performance goals.

Why this matters

• Operational simplicity is a security property: fewer modes, fewer surprises.
• If your protocol isn’t testable under reordering, it isn’t deployable.
• State compaction and snapshots are where correctness goes to die quietly.
• Observability must explain protocol state, not just latency.

Key questions

• What is the failure model (crash, byzantine, partitions, reordering)?
• What does “read” mean under replication lag?
• Which safety property is non-negotiable (no double-commit, no forks, no split brain)?
• How do clients discover leaders safely (and what happens during flaps)?
• Which components require determinism for reproducibility?
• What is your reconfiguration model (joint consensus, epochs, leases)?

Assumptions

• Delays are unbounded during incidents; timeouts are guesses.
• Nodes restart with partial state unless you prove durability.
• Workload is skewed: hot keys exist and dominate.
• Packets can be duplicated and reordered; acks can be lost.

Non-goals

• Pretending backpressure is an implementation detail.
• Assuming the network eventually behaves “nicely” under load.

Model & invariants

A common safety shape for replicated logs:

Write down the safety property first. If it’s not written, it’s not implemented.

Liveness is always conditional: specify when progress is expected and what you do otherwise.

Security properties

• Replay resistance: duplicated inputs do not change outcomes.
• Authenticity: actions are bound to identity and purpose.
• Evidence: critical actions emit verifiable audit events.
• Least authority: privileges are scoped by purpose and time.

Failure modes

• Mixed-version behavior that violates assumptions silently.
• Resource exhaustion (CPU/bandwidth/storage) turning into correctness failures.
• Observability gaps during incidents (missing evidence).
• Recovery paths that only work when nothing is broken.

Design sketch

flowchart TD
  client["Client"] --> leader["Leader"]
  leader --> log["Replicated Log"]
  log --> snap["Snapshot"]
  snap --> recover["Recovery / Catch-up"]
  leader --> reconfig["Reconfiguration"]

Implementation notes

Make the state machine explicit; then make persistence and networking boring.

Operational invariants to monitor:
- leader_changes_per_minute
- commit_index_monotonic
- snapshot_install_failures
- quorum_acks_latency_p99
- rejected_requests_due_to_admission_control

Verification strategy

• Model checking the smallest core (timeouts, election, reconfiguration).
• Jepsen-style fault injection: partitions + reordering + client retries.
• Upgrade tests: mixed versions and rolling deploy invariants.
• Stress + skew tests: hot keys, slow disks, noisy neighbors.
• Linearizability checks for read/write APIs that claim it.

Operational notes

• Expose protocol state: term/epoch, leader, commit index, config version.
• Treat compaction and snapshot install as first-class SLOs.
• Rehearse region failover and reconfiguration under load.
• Prefer monotonic time sources for leases; alert on clock discontinuities.
• Make client behavior part of the system: document retry semantics.

What to monitor

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

Rollback plan

• Define an explicit rollback trigger (metrics + thresholds).
• 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.
• Prefer backward-compatible changes; avoid “flag day” upgrades.

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.
• In Search of an Understandable Consensus Algorithm (Raft) (2) — Consensus with explicit state machines and practical tradeoffs.
  • Evidence: Track term/commitIndex as explicit evidence; test leader changes and log conflicts as part of rollback behavior.

Open questions

• Which invariants are violated first under overload: latency, availability, or correctness?
• Where does your protocol assume synchrony without admitting it?
• How do you prevent “operator fixes” from changing safety properties?
• What is the worst-case recovery time after a leader + disk failure?

Checklist

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

Further reading

• Paxos Made Simple (Lamport) — Agreement basics and the invariants that matter.
• In Search of an Understandable Consensus Algorithm (Raft) — Consensus with explicit state machines and practical tradeoffs.
• Time, Clocks, and the Ordering of Events (Lamport) — Causality, ordering, and why clocks are tricky.
• Jepsen — Testing correctness under partitions and faults.
• Learn TLA+ — Practical entry point for specification and model checking.
• Site Reliability Engineering (Google) — Error budgets, incident response, and reliability as an engineering discipline.