NotesSystem
Temporal
Also: Temporal.io
An open-source durable-execution platform, founded in 2019 on a fork of Uber's Cadence. Its defining choice is event-history replay: workflow code is re-executed against a recorded history to recover from any crash.
Design choices
- Durable Execution: workflows are code; activities (side effects) are retried at least once and should be idempotent; results are appended to an event history.
- Determinism by contract: replay compares commands to history, so workflow code must avoid clocks, randomness and direct I/O; a mismatch is a non-determinism error.
- Versioning: patch markers (
Patched/GetVersion) or Worker Versioning route old executions to compatible code. - Workflow Orchestration: durable timers, signals and child workflows coordinate long-running, multi-service processes.
Notable trade-offs
- History limits: the service terminates a workflow whose history exceeds 51,200 events or 50 MB, and warns from 10,240 events or 10 MB (defaults; fixed on Temporal Cloud), so long-running ones must continue-as-new.
- A cluster (or Temporal Cloud) to run, and a programming model the whole team has to learn.
- In return: Netflix reported transient deployment failures falling from about 4% to 0.0001% after moving cloud operations onto it (2025).