<aside> 🧭

Буллеты про Temporal, Cadence, оркестрацию процессов, саги, Step Functions и надёжные long-running workflows.

</aside>

Temporal / Cadence

• Migrated X multi-step business processes to Temporal workflows, cutting stuck-process incidents from X/week to near zero.
• Built Temporal workflows orchestrating X activities across X services with automatic retries and durable state.
• Replaced a hand-rolled state machine with Temporal, removing X lines of retry and recovery code.
• Cut manual operator intervention X% by making X long-running processes self-healing through workflow retries.
• Designed activity retry policies with per-activity backoff and X-attempt limits, recovering X% of transient failures.
• Implemented long-running workflows spanning X days with durable timers, replacing X fragile cron-plus-database flows.
• Built signal and query handlers letting operators inspect and steer X in-flight workflows without database surgery.
• Used continue-as-new to keep event histories under X events on workflows running for X months.
• Implemented child workflows for per-item processing, fanning out to X parallel branches with independent retries.
• Deployed Temporal workers across X task queues with per-queue concurrency limits and isolated failure domains.
• Achieved deterministic workflow versioning across X releases, upgrading logic without breaking X in-flight executions.
• Cut order-processing failure rate from X% to X% by moving orchestration from ad-hoc queues to Temporal.
• Built local activities for X fast side-effect-free steps, cutting workflow latency by Xms per execution.
• Instrumented workflow metrics (latency, failure rate, backlog) across X workflow types, cutting MTTR from Xh to Xmin.
• Self-hosted a Temporal cluster on Kubernetes backed by X, supporting X workflow starts/sec.
• Migrated from self-hosted Temporal to Temporal Cloud, cutting operational toil X hours/week.
• Introduced a workflow testing framework with time-skipping, covering X workflows and cutting regression escapes X%.
• Modelled human-in-the-loop approvals as workflow signals, cutting average approval turnaround from Xd to Xh.