← ClaudeAtlas

yugabytedb-sizinglisted

Calculate and recommend optimal YugabyteDB cluster sizing based on workload inputs. Use this skill whenever the user asks about YugabyteDB cluster sizing, capacity planning, node count, vCPU requirements, storage sizing, memory recommendations, or wants to know how many nodes they need for a given workload. Trigger when user mentions "size my cluster", "how many nodes", "YugabyteDB capacity", "YugabyteDB hardware requirements", QPS with YugabyteDB, or asks about CPU/storage/memory for a YugabyteDB deployment. Also trigger when the user provides workload metrics (QPS, read/write ratio, table size) and asks for infrastructure recommendations. Covers both YSQL (PostgreSQL-compatible) and YCQL (Cassandra-compatible) APIs.
srinivasa-vasu/agent-skills · ★ 3 · DevOps & Infrastructure · score 60
Install: claude install-skill srinivasa-vasu/agent-skills
# YugabyteDB Cluster Sizing Calculator Size a YugabyteDB cluster from workload inputs: node count, vCPU, memory, storage, IOPS, disk throughput, network and failure resilience, targeting ≤65% CPU. **Always run the bundled script** (`scripts/sizing_calc.py`) — never compute by hand. Rules: - RF must be 1, 3, 5 or 7 (the script rejects others). Never recommend RF=1 for production. - **Always state that the sizing is indicative** and that the user should test with their actual workload to fine-tune production sizing. - CPU comes from measured CPU cost per operation (transactional, point-select and YCQL key-value benchmarks), not from latency. Details: [references/cpu-model.md](references/cpu-model.md). ## Workflow 1. **Pick the API.** `--api ysql` (default) or `--api ycql`. Cassandra, CQL, port 9042, partition keys or TTL → YCQL; PostgreSQL, SQL, joins or port 5433 → YSQL. Ask if unclear. 2. **Gather inputs** (table below). Ask for the required ones; use defaults for the rest. 3. **Choose the CPU basis, best first:** 1. *Calibration* — if the user has a running cluster or PoC with a similar mix, ask for its QPS and average CPU% (`--observed-qps`, `--observed-cpu-pct`; plus `--observed-nodes` / `--observed-vcpu-per-node` if its shape differs from the target). 2. *Workload profile* — pick `--workload` from the user's description of their queries. 3. Only if neither is available, let the script auto-select a profile from the read/write mix, and