kazeflow plan

Resolve a flow and render its selected dependency plan without invoking asset bodies after the entry has loaded.

Synopsis

kazeflow plan ENTRY \
  [--target NAME ...] \
  [--partition-key KEY ... | --partition-range START END | --empty-partitions] \
  [--max-concurrency N] [--verbose] \
  [--format text|json|mermaid|dot]

Selection options

Option	Meaning
--target NAME	Select a target; repeat for several targets.
--partition-key KEY, --partition KEY	Select one key; repeat as needed. The selected definition normalizes or rejects each key.
--partition-range START END	Select one inclusive bounded range when the selected definition supports ranges.
--empty-partitions	Deliberately select no partition attempts.
--max-concurrency N	Review a positive normalized execution concurrency.
--verbose	Add task/configuration detail to text output only.

Flows may declare named invocation limits with ResourcePolicy and associate assets via @asset(resource="network"). A text plan lists only resources used by the selected dependency closure and annotates each associated task. max_concurrency bounds simultaneous invocations. An optional RateLimit(requests, per_seconds) bounds asset body invocation starts in each rolling time window. Both limits are shared across matching assets and keys within one run. The rate limit counts an asset body start after a materialization miss; it does not count cache hits or individual API requests, and separate runs have independent limits. Resource names are application-chosen metadata and may reveal information, so do not use secrets as names.

Flows may also declare PartitionGroup values for disjoint linear chains of partitioned assets. A group bounds how many selected keys progress through all members before another key is admitted. This key limit is independent of global invocation concurrency and resource limits. The text plan shows selected groups; there are no CLI group flags. Define groups in the trusted Python flow and review them in the plan before running. See Bound progress through a partition chain.

The same selected resource can configure max_attempts and classify_failure for explicit retries. The plan shows the invocation budget and a boolean indicating whether a classifier is configured; it never serializes callback code. The budget includes the first invocation. No classifier means no retry, even if the budget is greater than one.

Partition selectors are mutually exclusive. A partitioned closure requires exactly one selection form; omitting one is a preflight configuration error. Passing any selector to an unpartitioned closure is a usage error. Neither case invokes an asset body.

Graph formats

kazeflow plan daily.py
kazeflow plan daily.py --format mermaid
kazeflow plan daily.py --format dot > flow.dot

Mermaid output is directly pasteable into compatible Markdown:

flowchart LR
    task_0["fetch"]
    task_1["publish (target)"]
    task_0 --> task_1

kazeflow emits graph source but does not install or invoke Mermaid or Graphviz.

JSON boundary

The JSON plan projection is deterministic and intentionally lossy. It reports the selection kind, stable domain, and safe counts rather than serializing arbitrary raw partition keys. Text, JSON, Mermaid, and DOT plans preserve this no-raw-key boundary. Use JSON instead of parsing text whitespace. Plan documents in this unreleased change use schema version 5 and include selected groups without raw keys; check the installed package and see the JSON automation contract for migration guidance and schemas.