Power Automate Flows

Cloud flows are the "citizen-maintainable reaction to a Dataverse event" cell in the where-does-the-logic-run table. Once a flow is part of a delivered solution, it is production code and follows the same ALM and quality expectations as a plugin — this chapter covers the flow-specific rules.

Solution-aware, always

DGT-FLW-010 — Create flows inside a Dataverse solution, never as standalone ("My flows") artifacts. Only solution-aware flows participate in ALM: they deploy through pipelines, use connection references instead of hard-wired personal connections, and read environment variables instead of hard-coded URLs and ids.

Error handling

DGT-FLW-020 — No production flow without error handling. Model a try/catch: group the main logic in a Scope, follow it with a scope configured to run after has failed / has timed out, and notify a monitored channel (not a personal mailbox) from the catch branch. An unhandled failed run that nobody notices is the flow equivalent of a swallowed exception. See Microsoft's error-handling guidance.

Also configure each action's retry policy deliberately rather than accepting the default (exponential, four attempts): for idempotent actions the default is fine or can be extended; for non-idempotent actions (creating records, sending mails) disable the retry and compensate in the catch branch instead — an automatic retry of a non-idempotent action is a duplicate.

Triggers

DGT-FLW-030 — Filter in the trigger, not with a condition as the first action: use trigger conditions, and on Dataverse triggers set Select columns and Filter rows. A flow that triggers on every update and immediately terminates still consumes a run and API requests — and an unfiltered update → flow → update pair is the standard recipe for a trigger loop. See the Dataverse trigger documentation.

Loops & concurrency

DGT-FLW-040 — Treat concurrency and pagination settings as code, not as defaults to discover in production: variables written inside an Apply to each are not thread-safe once concurrency is enabled (use Select/Filter array composition instead of a shared variable, or leave concurrency off); list actions return 100 items by default — raise the pagination limit explicitly when more rows are expected. See Microsoft's limits guidance.

Know what flows are not for

Flows are not a high-volume ETL tool. Power Automate has its own throttling and duration limits (actions per day per license, payload sizes, maximum run duration) that are easy to hit with data-migration-style workloads — those belong in the Azure/data tooling row of the where-does-the-logic-run decision, not in a flow with a loop.