workflows

Batch APIs: when HTTP 200 does not confirm every item

An outer 200 can mask item failures. Check endpoint semantics, reconcile responses by ID, and gate downstream actions on confirmed results.

An HTTP 200 for a batch may confirm the request envelope without confirming every operation inside it. Before your workflow updates a record, sends a notification or marks a job complete, check the endpoint’s contract and any per-item results.

Short answer

In Microsoft Graph JSON batching, an outer 200 does not mean every subrequest succeeded. Each inner response has its own id and status; responses can arrive in a different order. In contrast, Google Sheets spreadsheets.batchUpdate applies its requests atomically: an invalid request makes the update fail as a whole. Follow the specific endpoint’s semantics, reconcile by ID where applicable and release downstream work only for confirmed outcomes.

A per-item release card

Field Record without sensitive payloads
Submitted ID A key that matches the request to its result
Individual status Success, correctable failure, retry candidate or unknown
Confirmed effect State from the system of record when the response is insufficient
Next action Release, correct, check current state, or retry only a safe item
Owner Person who resolves unknown outcomes

Pulse’s original rule: if a material item lacks a correlatable response or has an uncertain effect, keep the aggregate job pending reconciliation. An outer 200 cannot clear that flag. Avoid storing credentials or full customer payloads in the card.

Read the response before retrying

  1. Identify the failure model. Is the endpoint atomic, partially successful or asynchronous? Graph documents a partially successful batch; Sheets documents an atomic batchUpdate. Do not assume either model for every integration used in the US, Canada or UK.
  2. Match results by ID. Graph says inner responses may be returned out of request order. A positional match could attribute a failure to the wrong customer record. Use the provider’s documented correlation field.
  3. Classify failures before replay. Separate confirmed successes from permission errors, rate limits and unknown outcomes. A permission error requires an access fix; a rate limit calls for the provider’s retry guidance. Check whether an unknown item already created an external effect before sending it again.
  4. Gate the next action. If all items must succeed, block the aggregate downstream step. If partial delivery is acceptable, explicitly list the IDs that advanced and the IDs requiring follow-up.

Example, not a tested integration

A UK support team submits three Microsoft Graph operations, labeled A, B and C. The batch envelope returns 200, while A and C succeed and B reports a permission error. The team does not mark all three as complete. It records A and C, fixes B’s permission and assesses only B for retry. If the next customer message depends on all three changes, it waits. The IDs and outcomes are illustrative; Pulse did not run this workflow.

Stop conditions

Stop if the provider offers no reliable way to match requests and results, if a timeout leaves the side effect unknown and there is no state lookup, or if replay could send another email, charge or record. Establish a system of record and a manual resolution path first.

This guide is about batch outcomes. For missing source records, verify API pagination. For an individual rate limit, handle 429. For repeated effects from redelivery, prevent duplicate actions.

Sources