Airbyte Activity: Facebook Comscore Streams and Connector Hardening

The Airbyte monorepo logged 300 commits in the last lookback window. The operator visible work clusters around marketing and commerce sources: Meta deprecated DMA breakdowns in Ads Insights, several connectors now validate config before sync, and a handful of streams were removed or retyped. This is not a platform release; it is connector maintenance that can break warehouse schemas if you upgrade blindly.

Meta’s Ads Insights API stopped serving Designated Market Area breakdowns. The Facebook Marketing source connector shipped v6.0.0 in replace deprecated DMA streams with Comscore Market. See the integration docs for stream tables and changelog entries.

The stream rename is a hard break, not an alias:

  • ads_insights_dma becomes ads_insights_comscore_market
  • ads_insights_demographics_dma_region becomes ads_insights_demographics_comscore_market_region
  • The record field dma becomes comscore_market
  • Custom Insights breakdowns drop dma entirely

If normalization is on, each renamed stream lands as a separate destination table. Downstream dbt models, Looker explores, and metric definitions that referenced dma need explicit rewrites. Historical DMA data does not forward migrate; you either backfill from archived tables or accept a gap at the cutover.

The documentation update aligned changelog dates and table names in the integration docs linked above. Read that page before you bump the connector image past 6.0.0.

One day after the Comscore cutover, add type guards in CursorPatch.load_next_page() landed as v6.0.1. Symptom: TypeError when Facebook returned a malformed JSON body during cursor pagination. The root cause was CursorPatch.load_next_page() assuming response_obj.json() always returned a dict with a well formed paging.next chain.

The fix adds isinstance checks on the parsed response and on paging before dereferencing. Non dict responses raise AirbyteTracedException with FailureType.transient_error, which lets the platform retry instead of crashing the worker. The commit also adds 207 lines of unit tests in test_patches.py covering list responses, missing paging keys, and corrupt data arrays.

if not isinstance(response, dict):
    raise AirbyteTracedException(
        message="Facebook Marketing API response is not a JSON object.",
        failure_type=FailureType.transient_error,
    )

Tradeoff: transient errors may surface as retried syncs rather than silent skips. That is preferable to a hard failure mid stream, but watch retry budgets on large Ads Insights backfills.

Two commerce connectors got config layer fixes that prevent footguns before the first API call.

Shopify (v3.5.1). validate and normalize the shop config value rewrote SourceShopify.get_shop_name() in source.py. The connector now accepts either a bare handle (my-store) or a full shop URL and strips scheme, host suffix, and trailing slashes. Malformed input (embedded paths, whitespace, invalid characters) raises ShopifyWrongShopNameError, mapped to config_error so the UI shows a fixable message instead of a vague 401 on the Admin API.

Amazon Seller Partner (v5.7.10). gate vendor/seller streams by account type restructured manifest.yaml with ConditionalStreams. Seller accounts see Orders, OrderItems, and Finances streams. Vendor accounts see VendorOrders, VendorOrdersStatus, and vendor report streams. Previously, a Vendor connection could select Seller only REST endpoints and get HTTP 403 on the connectivity check because CheckStream always probed Orders.

The manifest now uses CheckDynamicStream and documents that stream ordering matters: the first stream in each conditional block is the probe target. Vendor connectivity prefers VendorOrders over VendorDirectFulfillmentShipping because the latter needs a restricted IAM role.

Smaller but still warehouse visible commits rounded out the week.

Jira. remove deprecated issue_field_configurations stream is a breaking change (feat!). The stream was already deprecated; the commit deletes it from the catalog and ERD. Any sync still selecting issue_field_configurations will fail discovery after upgrade. The Jira docs commit also added issue_changelogs to the documented I/O usage list, which matters for capacity planning on busy instances.

MSSQL. map NUMERIC/DECIMAL with scale 0 to Airbyte integer fixes a type coercion edge case. Columns declared as NUMERIC(10,0) or DECIMAL(18,0) previously flowed as floating point JSON numbers. They now map to integer schema types, which avoids precision loss in destinations that respect Airbyte type hints.

Airtable. CDK bump to 7.23.3 in upgrade CDK to 7.23.3 is infrastructure only, but paired doc fixes for the data type map reduce onboarding friction.

  1. Facebook Marketing upgrades. Pin below 6.0.0 until warehouse models map Comscore streams. After cutover, run a full refresh on the two renamed Ads Insights tables and retire any dma keyed metrics.
  2. Amazon Vendor connections. Confirm account_type is set to Vendor in the connection config. Run the connectivity check again; it should probe VendorOrders, not Orders.
  3. Agent connector docs. Several commits synced agent SDK documentation (sync agent connector docs, Agents release notes for June 23). If you run Airbyte Agents alongside OSS connectors, read the release notes before enabling new agent sidecars.

Most other activity in the window is changelog date corrections, metadata URL updates, and automated dependency bumps. Important if you maintain a fork, but low blast radius for standard deployments.