hive@14.1.0

Minor Changes

• #8679
  bd8f0f2
  Thanks @jonathanawesome! - Record which targets a breaking
  change was used in. Schema checks and publishes with conditional breaking changes now store
  per-target request counts for each breaking change and for each of its top affected operations,
  exposed as affectedTargets on SchemaChangeUsageStatistics and
  SchemaChangeUsageStatisticsAffectedOperation in the public GraphQL API.

• #8734
  a6eaf2e
  Thanks @n1ru4l! - Support uploading preprocessed app deployment
  documents. This omits the need of workers and does not make worker availability the bottleneck of
  the app deployment upload process.

• #8736
  367c53b
  Thanks @n1ru4l! - Support routing server-side OAuth requests through
  the request broker so OAuth providers do not see the IP addresses of Hive cluster nodes. This
  applies to GitHub, Google, Okta, and organization OIDC token and user-information requests.

  Self-hosters who need to hide their cluster IP addresses must configure the server service with:

  • REQUEST_BROKER=1
  • REQUEST_BROKER_ENDPOINT=<broker URL>
  • REQUEST_BROKER_SIGNATURE=<shared broker signature>

• #8679
  bd8f0f2
  Thanks @jonathanawesome! - Expose SchemaChange.path and
  SchemaCheckConditionalBreakingChangeMetadata.settings on the public GraphQL API, so the targets
  and schema coordinate behind a breaking change's usage can be read without parsing the change
  message.

Patch Changes

• #8708
  d3bfbf4
  Thanks @jdolle! - Upload persisted documents of an app deployment to
  the primary and mirror object storage concurrently instead of one after the other. Slow object
  storage writes made hive app:create uploads fail with Unexpected error. or time out in the
  CLI.

• #8679
  bd8f0f2
  Thanks @jonathanawesome! - Check access before resolving
  BreakingChangeMetadataTarget.target; load a schema check's changes in Target.schemaChecks when
  only hasSchemaChanges, hasUnapprovedBreakingChanges or cliApprovalMetadata is selected;
  expose breakingChangeFormula, requestCount and the excluded app deployments on a check's
  conditional breaking change settings; and stop storing usage statistics for changes found safe by
  the nullability check.

• #8717
  4512feb
  Thanks @dotansimha! - Upgrade got to v16 to fix ClickHouse
  requests failing with write ECANCELED (Canceled because of SSL destruction) instead of being
  retried.

  Since Node.js v24.20.0, a connection that closes while a request body is still being sent reports
  the error to got in a way that got v14 turned into an immediate failure, skipping the retry
  logic of the ClickHouse clients.

  got v16.1.0 also retries requests with a FormData body, which the ClickHouse client uses for
  audit log inserts and some operations queries.

• #8714
  0a100e4
  Thanks @jonathanawesome! - Fix the schema check page
  freezing on checks with thousands of schema changes. Safe changes are now grouped by kind, and
  breaking changes and each group are paginated 50 at a time, so the 5 types a check added are no
  longer buried among thousands of description edits. A check's schemas are only loaded when a
  schema tab is opened, so the check itself loads without them.

• #8726
  e085b8e
  Thanks @jdolle! - Use one request id per API request. The id is the
  incoming x-request-id header, or a generated UUID when none was sent, and it is now the same
  value on the API's own log lines, the request.id trace attribute, the x-request-id response
  header, and the requests forwarded to the schema, policy and commerce services. Logs and traces of
  those services can therefore be correlated with the API request that triggered them even when no
  proxy supplied an id. Every response from a Hive service now carries that id in the x-request-id
  header, including error responses produced before a request handler runs. Background proposal
  compositions run by the workflows service now carry the id of the API request that scheduled them.
  The schema, policy, commerce and usage services log a warning when a request arrives without an
  x-request-id header: their callers (other Hive services, and the Hive client libraries reporting
  usage) always send one, so a missing id means a caller is not propagating it.

• #8709
  28023c3
  Thanks @jdolle! - When a persisted-document upload of an app
  deployment is aborted by the client, times out on the server, or one of its storage writes fails,
  every remaining object storage and ClickHouse request of that batch is cancelled, including
  pending retries. Previously the worker kept uploading the whole batch after the request was
  already over.

• #8730
  b3cb83a
  Thanks @jdolle! - Trace persisted document ingestion inside the
  GraphQL API's worker threads. The worker's spans (schema build, document validation, the
  ClickHouse insert, every S3 object write with its status and retry count, and how long a batch
  waited for a worker) now join the request's trace, every log line a batch produces in the worker
  carries the request id, and spans still buffered when a worker exits or the API shuts down are
  flushed first.

• #8714
  0a100e4
  Thanks @jonathanawesome! - Paged SchemaChangeConnection
  results (a schema check's breakingSchemaChanges, safeSchemaChanges, schemaChanges and the
  changes of a safe-change group, on the check and on its contract checks) now report their
  position: edges.cursor encodes the change's offset, hasNextPage and hasPreviousPage say
  whether more follow or precede the page, startCursor is empty on the first page and endCursor
  empty on the last, and total and totalCount count the whole list rather than the page.

• #8679
  bd8f0f2
  Thanks @jonathanawesome! - Listing Target.schemaChecks
  with only hasSchemaChanges or hasUnapprovedBreakingChanges selected no longer loads each
  check's change lists; both flags are computed in the database.

• #8679
  bd8f0f2
  Thanks @jonathanawesome! - Fix Target.schemaChecks
  returning no changes when they are selected through the SchemaCheck interface or through
  schemaChanges, which made the public API list every check without its changes.

• #8696
  27b04ab
  Thanks @Abhist17! - Fix links in emails sent with
  EMAIL_PROVIDER=sendmail losing a dot (for example …prod.proddsl.example… arriving as
  …prod.proddslexample…). The body is now base64-encoded, so no line can start with a "." that the
  receiving server would strip.

  Closes #7702

• #8712
  6c7ab24
  Thanks @jonathanawesome! - Fix the schema check and history
  pages freezing on the Public Schema and Supergraph diffs of large schemas. Both schemas are now
  sorted before they are compared, so definitions that only moved (for example when types move
  between subgraphs) no longer show as changes, and a diff too large to compute shows a notice
  instead of locking the page. The unchanged Public Schema view of a check now shows the public
  schema instead of the service SDL.

• #8721
  52eb8f1
  Thanks @jonathanawesome! - Stop calling breaking changes
  "safe based on usage data" when Hive doesn't record usage for what they change, such as a removed
  type, a removed union member or a new required argument. They now say that usage isn't tracked for
  them, and the version history summary counts them as not checked against usage instead of safe.
  Changes with nothing more to show no longer open to "No details available".