mirror of
https://github.com/RevenueCat/purchases-ios
synced 2026-11-11 22:55:55 +05:45
Toni RicoandClaude Fable 5
8925524293
* feat(remote-config): health-check-gated API source failover in HTTPClient Wires the API source failover into the request pipeline, still behind the usesRemoteConfigAPISources dangerous setting (internal and defaulting to false, so released behavior is unchanged): - Requests resolve their API source once (in start) and carry the handle, so a failure can report the exact source unhealthy and ETag/backoff retries keep targeting the same host. - On a failure that points at the host, the source is health-checked before failing over: a healthy source surfaces the original error, an unhealthy one is reported and the request transparently retries on the next source (capped at 5 source attempts per request). The static per-endpoint fallback hosts stay the last resort. - Unlike Android, device-connectivity errors (reusing NetworkError.isAllowedToRetryWithFallbackHost) skip the health check entirely and surface the original error: switching hosts can't fix a device that is offline. - Both Backend HTTPClients share one APISourceFailover, so they walk one source list and share one health-check cache. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * test(remote-config): end-to-end API source failover integration tests Port of purchases-android's BackendAPISourceFailoverIntegrationTest: real Backend.getCustomerInfo through the production wiring (real HTTPClients, shared APISourceFailover, real SourceHealthChecker probing over the URL loading system) and a real RemoteConfigSourceProvider fed by an in-memory sources topic, with OHHTTPStubs standing in for the source hosts and recording exact per-host request order. Adds the iOS-only device-connectivity scenario and a dangerous-setting-off wiring check on top of Android's seven scenarios. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * refactor: model the request's API source as a single state enum Replaces the three loosely coupled Request fields (apiSource, apiSourceResolved, apiSourceAttemptCount) with one APISourceState enum (unresolved / noSource / source(_, attemptCount:)), making inconsistent combinations unrepresentable. No behavior change. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * refactor: adopt the narrowed APISourceFailover initializer Passes just the usesRemoteConfigAPISources flag at the two construction sites, following the signature change in the base branch. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * docs: clarify why static fallback still runs when the source is healthy Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * feat(remote-config): raise the API failover restart interval to 2 hours The api source list now returns to the primary source 2 hours after it first failed over instead of 10 minutes. Timer tests derive their time values from the constant so future changes are single-line. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * test(remote-config): cover shared failover and probe-in-flight ordering Addresses review feedback: - Both HTTPClient lanes share one APISourceFailover: a failover on the main lane advances the source list the remote-config lane resolves. - A request enqueued while a health check is in flight stays queued until the probe resolves and runs after the front-queued failover retry. - Bind the source and attempt count in one pattern match at the failover guard. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * refactor: drop the APISourceState.attemptCount accessor Its only caller pattern-matches instead, surfacing the "must already target a source" assumption at the one place it matters rather than papering over impossible states with a 0 default. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>