Re-tier HTTP request timeouts around per-host memory

[ajpallares]

Checklist

• Unit tests

Motivation

Now that a single request can fail over across several hosts, the old timeout handling doesn't hold up. We only remember the most recent timeout in one global slot, so one slow host shortens timeouts for unrelated hosts (and a success on any host resets them all). The timeouts are also the same regardless of the request, so we wait too long on a host that's down before failing over to the next one.

Description

New per-attempt timeout tiers:

Request kind	Base	Reduced (source timed out in last 10 min)
Main-source, endpoint has NO fallback-URL support	15s	5s
Main-source, endpoint HAS fallback-URL support	5s	2s
Fallback-host request	30s	n/a (flat)
Proxied request	30s	n/a (flat)

• Replace the single AtomicLong last-timeout slot in HTTPTimeoutManager with a per-host ConcurrentHashMap (host -> lastTimeoutDate, 10-minute per-entry expiry, pruned on access).
• Re-tier the request timeouts as above; fallback-host and proxied requests never consult the memory.
• Record a timeout for any main-source request that timed out (keyed by attempt host), dropping the previous fallback-support-only condition; a successful main-source response clears only that host's entry.
• Rename the TIMEOUT_ON_MAIN_BACKEND_FOR_FALLBACK_SUPPORTED_ENDPOINT result to MAIN_SOURCE_TIMED_OUT; add AppConfig.hasProxyURL.

iOS counterpart: RevenueCat/purchases-ios#7179

---

Note

Medium Risk
Changes core HTTP timing and failover behavior for all backend calls; behavior is well covered by unit tests but mis-tuned timeouts could affect reliability under poor network conditions.

Overview
Reworks how the Android SDK picks connect timeouts when requests can fail over across multiple hosts. Per-host memory replaces a single global “last timeout” slot so one slow host no longer shortens timeouts for others, and a success on one host only clears that host’s entry (10-minute expiry per host).

New timeout tiers depend on attempt type: main-source without fallback-URL support uses 15s / 5s reduced; main-source with fallback support uses 5s / 2s; fallback-host and proxied requests use a flat 30s and ignore the memory.

HTTPClient records MAIN_SOURCE_TIMED_OUT for any main-source SocketTimeoutException (not only fallback-capable endpoints), passes resolved host into HTTPTimeoutManager, and uses AppConfig.hasProxyURL when computing timeouts. Test mocks and unit/integration tests are updated for the new API and enum name.

^{Reviewed by Cursor Bugbot for commit 7263274. Bugbot is set up for automated code reviews on this repo. Configure here.}

[codecov]

Codecov Report

❌ Patch coverage is 96.15385% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 80.61%. Comparing base (65f3fb5) to head (7263274).

Files with missing lines	Patch %	Lines
...tlin/com/revenuecat/purchases/common/HTTPClient.kt	85.71%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3746      +/-   ##
==========================================
+ Coverage   80.60%   80.61%   +0.01%     
==========================================
  Files         423      423              
  Lines       17401    17406       +5     
  Branches     2682     2687       +5     
==========================================
+ Hits        14026    14032       +6     
  Misses       2362     2362              
+ Partials     1013     1012       -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[AlvaroBrey]

Looks good and makes sense I think! Just one question