Difficulty: Expert
Type: Feature
Recommended labels (if available in this repo): performance, backend, consistency, expert
Background
Applications using GuildPass SDK may trigger identical read requests concurrently, for example when multiple components request the same membership or access data during the same render cycle.
Launching duplicate network requests wastes resources and can produce inconsistent timing behaviour.
This issue introduces a generic in-flight request deduplication primitive without coupling it to any GuildPass endpoint.
Problem
There is currently no SDK mechanism for sharing one in-flight Promise between callers requesting the same logical operation.
A naive cache is not sufficient because completed response caching and in-flight deduplication have different semantics.
Expected Outcome
Implement a reusable request deduplicator that ensures concurrent callers using the same deduplication key share one underlying asynchronous execution.
Suggested Implementation
The module should:
- accept a deterministic string key;
- accept an asynchronous producer function;
- execute the producer only once while the key is in flight;
- return the same logical result to all concurrent callers;
- remove entries after success;
- remove entries after failure;
- avoid permanently caching completed results;
- ensure a failed execution can be retried later;
- optionally support configurable short-lived result retention only if clearly separated from in-flight semantics;
- expose current in-flight count for diagnostics if useful;
- bound the number of in-flight entries;
- define behaviour when capacity is exceeded;
- avoid leaking unresolved entries when producers reject;
- remain independent of HTTP and GuildPass domain types.
Acceptance Criteria
- Two concurrent calls with the same key execute the producer exactly once.
- All callers receive the same successful result.
- All callers observe the same failure when the shared producer rejects.
- Failed executions are removed so a later attempt can retry.
- Successful executions are removed after completion unless explicit retention is configured.
- Different keys execute independently.
- Capacity limits are enforced deterministically.
- No unresolved entry remains after producer settlement.
- Unit tests cover success, failure, retry, parallel different keys, and capacity handling.
- Tests use controlled promises to prove exactly-once execution.
pnpm typecheck passes.pnpm build passes.pnpm test passes.- No network request or response caching implementation is included.
Likely Affected Files/Directories
src/transport/ or src/utils/
tests/
Independence Requirement
This issue must remain a generic concurrency primitive and must not depend on the HTTP transport, cache, or endpoint implementations.
Difficulty: Expert
Type: Feature
Recommended labels (if available in this repo): performance, backend, consistency, expert
Background
Applications using GuildPass SDK may trigger identical read requests concurrently, for example when multiple components request the same membership or access data during the same render cycle.
Launching duplicate network requests wastes resources and can produce inconsistent timing behaviour.
This issue introduces a generic in-flight request deduplication primitive without coupling it to any GuildPass endpoint.
Problem
There is currently no SDK mechanism for sharing one in-flight Promise between callers requesting the same logical operation.
A naive cache is not sufficient because completed response caching and in-flight deduplication have different semantics.
Expected Outcome
Implement a reusable request deduplicator that ensures concurrent callers using the same deduplication key share one underlying asynchronous execution.
Suggested Implementation
The module should:
Acceptance Criteria
pnpm typecheckpasses.pnpm buildpasses.pnpm testpasses.Likely Affected Files/Directories
Independence Requirement
This issue must remain a generic concurrency primitive and must not depend on the HTTP transport, cache, or endpoint implementations.