fix(elasticache): serverless endpoint port + replication-group members + snapshot export (bug-hunt)

[vieiralucas]

Summary

Bug-hunt batch fixing four ElastiCache compatibility issues in fakecloud-elasticache. Behavior-only; no SDK/doc/count changes except a minimal cross-service S3 write surface for snapshot export.

• 1.7 (MED) — serverless cache Port:0 when no Docker runtime. In non-Docker setups (the default locally/CI) CreateServerlessCache creates the cache directly as available but left endpoint.port = 0, because the port backfill only runs on the Docker-backed tick. Every connection string then ended in :0 (unconnectable). Now, when there is no runtime, the endpoint (and reader endpoint) is assigned the conventional engine port at create time (6379 redis/valkey, 11211 memcached) via the existing engine_default_port helper. The Docker path still backfills its real mapped host port.
• 1.8 (MED) — replication group emitted one NodeGroupMember regardless of replicas. Each <NodeGroup> emitted exactly one primary member, so the Terraform provider derived replicas_per_node_group = len(NodeGroupMembers) - 1 = 0 and produced a perpetual plan diff. Now each shard emits 1 primary + N replica members (CurrentRole primary/replica) matching the group's configured replica count (current_replicas_per_shard), with synthesized <rg>-NNN cluster ids, clamped at AWS's 5-replica ceiling.
• 1.25 (LOW) — ExportServerlessCacheSnapshot discarded S3BucketName. The required bucket was parsed then dropped (let _ = bucket;). It now performs a real cross-service S3 write: the snapshot export artifact is written into the target bucket via the shared S3 state (same pattern firehose/athena/config use), with a real MD5 ETag. A missing bucket is rejected with InvalidParameterValue, matching AWS. Wired with_s3(...) on the service in main.rs.
• P.2 (LOW) — snapshot /tmp hardcode. Already resolved in the current tree: snapshot_rdb_temp_path roots under std::env::temp_dir(). Verified no other hardcoded /tmp remains in the crate source.

Test plan

• Added unit tests (fakecloud-elasticache lib, 240 pass):
  • create_serverless_cache_without_runtime_assigns_default_port — no runtime → create + describe report <Port>6379</Port>, never <Port>0</Port>.
  • replication_group_emits_primary_and_replica_members — NumCacheClusters=3 → describe shows 1 primary + 2 replica members with the expected rg-repl-00N ids.
  • export_serverless_cache_snapshot_writes_object_to_s3 — export writes snap-x.rdb into the target bucket (non-zero size, quoted ETag); missing bucket → InvalidParameterValue.
• Extended the Docker-gated e2e elasticache_create_replication_group_round_trips_extended_fields to assert each shard exposes 1 primary + 1 replica NodeGroupMember over the real SDK.
• cargo clippy -p fakecloud-elasticache --all-targets -- -D warnings clean; cargo fmt; cargo build -p fakecloud (server) green. Docker-gated e2e tests are skipped in this environment (Docker unreachable); the e2e change is purely additive inside the existing require_docker_or_skip guard, so it is identical to main when Docker is absent, and the e2e crate compiles cleanly.

---

Summary by cubic

Fixes three ElastiCache compatibility issues: serverless caches now get a real port when no runtime is present, replication groups report correct primary/replica members, and serverless snapshot export actually writes to S3. Adds minimal S3 wiring to support the export.

• Bug Fixes

  • Serverless caches: set endpoint and reader port to the engine default (6379 for redis/valkey, 11211 for memcached) when no runtime is available; Docker path still backfills the mapped port.
  • Replication groups: emit 1 primary + N replica NodeGroupMembers per shard (clamped to 5) with stable IDs like <rg>-NNN, fixing Terraform’s perpetual diff.
  • Snapshot export: ExportServerlessCacheSnapshot writes a *.rdb object into the given S3 bucket and returns InvalidParameterValue when the bucket is missing; snapshot temp files use the system temp dir.

• Dependencies

  • Added fakecloud-s3, bytes, and md-5. Wired S3 into the service via with_s3(...) so exports are visible through shared S3 state.

^{Written for commit 3c6c948. Summary will update on new commits.}

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.