v0.3.6: Active CDN replication, device roles, budgets, tombstones, engagement fix, DOS hardening
Active CDN replication:
- All devices proactively replicate recent posts (<72h, <2 replicas) to peers
- Target priority: desktops (300) > anchors (200) > phones (100) + cache_pressure
- ReplicationRequest/Response (0xE1/0xE2) wire messages
- 10-min cycle, 2-min initial delay, cap 20 posts per request
- Graceful with small networks (1 peer = 1 replica, 0 peers = silent skip)
Device roles & budgets:
- Intermittent (phone), Available (desktop), Persistent (anchor)
- Advertised in InitialExchange, stored per-peer
- Replication budget: phones 100MB/hr, desktops/anchors 200MB/hr
- Delivery budget: phones 1GB/hr, desktops 2GB/hr, anchors 1GB/hr
- Hourly auto-reset, enforcement on blob serving
Cache management:
- 1GB default cache limit, configurable in settings UI
- Eviction cycle activated (was implemented but never started)
- Share-link priority boost (+100 for 3+ downstream)
- Cache pressure score (0-255) for replication targeting
Engagement distribution fix:
- BlobHeader JSON rebuilt after BlobHeaderDiff ops
- Previously reactions/comments stored in tables but header stayed stale
Tombstone system:
- deleted_at column on reactions and comments
- Tombstones propagate through pull sync (additive merge respects timestamps)
- UI queries filter WHERE deleted_at IS NULL
Persistent notifications:
- seen_engagement and seen_messages tables replace in-memory Sets
- Only notify on genuinely unseen content, survives restarts
DOS hardening:
- BlobHeaderDiff fan-out: single batched task, max 10 concurrent via JoinSet
- Blob prefetch: cap 20 per cycle, newest first
- PostDownstreamRegister: cap 50 per sync
- Delivery budget enforcement on BlobRequest handler
- Pull preference: non-anchors first to preserve anchor delivery budget
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
a7e632de88