v0.11.0-rc4

🚀 What's New

• eth_getLogs completeness verification (logsBloom check) + l1infotreesync self-healing escalated reorg (#1773): root-causes the corruption class behind the bokuto production incident — a mixed/flaky RPC view during an L1 reorg can silently omit a log from eth_getLogs while every observed header hash stays canonical, invisible to header-hash reorg detection by construction. Header logs blooms have no false negatives, so a block whose bloom is positive for a queried address but whose logs response has no matching entry is now flagged as a suspected omission and arbitrated with a single re-query by block hash; a confirmed omission retries loudly instead of silently committing corruption. Wired into multidownloader (l1infotreesync) and the legacy sync downloader (bridgesync, claimsync, l2gersync) for the unfinalized zone. Additionally, l1infotreesync now persists a last verified checkpoint on every passing sanity check, and escalates a second consecutive reorg-at-the-same-block to that checkpoint (or to InitialBlock for a pre-existing corrupted DB) — turning a permanent stuck-halt loop into an automatic full self-heal. Also adds bounded error-log throttling (common.ShouldLogRetryAtError) to indefinite retry loops across syncers.
• bridgetracker idle-timeout eviction (#1770): new [Tracker].IdleTimeout (default 30m) forgets a supervised bridge — terminal or still active — once nobody has polled or subscribed to it for that long, bounding registry memory for bridges nobody is watching. Complements the existing RetentionPeriod (terminal-only grace period).
• Configurable per-method agglayer client cache policy (#1771): replaces the hardcoded "only GetCertificateHeader is cached" behavior with a per-method cached/passthrough/forbidden policy on CacheConfig, covering all six AgglayerClientInterface methods. Lets the bridge tracker collapse repeated per-bridge agglayer calls into one cached call per network.
• claimsync GetClaimsPaged timeout fix (#1774): fixes context deadline exceeded on the bridge service claims endpoint for deployments with a large claim table (observed on v0.11.0-rc3), by restricting the ranking CTE to the requested page's global indexes and adding two new indexes (claimsync0002 migration).
• force_ger_update redundant-send race fix (#1778): closes a race where the in-flight guard could clear before the timer reset was applied, occasionally causing two bridgeMessage transactions to be sent back to back in production despite #1769's earlier dedup fix. The guard is now cleared exclusively by the single-threaded confirmation step that also advances the timer.
• Dependency bump: github.com/pion/dtls/v3 3.1.2 → 3.1.4 (#1754).

⚠️ Breaking Changes

• 🛠️ Config: [Tracker] (proxy config) gains IdleTimeout (default 30m) (#1770).
• 🛠️ Config: CacheConfig gains 6 new per-method fields (SendCertificate, GetCertificateHeader, GetEpochConfiguration, GetLatestSettledCertificateHeader, GetLatestPendingCertificateHeader, GetNetworkInfo), each cached/passthrough/forbidden. Cached=true alone no longer implies GetCertificateHeader caching — each client config must now explicitly opt individual methods into cached (#1771). See the PR for the exact per-deployment config each existing client (Tracker/AggSender/Validator) needs to preserve prior behavior.
• ⚙️ Operational: on a DB with pre-existing committed-state divergence (no checkpoint recorded yet), l1infotreesync's first reorg escalation purges back to InitialBlock — i.e. a full automatic resync — replacing today's "halted forever until manual DB wipe." multidownloader StepSafe now fetches headers for every block in a queried range (not just blocks with logs); persisted data volume is unchanged, RPC load increases during initial sync/catch-up only (#1773).
• 🗄️ Storage: new migrations l1infotreesync0005 (single-row l1info_checkpoint table, #1773) and claimsync0002 (indexes on claim/unset_claim global_index, #1774) — both apply automatically on startup.

📋 Config Updates

• 🧾 Tracker (proxy) — new IdleTimeout, shown with the rest of [Tracker] for context:

  [Tracker]
  RetentionPeriod = "10m"
  IdleTimeout = "30m"                # <- new field, default 30m
  RegisterResolveTimeout = "3s"
  L1BlockFinality = "LatestBlock"
  L2BlockFinality = "LatestBlock"
  MaxTrackedBridges = 100000

• 🧾 Tracker (proxy) agglayer client cache — explicit per-method policy:

  [Tracker.AgglayerClient.ConfigurationCache]
  TTL = "1s"
  Capacity = 100
  SendCertificate = "forbidden"
  GetCertificateHeader = "cached"
  GetEpochConfiguration = "cached"
  GetLatestPendingCertificateHeader = "cached"
  GetNetworkInfo = "cached"

• 🧾 AggSender agglayer client cache — unchanged behavior, now explicit:

  [AggSender.AgglayerClient.ConfigurationCache]
  TTL = "5m"
  Capacity = 100
  SendCertificate = "passthrough"
  GetCertificateHeader = "cached"
  GetEpochConfiguration = "passthrough"
  GetLatestSettledCertificateHeader = "passthrough"
  GetLatestPendingCertificateHeader = "passthrough"
  GetNetworkInfo = "passthrough"

🐛 Bug Fixes

• l1infotreesync self-heals a permanent stuck-halt loop caused by RPC log omission during an L1 reorg, instead of requiring a manual DB wipe (#1773).
• claimsync's GetClaimsPaged no longer times out against large claim tables (#1774).
• force_ger_update no longer occasionally double-sends a bridgeMessage transaction (#1778).

🔧 Maintenance

• Bump github.com/pion/dtls/v3 3.1.2 → 3.1.4 (#1754).

Full changelog: v0.11.0-rc3...v0.11.0-rc4