Three interrelated fixes for sync reliability on large wallets:
1. Prevent -101 "excessive resource usage":
ScanChainAsync uses GetHistoryAsync for discovery instead of
SubscribeScripthashAsync. Subscriptions are limited to the gap window
only (≤ 2×gapLimit addresses), so the session subscription count stays
constant regardless of wallet history size — eliminating -101 on
wallets with 1000+ historical addresses.
2. Persist and resume tx cache across sessions and disconnects:
WalletSynchronizer gains ExportCaches() and PreloadCaches(). On every
successful sync (and on partial failure) the downloaded raw tx hex and
verified Merkle heights are written to SyncCache.RawTxHex/VerifiedAt
in the wallet file. On reconnect or app restart the new synchronizer
is pre-populated from disk so only new transactions are downloaded.
_synchronizer is always nulled on reconnect (new client = new instance)
while caches survive via the wallet file.
3. Auto-connect with multi-server fallback and retry:
ConnectAndSync iterates BuildServerCandidates() (current server first,
then all KnownServers in order) and stops at the first that responds.
_autoReconnect and _syncFailed are set in OpenLoaded() so the keepalive
timer (20 s) retries automatically on any failure — connection error,
-101 mid-download, or timeout. PersistPartialTxCache() saves whatever
was downloaded so each retry resumes from its checkpoint.
WalletSynchronizer now pipelines JSON-RPC requests (address subscribes,
histories, tx downloads and Merkle proofs run in parallel batches) and
keeps per-connection caches of downloaded transactions and verified
proofs, so notification-triggered resyncs only cost the delta.
The app connects automatically when a wallet opens, keeps one
synchronizer per connection, and queues notifications that arrive
during a sync instead of dropping them: incoming transactions now show
up within seconds.
davide