db_sync: add author_sig VERIFY FAIL diagnostics + SYNC_DONE retry limit

- db_verify_author_sig: show ed_pubkey and sig first bytes on failure
- SI_PEER: add sync_retry_count field
- db_handle_sync_done: limit re-initiate to 3 retries before giving up
- db_sync_initiate_sync / db_sync_on_conn_up: reset retry_count

src/db_sync.c

@@ -38,6 +38,7 @@ struct SI_PEER {
     uint64_t node_id;
     uint32_t synced_pos;
     uint8_t  sync_state;
+    uint8_t  sync_retry_count;
     uint64_t sync_start_tb;
 };
 
@@ -400,7 +401,8 @@ static int db_verify_author_sig(struct DB_SYNC_INSTANCE* si,
     uint8_t pubkey[32];
     if (db_get_ed25519_pubkey(si->db_sync, author_node_id, pubkey) != 0) {
         DEBUG_WARN(DEBUG_CATEGORY_DB_SYNC,
-                   "cannot get Ed25519 pubkey for author=%016llx", (unsigned long long)author_node_id);
+                   "cannot get Ed25519 pubkey for author=%016llx — discarding",
+                   (unsigned long long)author_node_id);
         return -1;
     }
     uint8_t msg[8192]; size_t off = 0;
@@ -408,8 +410,8 @@ static int db_verify_author_sig(struct DB_SYNC_INSTANCE* si,
     if (jlen > 0) { memcpy(msg + off, json, jlen); off += jlen; }
     if (sc_ed25519_verify(pubkey, msg, off, sig) != SC_OK) {
         DEBUG_WARN(DEBUG_CATEGORY_DB_SYNC,
-                   "author_sig VERIFY FAIL node=%016llx — discarding as forgery",
+                   "author_sig VERIFY FAIL author=%016llx ed_pubkey=%016llx... msg_len=%zu sig=%016llx... — discarding as forgery",
-                   (unsigned long long)author_node_id);
+                   (unsigned long long)author_node_id, *(uint64_t*)pubkey, off, *(uint64_t*)sig);
         return -1;
     }
     return 0;
@@ -981,14 +983,24 @@ static void db_handle_sync_done(struct DB_SYNC_INSTANCE* si, uint64_t src, const
                (mc == pc && mch8 == pch8) ? "matched" : "MISMATCH");
 
     if (mc != pc || mch8 != pch8) {
+        struct SI_PEER* sp = si_peer_find(si, src);
+        if (sp) sp->sync_retry_count++;
+        if (sp && sp->sync_retry_count > 3) {
+            DEBUG_WARN(DEBUG_CATEGORY_DB_SYNC,
+                       "SYNC_DONE mismatch retry limit exceeded (%u/3) with %016llx — giving up",
+                       sp->sync_retry_count, (unsigned long long)src);
+            sp->sync_state = 2; sp->synced_pos = (mc > 0) ? mc - 1 : 0;
+            return;
+        }
         DEBUG_INFO(DEBUG_CATEGORY_DB_SYNC,
-                   "SYNC_DONE mismatch with %016llx my_count=%u peer_count=%u my_ch8=%016llx peer_ch8=%016llx — re-initiating",
+                   "SYNC_DONE mismatch with %016llx my_count=%u peer_count=%u my_ch8=%016llx peer_ch8=%016llx — re-initiating (retry %u/3)",
-                   (unsigned long long)src, mc, pc, (unsigned long long)mch8, (unsigned long long)pch8);
+                   (unsigned long long)src, mc, pc, (unsigned long long)mch8, (unsigned long long)pch8,
+                   sp ? sp->sync_retry_count : 0);
         db_sync_initiate_sync(si, src);
         return;
     }
     struct SI_PEER* sp = si_peer_find(si, src);
-    if (sp) { sp->synced_pos = (mc > 0) ? mc - 1 : 0; sp->sync_state = 2; }
+    if (sp) { sp->synced_pos = (mc > 0) ? mc - 1 : 0; sp->sync_state = 2; sp->sync_retry_count = 0; }
     DEBUG_INFO(DEBUG_CATEGORY_DB_SYNC, "sync confirmed with %016llx count=%u ch8=%016llx",
                (unsigned long long)src, mc, (unsigned long long)mch8);
 }
@@ -1160,7 +1172,7 @@ static void db_sync_on_conn_up(struct ETCP_CONN* conn, void* arg)
         if (!p) continue;
         if (p->sync_state != 0) { skipped_state++; continue; }
         if (!conn->initialized || !conn->links_up) { skipped_not_ready++; continue; }
-        p->sync_state = 1;
+        p->sync_state = 1; p->sync_retry_count = 0;
         db_sync_initiate_sync(si, pid);
         synced++;
     }
@@ -1206,7 +1218,7 @@ static void db_sync_initiate_sync(struct DB_SYNC_INSTANCE* si, uint64_t pid)
     uint32_t mc = db_count(si);
     DEBUG_TRACE(DEBUG_CATEGORY_DB_SYNC, "pid=%016llx my=%u tbl=%s", (unsigned long long)pid, mc, SI_TBL(si));
     struct SI_PEER* p = si_peer_find(si, pid);
-    if (p) p->sync_start_tb = get_time_tb();
+    if (p) { p->sync_start_tb = get_time_tb(); p->sync_retry_count = 0; }
 
     uint8_t msg[5];
     msg[0] = DB_MSG_INIT_SYNC;