protocol: speed up sync by minimizing duplicate work
In particular, the prepare_handle_incoming_blocks call is pretty lengthy, and entirely pointless in the common case where several different connections will prepare the exact same blocks.
This commit is contained in:
parent
61dfa310d7
commit
8bdc86beb4
|
@ -135,6 +135,7 @@ namespace cryptonote
|
|||
std::atomic<bool> m_synchronized;
|
||||
bool m_one_request = true;
|
||||
std::atomic<bool> m_stopping;
|
||||
epee::critical_section m_sync_lock;
|
||||
|
||||
boost::mutex m_buffer_mutex;
|
||||
double get_avg_block_size();
|
||||
|
|
|
@ -790,6 +790,8 @@ namespace cryptonote
|
|||
context.m_remote_blockchain_height = arg.current_blockchain_height;
|
||||
|
||||
size_t count = 0;
|
||||
std::vector<crypto::hash> block_hashes;
|
||||
block_hashes.reserve(arg.blocks.size());
|
||||
for(const block_complete_entry& block_entry: arg.blocks)
|
||||
{
|
||||
if (m_stopping)
|
||||
|
@ -807,9 +809,10 @@ namespace cryptonote
|
|||
return 1;
|
||||
}
|
||||
//to avoid concurrency in core between connections, suspend connections which delivered block later then first one
|
||||
const crypto::hash block_hash = get_block_hash(b);
|
||||
if(count == 2)
|
||||
{
|
||||
if(m_core.have_block(get_block_hash(b)))
|
||||
if(m_core.have_block(block_hash))
|
||||
{
|
||||
context.m_state = cryptonote_connection_context::state_idle;
|
||||
context.m_needed_objects.clear();
|
||||
|
@ -819,7 +822,7 @@ namespace cryptonote
|
|||
}
|
||||
}
|
||||
|
||||
auto req_it = context.m_requested_objects.find(get_block_hash(b));
|
||||
auto req_it = context.m_requested_objects.find(block_hash);
|
||||
if(req_it == context.m_requested_objects.end())
|
||||
{
|
||||
LOG_ERROR_CCONTEXT("sent wrong NOTIFY_RESPONSE_GET_OBJECTS: block with id=" << epee::string_tools::pod_to_hex(get_blob_hash(block_entry.block))
|
||||
|
@ -836,6 +839,7 @@ namespace cryptonote
|
|||
}
|
||||
|
||||
context.m_requested_objects.erase(req_it);
|
||||
block_hashes.push_back(block_hash);
|
||||
}
|
||||
|
||||
if(context.m_requested_objects.size())
|
||||
|
@ -858,7 +862,31 @@ namespace cryptonote
|
|||
|
||||
uint64_t previous_height = m_core.get_current_blockchain_height();
|
||||
|
||||
// we lock all the rest to avoid having multiple connections redo a lot
|
||||
// of the same work, and one of them doing it for nothing: subsequent
|
||||
// connections will wait until the current one's added its blocks, then
|
||||
// will add any extra it has, if any
|
||||
CRITICAL_REGION_LOCAL(m_sync_lock);
|
||||
|
||||
// dismiss what another connection might already have done (likely everything)
|
||||
uint64_t top_height;
|
||||
crypto::hash top_hash;
|
||||
if (m_core.get_blockchain_top(top_height, top_hash)) {
|
||||
uint64_t dismiss = 1;
|
||||
for (const auto &h: block_hashes) {
|
||||
if (top_hash == h) {
|
||||
LOG_DEBUG_CC(context, "Found current top block in synced blocks, dismissing "
|
||||
<< dismiss << "/" << arg.blocks.size() << " blocks");
|
||||
while (dismiss--)
|
||||
arg.blocks.pop_front();
|
||||
break;
|
||||
}
|
||||
++dismiss;
|
||||
}
|
||||
}
|
||||
|
||||
m_core.prepare_handle_incoming_blocks(arg.blocks);
|
||||
|
||||
for(const block_complete_entry& block_entry: arg.blocks)
|
||||
{
|
||||
if (m_stopping)
|
||||
|
|
Loading…
Reference in New Issue