EOS原始碼備忘-Push Transaction機制
這裡我們討論EOS Push Transaction 的邏輯，這塊EOS與Eosforce實現有一些區別，我們會著重點出。 關於wasm相關的內容我們會有一片專門的文件分析。
我們這裡通常將Transaction譯做交易，其實這裡應該是事務的意思。
1. Transaction與Action
在EOS中Transaction與Action是最重要的幾個型別， 在EOS中，所有的鏈上行為都是Action，Transaction是一系列Action組成的事務。
EOS中使用繼承體系劃分trx與action結構，關係圖如下：
transaction_header <- transaction <- signed_transaction <- deferred_transaction
|
packed_transaction
1.1 Action
我們這裡先看一下Action的宣告：
// 許可權結構structpermission_level{account_name actor; permission_name permission; }; ...structaction{account_name account; action_name name;// 執行所需的許可權vector authorization; bytes data; ...// 打包成二進位制templateTdata_as()const{ ... } };
Action沒有什麼特別的內容，但要注意：
!> 在EOS中一個transaction中包含很多個action，而在Eosforce中一個trx只能包括一個action。
1.2 Transaction
下面我們分析一下transaction，這裡簡寫為trx。
首先看下
/**
* The transaction header contains the fixed-sized data
* associated with each transaction. It is separated from
* the transaction body to facilitate partial parsing of
* transactions without requiring dynamic memory allocation.
*
* All transactions have an expiration time after which they
* may no longer be included in the blockchain. Once a block
* with a block_header::timestamp greater than expiration is
* deemed irreversible, then a user can safely trust the transaction
* will never be included.
*
* Each region is an independent blockchain, it is included as routing
* information for inter-blockchain communication. A contract in this
* region might generate or authorize a transaction intended for a foreign
* region.
*/structtransaction_header{time_point_sec expiration;///< trx超時時間uint16_tref_block_num =0U;// 包含trx的block num 注意這個值是後2^16個塊中uint32_tref_block_prefix =0UL;// blockid的低32位fc::unsigned_int max_net_usage_words =0UL;// 網路資源上限uint8_tmax_cpu_usage_ms =0;// cpu資源上限fc::unsigned_int delay_sec =0UL;/// 延遲交易的延遲時間/**
* @return the absolute block number given the relative ref_block_num
* 計算ref_block_num
*/block_num_typeget_ref_blocknum( block_num_type head_blocknum )const{return((head_blocknum/0xffff)*0xffff) head_blocknum%0xffff; }voidset_reference_block(constblock_id_type& reference_block );boolverify_reference_block(constblock_id_type& reference_block )const;voidvalidate()const; };
transaction_header包含一個trx中固定長度的資料，這裡之所以要單獨提出來主要是為了優化。
transaction視為交易體資料，這裡主要是儲存這個trx包含的action。
/**
* A transaction consits of a set of messages which must all be applied or
* all are rejected. These messages have access to data within the given
* read and write scopes.
*/
// 在EOS中一個交易中 action要麼全部執行，要麼都不執行
struct transaction : public transaction_header {
vector<action> context_free_actions;
vector<action> actions;
extensions_type transaction_extensions;
// 獲取trx id
transaction_id_type id()const;
digest_type sig_digest( const chain_id_type& chain_id, const vector<bytes>& cfd = vector<bytes>() )const;
...
};
注意這裡的context_free_actions，這裡指上下文無關的Action，具體資訊可以參見這裡： https://medium.com/@bytemaster/eosio-development-update-272198df22c1 和 https://github.com/EOSIO/eos/issues/1387。 如果一個Action執行時只依賴與transaction的資料，而不依賴與鏈上的狀態，這樣的action可以併發的執行。
另外一個值得注意的是trx id：
transaction_id_type transaction::id()const{ digest_type::encoder enc; fc::raw::pack( enc, *this);returnenc.result();}
!> Eosforce不同
在Eosforce中為了新增手續費資訊，trx與EOS結構不同，主要是增加了fee， 在transaction中：
struct transaction : public transaction_header {
vector<action> context_free_actions;
vector<action> actions;
extensions_type transaction_extensions;
asset fee; // EOSForce 增加的手續費，在客戶端push trx時需要寫入
transaction_id_type id()const;
digest_type sig_digest( const chain_id_type& chain_id, const vector<bytes>& cfd = vector<bytes>() )const;
flat_set<public_key_type> get_signature_keys( const vector<signature_type>& signatures,
const chain_id_type& chain_id,
const vector<bytes>& cfd = vector<bytes>(),
bool allow_duplicate_keys = false,
bool use_cache = true )const;
uint32_t total_actions()const { return context_free_actions.size() actions.size(); }
account_name first_authorizor()const {
for( const auto& a : actions ) {
for( const auto& u : a.authorization )
return u.actor;
}
return account_name();
}
};
在 https://eosforce.github.io/Documentation/#/zh-cn/eosforce_client_develop_guild 這篇文件裡也有說明。
這裡計算trx id時完全使用trx的資料，這意味著，如果是兩個trx資料完全一致，特別的他們在一個區塊中，那麼這兩個trx的id就會是一樣的。
1.3 signed_transaction
一個trx簽名之後會得到一個signed_transaction，
structsigned_transaction:publictransaction { ...vector signatures;// 簽名vector context_free_data;// 上下文無關的action所使用的資料// 簽名constsignature_type&sign(constprivate_key_type& key,constchain_id_type& chain_id);signature_typesign(constprivate_key_type& key,constchain_id_type& chain_id)const; flat_set get_signature_keys(constchain_id_type& chain_id,boolallow_duplicate_keys =false,booluse_cache =true)const; };
signed_transaction包含簽名資料和上下文無關的action所使用的資料，
這裡要談一下context_free_data，可以參見 https://github.com/EOSIO/eos/commit/a41b4d56b5cbfd0346de34b0e03819f72e834041 ，之前我們看過context_free_actions， 在上下文無關的action中可以去從context_free_data獲取資料，可以參見在api_tests.cpp中的測試用例：
... {// back to normal actionactionact1(pl, da); signed_transaction trx; trx.context_free_actions.push_back(act); trx.context_free_data.emplace_back(fc::raw::pack(100));// verify payload matches context free datatrx.context_free_data.emplace_back(fc::raw::pack(200)); trx.actions.push_back(act1);// attempt to access non context free apifor(uint32_ti =200; i <=211; i) { trx.context_free_actions.clear(); trx.context_free_data.clear(); cfa.payload = i; cfa.cfd_idx =1;actioncfa_act({}, cfa); trx.context_free_actions.emplace_back(cfa_act); trx.signatures.clear(); set_transaction_headers(trx); sigs = trx.sign(get_private_key(N(testapi),"active"), control->get_chain_id()); BOOST_CHECK_EXCEPTION(push_transaction(trx), unaccessible_api, [](constfc::exception& e) {returnexpect_assert_message(e,"only context free api's can be used in this context"); } ); }...
這裡可以作為context_free_action的一個例子，在test_api.cpp中的合約會呼叫void test_action::test_cf_action()函式：
// 這個是測試`context_free_action`的action
void test_action::test_cf_action() {
eosio::action act = eosio::get_action( 0, 0 );
cf_action cfa = act.data_as<cf_action>();
if ( cfa.payload == 100 ) {
// verify read of get_context_free_data, also verifies system api access
// 測試在合約中通過 get_context_free_data 獲取 context_free_data
int size = get_context_free_data( cfa.cfd_idx, nullptr, 0 );
eosio_assert( size > 0, "size determination failed" );
eosio::bytes cfd( static_cast<size_t>(size) );
size = get_context_free_data( cfa.cfd_idx, &cfd[0], static_cast<size_t>(size) );
eosio_assert(static_cast<size_t>(size) == cfd.size(), "get_context_free_data failed" );
uint32_t v = eosio::unpack<uint32_t>( &cfd[0], cfd.size() );
eosio_assert( v == cfa.payload, "invalid value" );
// 以下是測試一些功能
// verify crypto api access
checksum256 hash;
char test[] = "test";
...
// verify context_free_system_api
eosio_assert( true, "verify eosio_assert can be called" );
// 下面是測試一些在上下文無關action中不能使用的功能
} else if ( cfa.payload == 200 ) {
// attempt to access non context free api, privileged_api
is_privileged(act.name);
eosio_assert( false, "privileged_api should not be allowed" );
} else if ( cfa.payload == 201 ) {
// attempt to access non context free api, producer_api
get_active_producers( nullptr, 0 );
eosio_assert( false, "producer_api should not be allowed" );
...
} else if ( cfa.payload == 211 ) {
send_deferred( N(testapi), N(testapi), "hello", 6 );
eosio_assert( false, "transaction_api should not be allowed" );
}
}
接下來我們來看一看packed_transaction，通過這個類我們可以將trx打包，這樣可以最大的節省空間，關於它的功能，會在下面使用的提到。
2. Transaction的接收和轉發流程
瞭解Transaction類定義之後，我們先來看一下trx在EOS系統中的接收和轉發流程，確定發起trx的入口， 在EOS中，大部分trx都是由使用者所操縱的客戶端發向同步節點，再通過同步網路傳送給超級節點，超級節點會把trx打包進塊，這裡我們梳理一下這裡的邏輯，
首先，關於客戶端提交trx的流程，可以參見 https://eosforce.github.io/Documentation/#/zh-cn/eosforce_client_develop_guild ， 我們這裡從node的角度看是怎麼處理收到的trx的。
對於一個節點，trx可能是其他節點同步過來的，也可能是客戶端通過api請求的，我們先看看api：
EOS中通過http_plugin外掛響應http請求，這裡我們只看處理邏輯，在chain_api_plugin.cpp中註冊的這兩個：
voidchain_api_plugin::plugin_startup() { ilog("starting chain_api_plugin"); my.reset(newchain_api_plugin_impl(app().get_plugin().chain()));autoro_api = app().get_plugin().get_read_only_api();autorw_api = app().get_plugin().get_read_write_api(); app().get_plugin().add_api({ ... CHAIN_RW_CALL_ASYNC(push_transaction, chain_apis::read_write::push_transaction_results,202), CHAIN_RW_CALL_ASYNC(push_transactions, chain_apis::read_write::push_transactions_results,202) });}
最終實際呼叫的是這裡：
// 呼叫流程 push_transactions -> push_recurse -> push_transactionvoidread_write::push_transaction(constread_write::push_transaction_params& params, next_function next) {try{autopretty_input =std::make_shared();autoresolver = make_resolver(this, abi_serializer_max_time);try{// 這裡在使用 packed_transaction 解包abi_serializer::from_variant(params, *pretty_input, resolver, abi_serializer_max_time); } EOS_RETHROW_EXCEPTIONS(chain::packed_transaction_type_exception,"Invalid packed transaction")// 這裡呼叫 incoming::methods::transaction_async 函式app().get_method()(pretty_input,true, [this, next](constfc::static_variant& result) ->void{ ...// 返回返回值， 略去}); }catch( boost::interprocess::bad_alloc& ) { raise(SIGUSR1); } CATCH_AND_CALL(next);}
注意這裡的 persist_until_expired 引數，我們在 EOS原始碼備忘-Block Produce機制 這篇文件中分析過。 incoming::methods::transaction_async註冊的是on_incoming_transaction_async函式：
my->_incoming_transaction_async_provider = app().get_method().register_provider([this](constpacked_transaction_ptr& trx,boolpersist_until_expired, next_function next) ->void{returnmy->on_incoming_transaction_async(trx, persist_until_expired, next ); });
on_incoming_transaction_async如下：
voidon_incoming_transaction_async(constpacked_transaction_ptr& trx,boolpersist_until_expired, next_function next){ chain::controller& chain = app().get_plugin().chain();if(!chain.pending_block_state()) { _pending_incoming_transactions.emplace_back(trx, persist_until_expired, next);return; }autoblock_time = chain.pending_block_state()->header.timestamp.to_time_point();// 返回結果的回撥autosend_response = [this, &trx, &next](constfc::static_variant& response) { next(response);if(response.contains()) { _transaction_ack_channel.publish(std::pair(response.get(), trx)); }else{ _transaction_ack_channel.publish(std::pair(nullptr, trx)); } };autoid = trx->id();// 超時時間檢查if( fc::time_point(trx->expiration()) < block_time ) { send_response(std::static_pointer_cast(std::make_shared(FC_LOG_MESSAGE(error,"expired transaction ${id}", ("id", id)) )));return; }// 檢查是否是已處理過的trxif( chain.is_known_unexpired_transaction(id) ) { send_response(std::static_pointer_cast(std::make_shared(FC_LOG_MESSAGE(error,"duplicate transaction ${id}", ("id", id)) )));return; }// 看看是否超過最大的執行時間了autodeadline = fc::time_point::now() fc::milliseconds(_max_transaction_time_ms);booldeadline_is_subjective =false;if(_max_transaction_time_ms <0|| (_pending_block_mode == pending_block_mode::producing && block_time < deadline) ) { deadline_is_subjective =true; deadline = block_time; }try{// 這裡直接呼叫`push_transaction`來執行trxautotrace = chain.push_transaction(std::make_shared(*trx), deadline);if(trace->except) {if(failure_is_subjective(*trace->except, deadline_is_subjective)) { _pending_incoming_transactions.emplace_back(trx, persist_until_expired, next); }else{autoe_ptr = trace->except->dynamic_copy_exception(); send_response(e_ptr); } }else{if(persist_until_expired) {// if this trx didnt fail/soft-fail and the persist flag is set, store its ID so that we can// ensure its applied to all future speculative blocks as well._persistent_transactions.insert(transaction_id_with_expiry{trx->id(), trx->expiration()}); } send_response(trace); } }catch(constguard_exception& e ) { app().get_plugin().handle_guard_exception(e); }catch( boost::interprocess::bad_alloc& ) { raise(SIGUSR1); } CATCH_AND_CALL(send_response); }
注意上面的is_known_unexpired_transaction，程式碼如下：
boolcontroller::is_known_unexpired_transaction(consttransaction_id_type& id)const{returndb().find(id);}
與之對應的是這個函式：
voidtransaction_context::record_transaction(consttransaction_id_type& id, fc::time_point_sec expire ) {try{ control.db().create([&](transaction_object& transaction) { transaction.trx_id = id; transaction.expiration = expire; }); }catch(constboost::interprocess::bad_alloc& ) {throw; }catch( ... ) { EOS_ASSERT(false, tx_duplicate,"duplicate transaction ${id}", ("id", id ) ); } }/// record_transaction
在push_transaction中會呼叫到，記錄trx已經被處理過了。
下面我們來看看send_response這個回撥：
autosend_response = [this, &trx, &next](constfc::static_variant& response) { next(response);if(response.contains()) { _transaction_ack_channel.publish(std::pair(response.get(), trx)); }else{ _transaction_ack_channel.publish(std::pair(nullptr, trx)); } };
在執行之後會呼叫send_response，這裡是將結果傳送到_transaction_ack_channel中，對於_transaction_ack_channel， 這個實際對應的是下面這個型別：
namespacecompat {namespacechannels {usingtransaction_ack = channel_decl>; } }
在EOS中在net_plugin註冊響應這個channel的函式:
my->incoming_transaction_ack_subscription =
app().get_channel<channels::transaction_ack>().subscribe(
boost::bind(&net_plugin_impl::transaction_ack, my.get(), _1));
處理的函式如下：
voidnet_plugin_impl::transaction_ack(conststd::pair& results) { transaction_id_type id = results.second->id();if(results.first) { fc_ilog(logger,"signaled NACK, trx-id = ${id} : ${why}",("id", id)("why", results.first->to_detail_string())); dispatcher->rejected_transaction(id); }else{ fc_ilog(logger,"signaled ACK, trx-id = ${id}",("id", id)); dispatcher->bcast_transaction(*results.second); } }
這裡會將執行正常的廣播給其他節點，這其中會發送給超級節點打包入塊，打包過程可以參見 https://eosforce.github.io/Documentation/#/zh-cn/code/block_produce 。
3. push_transaction程式碼分析
這裡我們來分析下push_transaction的過程，作為執行trx的入口，這個函式在EOS中非常重要，另一方面，這裡EOS與Eosforce有一定區別，這裡會具體介紹。
TODO 需要一個流程圖，不過部落格還不支援
3.1 transaction_metadata
我們先來看下push_transaction的transaction_metadata引數， 這個引數統一了各種不同型別，不同行為的trx：
/**
* This data structure should store context-free cached data about a transaction such as
* packed/unpacked/compressed and recovered keys
*/
class transaction_metadata {
public:
transaction_id_type id; // trx ID
transaction_id_type signed_id; // signed trx ID
signed_transaction trx;
packed_transaction packed_trx;
optional<pair<chain_id_type, flat_set<public_key_type>>> signing_keys;
bool accepted = false; // 標註是否呼叫了accepted訊號，確保只調用一次
bool implicit = false; // 是否忽略檢查
bool scheduled = false; // 是否是延遲trx
explicit transaction_metadata( const signed_transaction& t, packed_transaction::compression_type c = packed_transaction::none )
:trx(t),packed_trx(t, c) {
id = trx.id();
//raw_packed = fc::raw::pack( static_cast<const transaction&>(trx) );
signed_id = digest_type::hash(packed_trx);
}
explicit transaction_metadata( const packed_transaction& ptrx )
:trx( ptrx.get_signed_transaction() ), packed_trx(ptrx) {
id = trx.id();
//raw_packed = fc::raw::pack( static_cast<const transaction&>(trx) );
signed_id = digest_type::hash(packed_trx);
}
const flat_set<public_key_type>& recover_keys( const chain_id_type& chain_id );
uint32_t total_actions()const { return trx.context_free_actions.size() trx.actions.size(); }
};
using transaction_metadata_ptr = std::shared_ptr<transaction_metadata>;
先看一下implicit，這個引數指示下面的邏輯是否要忽略對於trx的各種檢查，一般用於系統內部的trx， 對於EOS，主要是處理on_block_transaction（可以參見出塊文件），在start_block呼叫：
...autoonbtrx =std::make_shared( get_on_block_transaction() ); onbtrx->implicit =true;// on_block trx 會被無條件接受autoreset_in_trx_requiring_checks = fc::make_scoped_exit([old_value=in_trx_requiring_checks,this](){ in_trx_requiring_checks = old_value; }); in_trx_requiring_checks =true;// 修改in_trx_requiring_checks變數達到不將trx寫入區塊，一些系統的trx沒有必要寫入區塊。push_transaction( onbtrx, fc::time_point::maximum(), self.get_global_properties().configuration.min_transaction_cpu_usage,true);...
!> Eosforce不同之處
而對於EOSForce中，除了on_block action之外，onfee合約也是被設定為implicit==true的，onfee合約是eosforce的系統合約，設計用來收取交易的手續費。
3.2 push_transaction函式
下面我們逐行分析下程式碼，EOS中push_transaction程式碼如下：
/**
* This is the entry point for new transactions to the block state. It will check authorization and
* determine whether to execute it now or to delay it. Lastly it inserts a transaction receipt into
* the pending block.
*/transaction_trace_ptrpush_transaction(consttransaction_metadata_ptr& trx, fc::time_point deadline,uint32_tbilled_cpu_time_us,boolexplicit_billed_cpu_time =false){// deadline必須不為空// deadline是trx執行時間的一個大上限，為了防止某些trx執行時間過長導致出塊失敗等問題，// 這裡必須有一個嚴格的上限，一旦超過上限，交易會立即失敗。EOS_ASSERT(deadline != fc::time_point(), transaction_exception,"deadline cannot be uninitialized"); transaction_trace_ptr trace;// trace主要用來儲存執行中的一些錯誤資訊。try{// trx_context是執行trx的上下文狀態，下面會專門說明transaction_contexttrx_context(self, trx->trx, trx->id);if((bool)subjective_cpu_leeway && pending->_block_status == controller::block_status::incomplete) { trx_context.leeway = *subjective_cpu_leeway; }// 設定資料trx_context.deadline = deadline; trx_context.explicit_billed_cpu_time = explicit_billed_cpu_time; trx_context.billed_cpu_time_us = billed_cpu_time_us; trace = trx_context.trace;try{if( trx->implicit ) {// 如果是implicit的就沒有必要做下面的一些檢查和記錄，這裡的檢查主要是資源方面的trx_context.init_for_implicit_trx(); trx_context.can_subjectively_fail =false; }else{// 如果是重放並且不是重放過程中接到的新交易，則不去使用`record_transaction`記錄boolskip_recording = replay_head_time && (time_point(trx->trx.expiration) <= *replay_head_time);// 一些trx_context的初始化操作trx_context.init_for_input_trx( trx->packed_trx.get_unprunable_size(), trx->packed_trx.get_prunable_size(), trx->trx.signatures.size(), skip_recording); }if( trx_context.can_subjectively_fail && pending->_block_status == controller::block_status::incomplete ) { check_actor_list( trx_context.bill_to_accounts );// Assumes bill_to_accounts is the set of actors authorizing the transaction} trx_context.delay = fc::seconds(trx->trx.delay_sec);if( !self.skip_auth_check() && !trx->implicit ) {// 檢測交易所需要的許可權authorization.check_authorization( trx->trx.actions, trx->recover_keys( chain_id ), {}, trx_context.delay, [](){}/*std::bind(&transaction_context::add_cpu_usage_and_check_time, &trx_context,
std::placeholders::_1)*/,false); }// 執行，注意這時trx_context包括所有資訊和狀態trx_context.exec(); trx_context.finalize();// Automatically rounds up network and CPU usage in trace and bills payers if successfulautorestore = make_block_restore_point();if(!trx->implicit) {// 如果是非implicit的交易，則需要進入區塊。transaction_receipt::status_enum s = (trx_context.delay == fc::seconds(0)) ? transaction_receipt::executed : transaction_receipt::delayed; trace->receipt = push_receipt(trx->packed_trx, s, trx_context.billed_cpu_time_us, trace->net_usage); pending->_pending_block_state->trxs.emplace_back(trx); }else{// 注意，這裡implicit類的交易是不會進入區塊的，只會計入資源消耗// 因為這類的trx無條件執行，所以不需要另行記錄。transaction_receipt_header r; r.status = transaction_receipt::executed; r.cpu_usage_us = trx_context.billed_cpu_time_us; r.net_usage_words = trace->net_usage /8; trace->receipt = r; }// 這裡會將執行過的action寫入待出塊狀態的_actions之中fc::move_append(pending->_actions, move(trx_context.executed));// call the accept signal but only once for this transaction// 為這個交易呼叫accept訊號，保證只調用一次if(!trx->accepted) { trx->accepted =true; emit( self.accepted_transaction, trx); }// 觸發applied_transaction訊號emit(self.applied_transaction, trace);if( read_mode != db_read_mode::SPECULATIVE && pending->_block_status == controller::block_status::incomplete ) {//this may happen automatically in destructor, but I prefere make it more explicittrx_context.undo(); }else{ restore.cancel(); trx_context.squash(); }// implicit的trx壓根沒有在unapplied_transactions中if(!trx->implicit) { unapplied_transactions.erase( trx->signed_id ); }returntrace; }catch(constfc::exception& e) { trace->except = e; trace->except_ptr =std::current_exception(); }// 注意這裡，如果成功的話上面就返回了這裡是失敗的情況// failure_is_subjective 表明if(!failure_is_subjective(*trace->except)) { unapplied_transactions.erase( trx->signed_id ); } emit( self.accepted_transaction, trx ); emit( self.applied_transaction, trace );returntrace; } FC_CAPTURE_AND_RETHROW((trace)) }/// push_transaction
上面註釋中闡述了大致的流程，下面仔細分析一下：
首先是trx_context，這個物件的類宣告如下：
classtransaction_context{...// 省略voiddispatch_action( action_trace& trace,constaction& a, account_name receiver,boolcontext_free =false,uint32_trecurse_depth =0);inlinevoiddispatch_action( action_trace& trace,constaction& a,boolcontext_free =false){ dispatch_action(trace, a, a.account, context_free); };voidschedule_transaction();voidrecord_transaction(consttransaction_id_type& id, fc::time_point_sec expire );voidvalidate_cpu_usage_to_bill(int64_tu,boolcheck_minimum =true)const;public: controller& control;// controller類的引用constsigned_transaction& trx;// 要執行的trxtransaction_id_type id; optional undo_session; transaction_trace_ptr trace;// 記錄錯誤的tracefc::time_point start;// 起始時刻fc::time_point published;// publish的時刻vector executed;// 執行完成的actionflat_set bill_to_accounts; flat_set validate_ram_usage;/// the maximum number of virtual CPU instructions of the transaction that can be safely billed to the billable accountsuint64_tinitial_max_billable_cpu =0; fc::microseconds delay;boolis_input =false;boolapply_context_free =true;boolcan_subjectively_fail =true; fc::time_point deadline = fc::time_point::maximum(); fc::microseconds leeway = fc::microseconds(3000);int64_tbilled_cpu_time_us =0;boolexplicit_billed_cpu_time =false;private:boolis_initialized =false;uint64_tnet_limit =0;boolnet_limit_due_to_block =true;boolnet_limit_due_to_greylist =false;uint64_teager_net_limit =0;uint64_t& net_usage;/// reference to trace->net_usageboolcpu_limit_due_to_greylist =false; fc::microseconds initial_objective_duration_limit; fc::microseconds objective_duration_limit; fc::time_point _deadline = fc::time_point::maximum();int64_tdeadline_exception_code = block_cpu_usage_exceeded::code_value;int64_tbilling_timer_exception_code = block_cpu_usage_exceeded::code_value; fc::time_point pseudo_start; fc::microseconds billed_time; fc::microseconds billing_timer_duration_limit; };
我們先看一下init_for_input_trx：
voidtransaction_context::init_for_input_trx(uint64_tpacked_trx_unprunable_size,// 這個是指trx打包後完整的大小uint64_tpacked_trx_prunable_size,// 這個指trx額外資訊的大小uint32_tnum_signatures,// 這個引數沒用上boolskip_recording )// 是否要跳過記錄{// 根據cfg和trx初始化資源constauto& cfg = control.get_global_properties().configuration;// 利用packed_trx_unprunable_size和packed_trx_prunable_size 計算net資源消耗uint64_tdiscounted_size_for_pruned_data = packed_trx_prunable_size;if( cfg.context_free_discount_net_usage_den >0&& cfg.context_free_discount_net_usage_num < cfg.context_free_discount_net_usage_den ) { discounted_size_for_pruned_data *= cfg.context_free_discount_net_usage_num; discounted_size_for_pruned_data = ( discounted_size_for_pruned_data cfg.context_free_discount_net_usage_den -1) / cfg.context_free_discount_net_usage_den;// rounds up}uint64_tinitial_net_usage =static_cast(cfg.base_per_transaction_net_usage) packed_trx_unprunable_size discounted_size_for_pruned_data;// 對於delay trx需要額外的net資源if( trx.delay_sec.value >0) {// If delayed, also charge ahead of time for the additional net usage needed to retire the delayed transaction// whether that be by successfully executing, soft failure, hard failure, or expiration.initial_net_usage =static_cast(cfg.base_per_transaction_net_usage) static_cast(config::transaction_id_net_usage); }// 初始化一些資訊published = control.pending_block_time(); is_input =true;if(!control.skip_trx_checks()) { control.validate_expiration(trx); control.validate_tapos(trx); control.validate_referenced_accounts(trx); } init( initial_net_usage);// 這裡呼叫init函式， 在這個函式中會處理cpu資源和ram資源if(!skip_recording)// 將trx新增入記錄中record_transaction( id, trx.expiration );/// checks for dupes}
這裡會先計算net，再在init函式中處理其他資源：
voidtransaction_context::init(uint64_tinitial_net_usage) { EOS_ASSERT( !is_initialized, transaction_exception,"cannot initialize twice");conststaticint64_tlarge_number_no_overflow =std::numeric_limits::max()/2;constauto& cfg = control.get_global_properties().configuration;auto& rl = control.get_mutable_resource_limits_manager(); net_limit = rl.get_block_net_limit(); objective_duration_limit = fc::microseconds( rl.get_block_cpu_limit() ); _deadline = start objective_duration_limit;// Possibly lower net_limit to the maximum net usage a transaction is allowed to be billedif( cfg.max_transaction_net_usage <= net_limit ) { net_limit = cfg.max_transaction_net_usage; net_limit_due_to_block =false; }// Possibly lower objective_duration_limit to the maximum cpu usage a transaction is allowed to be billedif( cfg.max_transaction_cpu_usage <= objective_duration_limit.count() ) { objective_duration_limit = fc::microseconds(cfg.max_transaction_cpu_usage); billing_timer_exception_code = tx_cpu_usage_exceeded::code_value; _deadline = start objective_duration_limit; }// Possibly lower net_limit to optional limit set in the transaction headeruint64_ttrx_specified_net_usage_limit =static_cast(trx.max_net_usage_words.value) *8;if( trx_specified_net_usage_limit >0&& trx_specified_net_usage_limit <= net_limit ) { net_limit = trx_specified_net_usage_limit; net_limit_due_to_block =false; }// Possibly lower objective_duration_limit to optional limit set in transaction headerif( trx.max_cpu_usage_ms >0) {autotrx_specified_cpu_usage_limit = fc::milliseconds(trx.max_cpu_usage_ms);if( trx_specified_cpu_usage_limit <= objective_duration_limit ) { objective_duration_limit = trx_specified_cpu_usage_limit; billing_timer_exception_code = tx_cpu_usage_exceeded::code_value; _deadline = start objective_duration_limit; } } initial_objective_duration_limit = objective_duration_limit;if( billed_cpu_time_us >0)// could also call on explicit_billed_cpu_time but it would be redundantvalidate_cpu_usage_to_bill( billed_cpu_time_us,false);// Fail early if the amount to be billed is too high// Record accounts to be billed for network and CPU usagefor(constauto& act : trx.actions ) {for(constauto& auth : act.authorization ) { bill_to_accounts.insert( auth.actor ); } } validate_ram_usage.reserve( bill_to_accounts.size() );// Update usage values of accounts to reflect new timerl.update_account_usage( bill_to_accounts, block_timestamp_type(control.pending_block_time()).slot );// Calculate the highest network usage and CPU time that all of the billed accounts can afford to be billedint64_taccount_net_limit =0;int64_taccount_cpu_limit =0;boolgreylisted_net =false, greylisted_cpu =false;std::tie( account_net_limit, account_cpu_limit, greylisted_net, greylisted_cpu) = max_bandwidth_billed_accounts_can_pay(); net_limit_due_to_greylist |= greylisted_net; cpu_limit_due_to_greylist |= greylisted_cpu; eager_net_limit = net_limit;// Possible lower eager_net_limit to what the billed accounts can pay plus some (objective) leewayautonew_eager_net_limit =std::min( eager_net_limit,static_cast(account_net_limit cfg.net_usage_leeway) );if( new_eager_net_limit < eager_net_limit ) { eager_net_limit = new_eager_net_limit; net_limit_due_to_block =false; }// Possibly limit deadline if the duration accounts can be billed for ( a subjective leeway) does not exceed current deltaif( (fc::microseconds(account_cpu_limit) leeway) <= (_deadline - start) ) { _deadline = start fc::microseconds(account_cpu_limit) leeway; billing_timer_exception_code = leeway_deadline_exception::code_value; } billing_timer_duration_limit = _deadline - start;// Check if deadline is limited by caller-set deadline (only change deadline if billed_cpu_time_us is not set)if( explicit_billed_cpu_time || deadline < _deadline ) { _deadline = deadline; deadline_exception_code = deadline_exception::code_value; }else{ deadline_exception_code = billing_timer_exception_code; } eager_net_limit = (eager_net_limit/8)*8;// Round down to nearest multiple of word size (8 bytes) so check_net_usage can be efficientif( initial_net_usage >0) add_net_usage( initial_net_usage );// Fail early if current net usage is already greater than the calculated limitchecktime();// Fail early if deadline has already been exceededis_initialized =true; }
以上就是transaction_context初始化過程，這裡主要是處理資源消耗。
下面是exec函式，這個函式很簡單：
voidtransaction_context::exec() { EOS_ASSERT( is_initialized, transaction_exception,"must first initialize");// 呼叫`dispatch_action`，這裡並沒有對上下文無關trx進行特別的操作，只是引數不同if( apply_context_free ) {for(constauto& act : trx.context_free_actions ) { trace->action_traces.emplace_back(); dispatch_action( trace->action_traces.back(), act,true); } }if( delay == fc::microseconds() ) {for(constauto& act : trx.actions ) { trace->action_traces.emplace_back(); dispatch_action( trace->action_traces.back(), act ); } }else{// 對於延遲交易，這裡特別處理schedule_transaction(); } }
主要執行在dispatch_action中，這裡會根據action不同分別觸發對應的呼叫：
voidtransaction_context::dispatch_action( action_trace& trace,constaction& a, account_name receiver,boolcontext_free,uint32_trecurse_depth ) {// 構建apply_context執行action， apply_context的分析在下節進行apply_contextacontext( control, *this, a, recurse_depth ); acontext.context_free = context_free; acontext.receiver = receiver;try{ acontext.exec(); }catch( ... ) { trace = move(acontext.trace);throw; }// 彙總結果到tracetrace = move(acontext.trace); }
對於延遲交易，執行schedule_transaction：
voidtransaction_context::schedule_transaction() {// 因為交易延遲執行，會消耗額外的net和ram資源// Charge ahead of time for the additional net usage needed to retire the delayed transaction// whether that be by successfully executing, soft failure, hard failure, or expiration.if( trx.delay_sec.value ==0) {// Do not double bill. Only charge if we have not already charged for the delay.constauto& cfg = control.get_global_properties().configuration; add_net_usage(static_cast(cfg.base_per_transaction_net_usage) static_cast(config::transaction_id_net_usage) );// Will exit early if net usage cannot be payed.}autofirst_auth = trx.first_authorizor();// 將延遲交易寫入節點執行時狀態資料庫中，到時會從這裡查找出來執行uint32_ttrx_size =0;constauto& cgto = control.db().create( [&](auto& gto ) { gto.trx_id = id; gto.payer = first_auth; gto.sender = account_name();/// delayed transactions have no sendergto.sender_id = transaction_id_to_sender_id( gto.trx_id ); gto.published = control.pending_block_time(); gto.delay_until = gto.published delay; gto.expiration = gto.delay_until fc::seconds(control.get_global_properties().configuration.deferred_trx_expiration_window); trx_size = gto.set( trx ); });// 因為要寫記憶體記錄，所以也消耗了一定的ramadd_ram_usage( cgto.payer, (config::billable_size_v trx_size) ); }
呼叫完exec之後會呼叫transaction_context::finalize()：
// 這裡主要是處理資源消耗voidtransaction_context::finalize() { EOS_ASSERT( is_initialized, transaction_exception,"must first initialize");if( is_input ) {auto& am = control.get_mutable_authorization_manager();for(constauto& act : trx.actions ) {for(constauto& auth : act.authorization ) { am.update_permission_usage( am.get_permission(auth) ); } } }auto& rl = control.get_mutable_resource_limits_manager();for(autoa : validate_ram_usage ) { rl.verify_account_ram_usage( a ); }// Calculate the new highest network usage and CPU time that all of the billed accounts can afford to be billedint64_taccount_net_limit =0;int64_taccount_cpu_limit =0;boolgreylisted_net =false, greylisted_cpu =false;std::tie( account_net_limit, account_cpu_limit, greylisted_net, greylisted_cpu) = max_bandwidth_billed_accounts_can_pay(); net_limit_due_to_greylist |= greylisted_net; cpu_limit_due_to_greylist |= greylisted_cpu;// Possibly lower net_limit to what the billed accounts can payif(static_cast(account_net_limit) <= net_limit ) {//NOTE:net_limit may possibly not be objective anymore due to net greylisting, but it should still be no greater than the truly objective net_limitnet_limit =static_cast(account_net_limit); net_limit_due_to_block =false; }// Possibly lower objective_duration_limit to what the billed accounts can payif( account_cpu_limit <= objective_duration_limit.count() ) {//NOTE:objective_duration_limit may possibly not be objective anymore due to cpu greylisting, but it should still be no greater than the truly objective objective_duration_limitobjective_duration_limit = fc::microseconds(account_cpu_limit); billing_timer_exception_code = tx_cpu_usage_exceeded::code_value; } net_usage = ((net_usage 7)/8)*8;// Round up to nearest multiple of word size (8 bytes)eager_net_limit = net_limit; check_net_usage();autonow = fc::time_point::now(); trace->elapsed = now - start; update_billed_cpu_time( now ); validate_cpu_usage_to_bill( billed_cpu_time_us ); rl.add_transaction_usage( bill_to_accounts,static_cast(billed_cpu_time_us), net_usage, block_timestamp_type(control.pending_block_time()).slot );// Should never fail}
接下來make_block_restore_point，這裡添加了一個檢查點：
// The returned scoped_exit should not exceed the lifetime of the pending which existed when make_block_restore_point was called.fc::scoped_exit> make_block_restore_point() {autoorig_block_transactions_size = pending->_pending_block_state->block->transactions.size();autoorig_state_transactions_size = pending->_pending_block_state->trxs.size();autoorig_state_actions_size = pending->_actions.size();std::function callback = [this, orig_block_transactions_size, orig_state_transactions_size, orig_state_actions_size]() { pending->_pending_block_state->block->transactions.resize(orig_block_transactions_size); pending->_pending_block_state->trxs.resize(orig_state_transactions_size); pending->_actions.resize(orig_state_actions_size); };returnfc::make_scoped_exit(std::move(callback) ); }
而後對於不是implicit的交易會呼叫push_receipt，這裡會將trx寫入區塊資料中，這也意味著implicit為true的交易雖然執行了，但不會在區塊中。
/**
* Adds the transaction receipt to the pending block and returns it.
*/templateconsttransaction_receipt&push_receipt(constT& trx, transaction_receipt_header::status_enum status,uint64_tcpu_usage_us,uint64_tnet_usage ){uint64_tnet_usage_words = net_usage /8; EOS_ASSERT( net_usage_words*8== net_usage, transaction_exception,"net_usage is not divisible by 8"); pending->_pending_block_state->block->transactions.emplace_back( trx ); transaction_receipt& r = pending->_pending_block_state->block->transactions.back(); r.cpu_usage_us = cpu_usage_us; r.net_usage_words = net_usage_words; r.status = status;returnr; }
上面的邏輯很大程度上和implicit為true時的邏輯重複，估計以後會重構。
接下來值得注意的是這裡：
if( read_mode != db_read_mode::SPECULATIVE && pending->_block_status == controller::block_status::incomplete ) {//this may happen automatically in destructor, but I prefere make it more explicittrx_context.undo(); }else{ restore.cancel(); trx_context.squash(); }
TODO trx_context.undo
這裡呼叫database::session對應的函式，
!> Eosforce不同之處
以上是EOS的流程，這裡我們再來看看Eosforce的不同之處，Eosforce與EOS一個明顯的不同是Eosforce採用了基於手續費的資源模型， 這種模型意味著，如果一個交易在超級節點打包進塊時失敗了，此時也要收取手續費，否則會造成潛在的攻擊風險，所以Eosforce中，執行失敗的交易也會寫入區塊中，這樣每次執行時會呼叫對應onfee。 另一方面, Eosforce雖然使用手續費，但是還是區分cpu，net，ram資源，並且在大的限制上依然進行檢查。 後續Eosforce會完成新的資源模型，這裡會有所改動。
Eosforce中的push_transaction函式如下：
transaction_trace_ptrpush_transaction(consttransaction_metadata_ptr& trx, fc::time_point deadline,uint32_tbilled_cpu_time_us,boolexplicit_billed_cpu_time =false){ EOS_ASSERT(deadline != fc::time_point(), transaction_exception,"deadline cannot be uninitialized");// eosforce暫時沒有開放延遲交易和上下文無關交易EOS_ASSERT(trx->trx.delay_sec.value ==0UL, transaction_exception,"delay,transaction failed"); EOS_ASSERT(trx->trx.context_free_actions.size()==0, transaction_exception,"context free actions size should be zero!");// 在eosforce中，為了安全性，對於特定一些交易進行了額外的驗證，主要是考慮到，系統會將執行錯誤的交易寫入區塊// 此時就要先驗證下交易內容，特別是大小上有沒有超出限制，否則將會帶來安全問題。check_action(trx->trx.actions); transaction_trace_ptr trace;try{// 一樣的程式碼 略去...try{// 一樣的程式碼 略去...// 處理手續費EOS_ASSERT(trx->trx.fee == txfee.get_required_fee(trx->trx), transaction_exception,"set tx fee failed"); EOS_ASSERT(txfee.check_transaction(trx->trx) ==true, transaction_exception,"transaction include actor more than one");try{// 這裡會執行onfee合約，也是通過`push_transaction`實現的autoonftrx =std::make_shared( get_on_fee_transaction(trx->trx.fee, trx->trx.actions[0].authorization[0].actor) ); onftrx->implicit =true;autoonftrace = push_transaction( onftrx, fc::time_point::maximum(), config::default_min_transaction_cpu_usage,true);// 這裡如果執行失敗直接丟擲異常，不會執行下面的東西if( onftrace->except )throw*onftrace->except; }catch(constfc::exception &e) { EOS_ASSERT(false, transaction_exception,"on fee transaction failed, exception: ${e}", ("e", e)); }catch( ... ) { EOS_ASSERT(false, transaction_exception,"on fee transaction failed, but shouldn't enough asset to pay for transaction fee"); } }// 注意這一層try catch，因為eos中出錯的交易會被拋棄，所以eos的異常會被直接丟擲到外層// 而在eosforce中出錯的交易會進入區塊// 但是要注意，這裡如果這裡並不是在超級節點出塊時呼叫，雖然也會執行下面的邏輯，但是不會被轉發給超級節點。try{if(explicit_billed_cpu_time && billed_cpu_time_us ==0){// 在eosforce中 因為超級節點打包區塊時失敗的交易也會被寫入區塊中，// 而很多交易失敗的原因不是交易本身有問題，而是在執行交易時，資源上限被觸發，導致交易被直接判定為失敗，// 這時寫入區塊的交易的cpu消耗是0, 這裡是需要失敗的，否則重跑區塊時會出現不同步的情況EOS_ASSERT(false, transaction_exception,"billed_cpu_time_us is 0"); } trx_context.exec(); trx_context.finalize();// Automatically rounds up network and CPU usage in trace and bills payers if successful}catch(constfc::exception &e) { trace->except = e; trace->except_ptr =std::current_exception();// eosforce加了一些日誌if(head->block_num !=1) { elog("---trnasction exe failed--------trace: ${trace}", ("trace", trace)); } }autorestore = make_block_restore_point();if(!trx->implicit) {// 這裡不太好的地方是，對於出錯的交易也被標為`executed`（嚴格說也確實是executed），後續eosforce將會重構這裡transaction_receipt::status_enum s = (trx_context.delay == fc::seconds(0)) ? transaction_receipt::executed :