Merge pull request #1153 from AntelopeIO/GH-1145-replay-errors-main

[1.1.0 -> main] Fix deferred trx processing

Author: heifner

benchmark/bls.cpp

@@ -78,7 +78,8 @@ struct interface_in_benchmark {
       timer = std::make_unique<platform_timer>();
       trx_timer = std::make_unique<transaction_checktime_timer>(*timer);
       trx_ctx = std::make_unique<transaction_context>(*chain->control.get(), *ptrx, ptrx->id(), std::move(*trx_timer),
-                                                      action_digests_t::store_which_t::legacy);
+                                                      action_digests_t::store_which_t::legacy, fc::time_point::now(),
+                                                      transaction_metadata::trx_type::input);
       trx_ctx->max_transaction_time_subjective = fc::microseconds::maximum();
       trx_ctx->init_for_input_trx( ptrx->get_unprunable_size(), ptrx->get_prunable_size() );
       trx_ctx->exec(); // this is required to generate action traces to be used by apply_context constructor

libraries/chain/controller.cpp

@@ -2644,7 +2644,8 @@ struct controller_impl {
 
       transaction_checktime_timer trx_timer(timer);
       const packed_transaction trx( std::move( etrx ) );
-      transaction_context trx_context( self, trx, trx.id(), std::move(trx_timer), bb.action_receipt_digests().store_which(), start );
+      transaction_context trx_context( self, trx, trx.id(), std::move(trx_timer), bb.action_receipt_digests().store_which(),
+                                       start, transaction_metadata::trx_type::implicit );
 
       if (auto dm_logger = get_deep_mind_logger(trx_context.is_transient())) {
          dm_logger->on_onerror(etrx);
@@ -2690,6 +2691,8 @@ struct controller_impl {
       } catch ( const boost::interprocess::bad_alloc& ) {
          throw;
       } catch( const fc::exception& e ) {
+         // apply_onerror for deferred trxs is implicit so interrupt oc not allowed
+         assert(e.code() != interrupt_oc_exception::code_value);
          handle_exception(e);
       } catch ( const std::exception& e ) {
          auto wrapper = fc::std_exception_wrapper::from_current_exception(e);
@@ -2815,7 +2818,8 @@ struct controller_impl {
       auto& bb = std::get<building_block>(pending->_block_stage);
 
       transaction_checktime_timer trx_timer( timer );
-      transaction_context trx_context( self, *trx->packed_trx(), gtrx.trx_id, std::move(trx_timer), bb.action_receipt_digests().store_which() );
+      transaction_context trx_context( self, *trx->packed_trx(), gtrx.trx_id, std::move(trx_timer), bb.action_receipt_digests().store_which(),
+                                       start, transaction_metadata::trx_type::scheduled );
       trx_context.leeway =  fc::microseconds(0); // avoid stealing cpu resource
       trx_context.block_deadline = block_deadline;
       trx_context.max_transaction_time_subjective = max_transaction_time;

libraries/chain/include/eosio/chain/transaction_context.hpp

@@ -114,8 +114,8 @@ namespace eosio::chain {
                               const transaction_id_type& trx_id, // trx_id diff than t.id() before replace_deferred
                               transaction_checktime_timer&& timer,
                               action_digests_t::store_which_t sad,
-                              fc::time_point start = fc::time_point::now(),
-                              transaction_metadata::trx_type type = transaction_metadata::trx_type::input);
+                              fc::time_point start,
+                              transaction_metadata::trx_type type);
          ~transaction_context();
 
          void init_for_implicit_trx();
@@ -162,6 +162,8 @@ namespace eosio::chain {
          bool is_dry_run()const { return trx_type == transaction_metadata::trx_type::dry_run; };
          bool is_read_only()const { return trx_type == transaction_metadata::trx_type::read_only; };
          bool is_transient()const { return trx_type == transaction_metadata::trx_type::read_only || trx_type == transaction_metadata::trx_type::dry_run; };
+         bool is_implicit()const { return trx_type == transaction_metadata::trx_type::implicit; };
+         bool is_scheduled()const { return trx_type == transaction_metadata::trx_type::scheduled; };
          bool has_undo()const;
 
          int64_t set_proposed_producers(vector<producer_authority> producers);

libraries/chain/include/eosio/chain/wasm_interface_private.hpp

@@ -164,7 +164,14 @@ struct eosvmoc_tier {
             }
          }
 #endif
-         const bool allow_oc_interrupt = attempt_tierup && context.is_applying_block() && context.trx_context.has_undo();
+         // Do not allow oc interrupt if no undo as the transaction needs to be undone to restart it.
+         // Do not allow oc interrupt if implicit or scheduled. There are two implicit trxs: onblock and onerror.
+         //   The onerror trx of deferred trxs is implicit. Interrupt needs to be disabled for deferred trxs because
+         //   they capture all exceptions, explicitly handle undo stack, and directly call trx_context.execute_action.
+         //   Not allowing interrupt for onblock seems rather harmless, so instead of distinguishing between onerror and
+         //   onblock, just disallow for all implicit.
+         const bool allow_oc_interrupt = attempt_tierup && context.is_applying_block() &&
+                                         context.trx_context.has_undo() && !context.trx_context.is_implicit() && !context.trx_context.is_scheduled();
          auto ex = fc::make_scoped_exit([&]() {
             if (allow_oc_interrupt) {
                eos_vm_oc_compile_interrupt = false;

libraries/chain/transaction_context.cpp

@@ -59,6 +59,7 @@ namespace eosio::chain {
       undo();
       *trace = transaction_trace{}; // reset trace
       initialize();
+      transaction_timer.stop();
       resume_billing_timer(start);
 
       auto sw = executed_action_receipts.store_which();

libraries/chain/webassembly/runtimes/eos-vm-oc/executor.cpp

@@ -231,7 +231,6 @@ void executor::execute(const code_descriptor& code, memory& mem, apply_context&
       syscall(SYS_mprotect, self->code_mapping, self->code_mapping_size, PROT_NONE);
       self->mapping_is_executable = false;
    }, this);
-   context.trx_context.checktime(); //catch any expiration that might have occurred before setting up callback
 
    auto cleanup = fc::make_scoped_exit([cb, &tt=context.trx_context.transaction_timer, &mem=mem](){
       cb->is_running = false;
@@ -245,6 +244,8 @@ void executor::execute(const code_descriptor& code, memory& mem, apply_context&
       }
    });
 
+   context.trx_context.checktime(); //catch any expiration that might have occurred before setting up callback
+
    void(*apply_func)(uint64_t, uint64_t, uint64_t) = (void(*)(uint64_t, uint64_t, uint64_t))(cb->running_code_base + code.apply_offset);
 
    switch(sigsetjmp(*cb->jmp, 0)) {