GH-985 Correctly monitor outstanding compiles

libraries/chain/include/eosio/chain/webassembly/eos-vm-oc/code_cache.hpp

@@ -118,6 +118,7 @@ class code_cache_base {
       std::mutex                             _mtx;
       queued_compilies_t                     _queued_compiles;                  // protected by _mtx
       std::unordered_map<code_tuple, bool>   _outstanding_compiles_and_poison;  // protected by _mtx
+      std::atomic<size_t>                    _outstanding_compiles{0};
 
       size_t _free_bytes_eviction_threshold;
       void check_eviction_threshold(size_t free_bytes);

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

@@ -78,6 +78,8 @@ void code_cache_async::wait_on_compile_monitor_message() {
          return;
       }
 
+      --_outstanding_compiles;
+
       const auto& msg = std::get<wasm_compilation_result_message>(message);
       _result_queue.push(msg);
 
@@ -92,10 +94,11 @@ void code_cache_async::wait_on_compile_monitor_message() {
 //called from non-main thread
 void code_cache_async::process_queued_compiles() {
    std::lock_guard g(_mtx);
-   while (_outstanding_compiles_and_poison.size() < _threads && !_queued_compiles.empty()) {
+   while (_outstanding_compiles < _threads && !_queued_compiles.empty()) {
       auto nextup = _queued_compiles.begin();
 
       _outstanding_compiles_and_poison.emplace(nextup->code(), false);
+      ++_outstanding_compiles;
       FC_ASSERT(write_message_with_fds(_compile_monitor_write_socket, nextup->msg, nextup->fds_to_pass), "EOS VM failed to communicate to OOP manager");
 
       _queued_compiles.erase(nextup);
@@ -212,6 +215,7 @@ code_cache_async::get_descriptor_for_code(mode m, uint64_t executing_action_id,
    }
 
    _outstanding_compiles_and_poison.emplace(ct, false);
+   ++_outstanding_compiles;
    write_message_with_fds(_compile_monitor_write_socket, msg, fds_to_pass);
    failure = get_cd_failure::temporary; // Compile might not be done yet
    return nullptr;

unittests/eosvmoc_interrupt_tests.cpp

@@ -40,16 +40,16 @@ BOOST_AUTO_TEST_CASE( wasm_interrupt_test ) { try {
    auto pre_id = t.control->get_wasm_interface().get_executing_action_id();
 
    // Use an infinite executing action. When oc compile completes it will kill the action and restart it under
-   // eosvmoc. That action will then fail when it hits the 75000ms deadline.
-   // 75000ms has to be long enough for oc compile to complete and kill the non-oc executing transaction
+   // eosvmoc. That action will then fail when it hits the 5000ms deadline.
+   // 5000ms has to be long enough for oc compile to complete and kill the non-oc executing transaction
    BOOST_CHECK_THROW( push_trx( t, test_api_action<WASM_TEST_ACTION("test_checktime", "checktime_failure")>{},
-                                0, 150, 75000, true, fc::raw::pack(10000000000000000000ULL) ),
+                                0, 150, 5000, true, fc::raw::pack(10000000000000000000ULL) ),
                       deadline_exception );
 
    auto post_id = t.control->get_wasm_interface().get_executing_action_id();
 
    // each action uses 1 id, 2 if retried because of oc compile completion interruption
-   // if post_id == pre_id + 1, then likely that 75000ms above was not long enough for oc compile to complete
+   // if post_id == pre_id + 1, then likely that 5000ms above was not long enough for oc compile to complete
    BOOST_TEST(post_id == pre_id + 2);
 
    BOOST_REQUIRE_EQUAL( t.validate(), true );