Pass runtime state instead of repeatedly fetching it from Thread.current #4621
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I ran the benchmarks with YJIT turned on and found that
Thread#[]was one of the top methods. I thought I might minimize the hits to that by refactoring a bit. In the end, the change was negligible. I think the reason is because, with YJIT running, most of the runtime ends in "misses," which I'm guessing is optimized code that stackprof can't see.Large Result Benchmark
Before, YJIT (9.453 i/s)
Calculating ------------------------------------- Querying for 1000 objects 9.453 (±10.6%) i/s - 95.000 in 10.096298s ================================== Mode: wall(1) Samples: 324246 (71.59% miss rate) GC: 11313 (3.49%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 25950 (8.0%) 25950 (8.0%) Kernel#class 20545 (6.3%) 20545 (6.3%) Thread#[] 19958 (6.2%) 19958 (6.2%) GraphQL::Execution::Lazy#initialize 13375 (4.1%) 13375 (4.1%) Kernel#hash 11176 (3.4%) 11176 (3.4%) GraphQL::Execution::Interpreter::Runtime#dead_result? 30562 (9.4%) 10012 (3.1%) GraphQL::Execution::Interpreter::Runtime#get_current_runtime_state 303235 (93.5%) 9393 (2.9%) GraphQL::Execution::Interpreter::Runtime#after_lazy 9290 (2.9%) 9290 (2.9%) GraphQL::Schema::Member::HasArguments#own_arguments 9076 (2.8%) 9076 (2.8%) GraphQL::Schema.lazy_methods 8739 (2.7%) 8739 (2.7%) Kernel#respond_to?Before, No YJIT (4.943 i/s)
Calculating ------------------------------------- Querying for 1000 objects 4.943 (± 0.0%) i/s - 50.000 in 10.147070s ================================== Mode: wall(1) Samples: 167739 (56.34% miss rate) GC: 7384 (4.40%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 104604 (62.4%) 9047 (5.4%) GraphQL::Execution::Interpreter::Runtime#continue_field 6650 (4.0%) 6650 (4.0%) (sweeping) 118832 (70.8%) 6473 (3.9%) GraphQL::Execution::Interpreter::Runtime#after_lazy 74223 (44.2%) 6092 (3.6%) GraphQL::Execution::Interpreter::Runtime#evaluate_selection_with_resolved_keyword_args 103127 (61.5%) 5410 (3.2%) GraphQL::Execution::Interpreter::Runtime#resolve_list_item 6590 (3.9%) 4702 (2.8%) GraphQL::Execution::Interpreter::Runtime#lazy? 82289 (49.1%) 4512 (2.7%) GraphQL::Execution::Interpreter::Runtime#evaluate_selection 7029 (4.2%) 4276 (2.5%) GraphQL::Execution::Interpreter::Runtime#get_current_runtime_state 19960 (11.9%) 4163 (2.5%) GraphQL::Schema::Field#resolve 3549 (2.1%) 3549 (2.1%) GraphQL::Execution::Lazy#initialize 8347 (5.0%) 3211 (1.9%) GraphQL::Execution::Interpreter::Runtime#set_result 14259 (8.5%) 2905 (1.7%) GraphQL::Schema::Field#with_extensions 121394 (72.4%) 2771 (1.7%) Array#eachAfter, YJIT (9.558 i/s)
Calculating ------------------------------------- Querying for 1000 objects 9.558 (±10.5%) i/s - 96.000 in 10.085145s ================================== Mode: wall(1) Samples: 299263 (71.11% miss rate) GC: 11417 (3.82%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 29462 (9.8%) 29462 (9.8%) Kernel#class 17072 (5.7%) 17072 (5.7%) GraphQL::Execution::Lazy#initialize 13158 (4.4%) 13158 (4.4%) Kernel#hash 11492 (3.8%) 11492 (3.8%) Thread#[] 8537 (2.9%) 8537 (2.9%) GraphQL::Schema.lazy_methods 35112 (11.7%) 8318 (2.8%) GraphQL::Execution::Interpreter::Runtime#lazy? 8161 (2.7%) 8161 (2.7%) Kernel#respond_to? 8101 (2.7%) 8101 (2.7%) GraphQL::Schema::Member::HasArguments#own_argumentsAfter, No YJIT (5.188 i/s)
Calculating ------------------------------------- Querying for 1000 objects 5.188 (± 0.0%) i/s - 52.000 in 10.032891s ================================== Mode: wall(1) Samples: 162494 (56.02% miss rate) GC: 9416 (5.79%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 8440 (5.2%) 8440 (5.2%) (sweeping) 99386 (61.2%) 8373 (5.2%) GraphQL::Execution::Interpreter::Runtime#continue_field 112920 (69.5%) 6398 (3.9%) GraphQL::Execution::Interpreter::Runtime#after_lazy 66843 (41.1%) 5444 (3.4%) GraphQL::Execution::Interpreter::Runtime#evaluate_selection_with_resolved_keyword_args 97897 (60.2%) 4906 (3.0%) GraphQL::Execution::Interpreter::Runtime#resolve_list_item 6704 (4.1%) 4662 (2.9%) GraphQL::Execution::Interpreter::Runtime#lazy?Introspection Benchmark
Before, YJIT (6.369 (± 0.0%) i/s)
Calculating ------------------------------------- Run large introspection 6.369 (± 0.0%) i/s - 64.000 in 10.065125s ================================== Mode: wall(1000) Samples: 161 (4.17% miss rate) GC: 4 (2.48%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 22 (13.7%) 22 (13.7%) Kernel#class 13 (8.1%) 13 (8.1%) Kernel#hash 57 (35.4%) 8 (5.0%) GraphQL::Schema::Field#resolve 8 (5.0%) 8 (5.0%) GraphQL::Execution::Interpreter::Runtime#dead_result? 12 (7.5%) 6 (3.7%) GraphQL::Schema::Warden#get_field 28 (17.4%) 6 (3.7%) GraphQL::Execution::Interpreter::Runtime#lazy? 6 (3.7%) 6 (3.7%) GraphQL::Schema::Member::HasArguments#own_arguments 4 (2.5%) 4 (2.5%) GraphQL::Schema::EnumValue#value 4 (2.5%) 4 (2.5%) Thread#[] 4 (2.5%) 4 (2.5%) Kernel#respond_to? 4 (2.5%) 4 (2.5%) GraphQL::Schema::Field#typeBefore, No YJIT (2.766 (± 0.0%) i/s)
Calculating ------------------------------------- Run large introspection 2.766 (± 0.0%) i/s - 28.000 in 10.129171s ================================== Mode: wall(1000) Samples: 375 (1.57% miss rate) GC: 3 (0.80%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 369 (98.4%) 42 (11.2%) GraphQL::Execution::Interpreter::Runtime#evaluate_selection_with_resolved_keyword_args 98 (26.1%) 41 (10.9%) GraphQL::Schema::Field#resolve 369 (98.4%) 28 (7.5%) GraphQL::Execution::Interpreter::Runtime#continue_field 369 (98.4%) 28 (7.5%) GraphQL::Execution::Interpreter::Runtime#evaluate_selection 16 (4.3%) 13 (3.5%) GraphQL::Execution::Interpreter::Runtime#get_current_runtime_state 91 (24.3%) 12 (3.2%) GraphQL::Query#after_lazy 13 (3.5%) 11 (2.9%) GraphQL::Execution::Interpreter::Runtime#lazy? 369 (98.4%) 10 (2.7%) GraphQL::Execution::Interpreter::Runtime#evaluate_selections 18 (4.8%) 9 (2.4%) GraphQL::Execution::Interpreter::Runtime#continue_value 9 (2.4%) 9 (2.4%) GraphQL::Execution::Interpreter::Runtime::GraphQLResultHash#set_leaf 24 (6.4%) 9 (2.4%) GraphQL::Execution::Interpreter::Runtime#set_result 9 (2.4%) 8 (2.1%) GraphQL::Schema::Warden#get_field 34 (9.1%) 7 (1.9%) Kernel#public_send 354 (94.4%) 7 (1.9%) GraphQL::Execution::Interpreter::Runtime#resolve_list_item 13 (3.5%) 7 (1.9%) GraphQL::Schema::Member::HasArguments::FieldConfigured#any_arguments? 369 (98.4%) 6 (1.6%) Hash#eachAfter, YJIT (6.450 (± 0.0%) i/s)
Calculating ------------------------------------- Run large introspection 6.450 (± 0.0%) i/s - 65.000 in 10.102631s ================================== Mode: wall(1000) Samples: 152 (5.00% miss rate) GC: 0 (0.00%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 17 (11.2%) 17 (11.2%) Kernel#class 12 (7.9%) 12 (7.9%) Kernel#respond_to? 9 (5.9%) 9 (5.9%) Kernel#hash 59 (38.8%) 9 (5.9%) GraphQL::Schema::Field#resolve 8 (5.3%) 8 (5.3%) BasicObject#== 6 (3.9%) 6 (3.9%) GraphQL::Schema::Field#type 6 (3.9%) 6 (3.9%) Kernel#is_a? 5 (3.3%) 5 (3.3%) GraphQL::Execution::Interpreter::Runtime#dead_result? 4 (2.6%) 4 (2.6%) Thread#[] 8 (5.3%) 4 (2.6%) GraphQL::Execution::Interpreter::Runtime#get_current_runtime_state 4 (2.6%) 4 (2.6%) GraphQL::Execution::Interpreter::Runtime::GraphQLResult#initialize 4 (2.6%) 4 (2.6%) GraphQL::Schema::Member::BaseDSLMethods#graphql_name 21 (13.8%) 4 (2.6%) GraphQL::Execution::Interpreter::Runtime#lazy? 3 (2.0%) 3 (2.0%) Hash#any?After, No YJIT (2.791 (± 0.0%) i/s)
Calculating ------------------------------------- Run large introspection 2.791 (± 0.0%) i/s - 28.000 in 10.036736s ================================== Mode: wall(1000) Samples: 375 (1.83% miss rate) GC: 7 (1.87%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 366 (97.6%) 41 (10.9%) GraphQL::Execution::Interpreter::Runtime#evaluate_selection_with_resolved_keyword_args 366 (97.6%) 38 (10.1%) GraphQL::Execution::Interpreter::Runtime#continue_field 90 (24.0%) 31 (8.3%) GraphQL::Schema::Field#resolve 366 (97.6%) 28 (7.5%) GraphQL::Execution::Interpreter::Runtime#evaluate_selection 15 (4.0%) 13 (3.5%) GraphQL::Execution::Interpreter::Runtime#lazy? 79 (21.1%) 12 (3.2%) GraphQL::Query#after_lazy 14 (3.7%) 12 (3.2%) GraphQL::Schema::Warden#get_field 366 (97.6%) 11 (2.9%) GraphQL::Execution::Interpreter::Runtime#evaluate_selections 11 (2.9%) 11 (2.9%) GraphQL::Execution::Interpreter::Runtime#get_current_runtime_state 22 (5.9%) 9 (2.4%) GraphQL::Schema::Object.authorized_new 18 (4.8%) 8 (2.1%) GraphQL::Execution::Interpreter::Runtime#continue_value 30 (8.0%) 8 (2.1%) Kernel#public_send 366 (97.6%) 7 (1.9%) Hash#eachIt seems like introspection even got slower, that's a bit of a surprise to me because I would have expected that case to benefit from fewer needless calls to re-fetch state.I updated the benchmark after removing another write to state and it's looking faster or the same now.