Add JSON writer benchmarks (#5314) #5317
Conversation
tustvold
force-pushed
the
json-writer-bench
branch
from
1eecd07
to
7f3c74c
Compare
tustvold
force-pushed
the
json-writer-bench
branch
from
7f3c74c
to
3c90e2b
Compare
| fn create_mixed(len: usize) -> RecordBatch { | ||
| let c1 = Arc::new(create_string_array::<i32>(len, 0.)); | ||
| let c2 = Arc::new(create_primitive_array::<Int32Type>(len, 0.)); | ||
| let c3 = Arc::new(create_primitive_array::<UInt32Type>(len, 0.)); |
There was a problem hiding this comment.
maybe we should also include a 64 bit type (like U64IntType rather than U32IntType)?
arrow/benches/json_writer.rs
Outdated
| } | ||
|
|
||
| fn bench_dict_array(c: &mut Criterion) { | ||
| let c1 = Arc::new(create_string_array::<i32>(NUM_ROWS, 0.)); |
There was a problem hiding this comment.
Is there a reason to include a string array in a benchmark for dictionary ? Maybe this is already covered by the separate benchmark for just strings?
| do_bench(c, "bench_struct", &batch) | ||
| } | ||
|
|
||
| fn bench_nullable_struct(c: &mut Criterion) { |
There was a problem hiding this comment.
💯 for including structs
Though again here I wonder if it is important to also have a string array or if we should just have a single struct array column)
There was a problem hiding this comment.
I wanted to have more than one column at the struct level, to break branch prediction
| } | ||
|
|
||
| fn bench_primitive(c: &mut Criterion) { | ||
| let c1 = Arc::new(create_primitive_array::<Float32Type>(NUM_ROWS, 0.)); |
There was a problem hiding this comment.
any reason to lump them together or would it make more sense to have a separate one for ints and flots?
There was a problem hiding this comment.
Habit, this is a good shout as float formatting is much more complex for JSON
Which issue does this PR close?
Part of #5314
Rationale for this change
We should get benchmarks before making changes to the serialization logic.
What changes are included in this PR?
Are there any user-facing changes?