Skip to content

Reduce allocs + improve benchmark table #74

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
csvance merged 2 commits into from
Dec 5, 2025
Merged

Reduce allocs + improve benchmark table #74

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4b87766
Reduce per request allocations
csvance Dec 5, 2025
e289687
Update benchmark table format + docs
csvance Dec 5, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 20 additions & 20 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,33 +22,33 @@ By default Julia 1.12 starts with just one thread. The closer to `@async` we get
However, it is unlikely Julia will be used this way in the real world.

```
╭──────────────────────────────────┬─────────┬────────┬─────────────┬──────────┬────────────┬──────────────┬─────────┬──────┬──────╮
│ Benchmark │ N │ Memory │ Allocations │ Duration │ Throughput │ Avg duration │ Std-dev │ Min │ Max │
│ │ calls │ MiB │ │ s │ calls/s │ μs │ μs │ μs │ μs │
├──────────────────────────────────┼─────────┼────────┼─────────────┼──────────┼────────────┼──────────────┼─────────┼──────┼──────┤
│ workload_smol │ 94000 │ 3.74 │ 85110 │ 5.01 │ 18756 │ 53 │ 4.2 │ 47 │ 71 │
│ workload_32_224_224_uint8 │ 2800 │ 63.78 │ 9230 │ 5.11 │ 548 │ 1826 │ 378.6 │ 1598 │ 2657 │
│ workload_streaming_request │ 2566000 │ 0.61 │ 6615 │ 4.99 │ 514001 │ 2 │ 0.61 │ 1 │ 16 │
│ workload_streaming_response │ 985000 │ 13.0 │ 27721 │ 5.0 │ 197101 │ 5 │ 0.48 │ 4 │ 7 │
│ workload_streaming_bidirectional │ 2568000 │ 1.98 │ 25503 │ 4.99 │ 514539 │ 2 │ 0.5 │ 1 │ 12 │
╰──────────────────────────────────┴─────────┴────────┴─────────────┴──────────┴────────────┴──────────────┴─────────┴──────┴──────╯
╭──────────────────────────────────┬─────────────┬────────────────┬────────────┬──────────────┬─────────┬──────┬──────╮
│ Benchmark │ Avg Memory │ Avg Allocs │ Throughput │ Avg duration │ Std-dev │ Min │ Max │
│ │ KiB/message │ allocs/message │ calls/s │ μs │ μs │ μs │ μs │
├──────────────────────────────────┼─────────────┼────────────────┼────────────┼──────────────┼─────────┼──────┼──────┤
│ workload_smol │ 2.95 │ 72.5 │ 18424 │ 54 │ 3.54 │ 48 │ 66 │
│ workload_32_224_224_uint8 │ 637.0 │ 79.1 │ 548 │ 1826 │ 405.48 │ 1602 │ 2730 │
│ workload_streaming_request │ 0.61 │ 6.6 │ 508983 │ 2 │ 0.67 │ 1 │ 15 │
│ workload_streaming_response │ 12.99 │ 27.6 │ 194689 │ 5 │ 0.52 │ 4 │ 9 │
│ workload_streaming_bidirectional │ 1.98 │ 25.5 │ 490718 │ 2 │ 0.59 │ 1 │ 13 │
╰──────────────────────────────────┴─────────────┴────────────────┴────────────┴──────────────┴─────────┴──────┴──────╯
```

### Real World: `julia -t auto`

Using more threads isn't great for async IO, but this is likely how most people will be using `gRPCClient.jl`.

```
╭──────────────────────────────────┬─────────┬────────┬─────────────┬──────────┬────────────┬──────────────┬─────────┬──────┬──────╮
│ Benchmark │ N │ Memory │ Allocations │ Duration │ Throughput │ Avg duration │ Std-dev │ Min │ Max │
│ │ calls │ MiB │ │ s │ calls/s │ μs │ μs │ μs │ μs │
├──────────────────────────────────┼─────────┼────────┼─────────────┼──────────┼────────────┼──────────────┼─────────┼──────┼──────┤
│ workload_smol │ 91000 │ 3.75 │ 85123 │ 5.03 │ 18079 │ 55 │ 3.96 │ 48 │ 67 │
│ workload_32_224_224_uint8 │ 2900 │ 63.78 │ 9188 │ 5.01 │ 579 │ 1728 │ 97.86 │ 1614 │ 1899 │
│ workload_streaming_request │ 1841000 │ 0.89 │ 6482 │ 4.99 │ 368669 │ 3 │ 1.35 │ 2 │ 21 │
│ workload_streaming_response │ 330000 │ 13.0 │ 27838 │ 5.02 │ 65771 │ 15 │ 5.2 │ 6 │ 37 │
│ workload_streaming_bidirectional │ 405000 │ 1.48 │ 25672 │ 5.0 │ 80948 │ 12 │ 8.52 │ 3 │ 62 │
╰──────────────────────────────────┴─────────┴────────┴─────────────┴──────────┴────────────┴──────────────┴─────────┴──────┴──────╯
╭──────────────────────────────────┬─────────────┬────────────────┬────────────┬──────────────┬─────────┬──────┬──────╮
│ Benchmark │ Avg Memory │ Avg Allocs │ Throughput │ Avg duration │ Std-dev │ Min │ Max │
│ │ KiB/message │ allocs/message │ calls/s │ μs │ μs │ μs │ μs │
├──────────────────────────────────┼─────────────┼────────────────┼────────────┼──────────────┼─────────┼──────┼──────┤
│ workload_smol │ 2.95 │ 72.5 │ 18014 │ 56 │ 3.08 │ 50 │ 64 │
│ workload_32_224_224_uint8 │ 637.0 │ 79.7 │ 567 │ 1762 │ 99.07 │ 1628 │ 1911 │
│ workload_streaming_request │ 0.86 │ 6.5 │ 341851 │ 3 │ 1.68 │ 2 │ 30 │
│ workload_streaming_response │ 13.0 │ 27.7 │ 64515 │ 16 │ 5.12 │ 6 │ 33 │
│ workload_streaming_bidirectional │ 1.41 │ 25.6 │ 102072 │ 10 │ 6.23 │ 4 │ 52 │
╰──────────────────────────────────┴─────────────┴────────────────┴────────────┴──────────────┴─────────┴──────┴──────╯
```

## Acknowledgement
Expand Down
20 changes: 10 additions & 10 deletions docs/src/index.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,16 +156,16 @@ benchmark_table()
```

```
╭──────────────────────────────────┬─────────┬────────┬─────────────┬──────────┬────────────┬──────────────┬─────────┬──────┬──────╮
│ Benchmark │ N │ Memory │ Allocations │ Duration │ Throughput │ Avg duration │ Std-dev │ Min │ Max │
│ │ calls │ MiB │ │ s │ calls/s │ μs │ μs │ μs │ μs │
├──────────────────────────────────┼─────────┼────────┼─────────────┼──────────┼────────────┼──────────────┼─────────┼──────┼──────┤
│ workload_smol │ 94000 │ 3.74 │ 85110 │ 5.01 │ 18756 │ 53 │ 4.2 │ 47 │ 71 │
│ workload_32_224_224_uint8 │ 2800 │ 63.78 │ 9230 │ 5.11 │ 548 │ 1826 │ 378.6 │ 1598 │ 2657 │
│ workload_streaming_request │ 2566000 │ 0.61 │ 6615 │ 4.99 │ 514001 │ 2 │ 0.61 │ 1 │ 16 │
│ workload_streaming_response │ 985000 │ 13.0 │ 27721 │ 5.0 │ 197101 │ 5 │ 0.48 │ 4 │ 7 │
│ workload_streaming_bidirectional │ 2568000 │ 1.98 │ 25503 │ 4.99 │ 514539 │ 2 │ 0.5 │ 1 │ 12 │
╰──────────────────────────────────┴─────────┴────────┴─────────────┴──────────┴────────────┴──────────────┴─────────┴──────┴──────╯
╭──────────────────────────────────┬─────────────┬────────────────┬────────────┬──────────────┬─────────┬──────┬──────╮
│ Benchmark │ Avg Memory │ Avg Allocs │ Throughput │ Avg duration │ Std-dev │ Min │ Max │
│ │ KiB/message │ allocs/message │ calls/s │ μs │ μs │ μs │ μs │
├──────────────────────────────────┼─────────────┼────────────────┼────────────┼──────────────┼─────────┼──────┼──────┤
│ workload_smol │ 2.95 │ 72.5 │ 18014 │ 56 │ 3.08 │ 50 │ 64 │
│ workload_32_224_224_uint8 │ 637.0 │ 79.7 │ 567 │ 1762 │ 99.07 │ 1628 │ 1911 │
│ workload_streaming_request │ 0.86 │ 6.5 │ 341851 │ 3 │ 1.68 │ 2 │ 30 │
│ workload_streaming_response │ 13.0 │ 27.7 │ 64515 │ 16 │ 5.12 │ 6 │ 33 │
│ workload_streaming_bidirectional │ 1.41 │ 25.6 │ 102072 │ 10 │ 6.23 │ 4 │ 52 │
╰──────────────────────────────────┴─────────────┴────────────────┴────────────┴──────────────┴─────────┴──────┴──────╯
```

### Stress Workloads
Expand Down
24 changes: 11 additions & 13 deletions src/Curl.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -138,21 +138,21 @@ end
function grpc_timeout_header_val(timeout::Real)
if round(Int, timeout) == timeout
timeout_secs = round(Int64, timeout)
return "$(timeout_secs)S"
return "$(string(timeout_secs))S"
end
timeout *= 1000
if round(Int, timeout) == timeout
timeout_millisecs = round(Int64, timeout)
return "$(timeout_millisecs)m"
return "$(string(timeout_millisecs))m"
end
timeout *= 1000
if round(Int, timeout) == timeout
timeout_microsecs = round(Int64, timeout)
return "$(timeout_microsecs)u"
return "$(string(timeout_microsecs))u"
end
timeout *= 1000
timeout_nanosecs = round(Int64, timeout)
return "$(timeout_nanosecs)n"
return "$(string(timeout_nanosecs))n"
end

mutable struct gRPCRequest
Expand Down Expand Up @@ -238,22 +238,19 @@ mutable struct gRPCRequest
# Uncomment this for debugging purposes
# curl_easy_setopt(easy_handle, CURLOPT_VERBOSE, UInt32(1))

http_url = replace(url, "grpc://" => "http://")
http_url = replace(http_url, "grpcs://" => "https://")

curl_easy_setopt(easy_handle, CURLOPT_URL, http_url)
curl_easy_setopt(easy_handle, CURLOPT_URL, url)
curl_easy_setopt(easy_handle, CURLOPT_TIMEOUT, deadline)
curl_easy_setopt(easy_handle, CURLOPT_PIPEWAIT, Clong(1))
curl_easy_setopt(easy_handle, CURLOPT_POST, Clong(1))
curl_easy_setopt(easy_handle, CURLOPT_CUSTOMREQUEST, "POST")

if startswith(http_url, "http://")
if startswith(url, "http://")
curl_easy_setopt(
easy_handle,
CURLOPT_HTTP_VERSION,
CURL_HTTP_VERSION_2_PRIOR_KNOWLEDGE,
)
elseif startswith(http_url, "https://")
elseif startswith(url, "https://")
curl_easy_setopt(easy_handle, CURLOPT_HTTP_VERSION, CURL_HTTP_VERSION_2TLS)
end

Expand All @@ -275,7 +272,7 @@ mutable struct gRPCRequest
easy_handle,
grpc.multi,
headers,
http_url,
url,
request,
0,
response,
Expand Down Expand Up @@ -797,8 +794,9 @@ function check_multi_info(grpc::gRPCCURL)
# The actual cleanup/notification happens here
cleanup_request(grpc, req)

# Remove from the list of requests associated
grpc.requests = filter(x -> x !== req, grpc.requests)
# Remove from the list of requests associated (in-place, no allocation)
idx = findfirst(x -> x === req, grpc.requests)
!isnothing(idx) && deleteat!(grpc.requests, idx)
else
@error("curl_multi_info_read: unknown message", message, maxlog = 1_000)
end
Expand Down
10 changes: 7 additions & 3 deletions src/gRPC.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,11 +88,15 @@ end

function url(client::gRPCServiceClient)
protocol = if client.secure
"grpcs"
"https"
else
"grpc"
"http"
end
"$protocol://$(client.host):$(client.port)$(client.path)"

# "$protocol://$(client.host):$(client.port)$(client.path)"
buffer = IOBuffer()
write(buffer, protocol, "://", client.host, ":", string(client.port), client.path)
String(take!(buffer))
end


Expand Down
14 changes: 5 additions & 9 deletions utils/gRPCClientUtils.jl/src/Benchmark.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,10 +15,8 @@ function perform_benchmark(f, N)

return [
f,
N_sample,
mem,
b.allocs,
round(timing, digits = 2),
1000*mem / N, # Avg Memory
round(b.allocs / N, digits = 1), # Avg Allocs
round(Int, N_sample/timing), # Throughput
round(Int, mean(timings_us) / N), # Avg duration
round(std(timings_us) / N, digits = 2),
Expand All @@ -33,17 +31,15 @@ function benchmark_table()
column_labels = [
[
"Benchmark",
"N",
"Memory",
"Allocations",
"Duration",
"Avg Memory",
"Avg Allocs",
"Throughput",
"Avg duration",
"Std-dev",
"Min",
"Max",
],
["", "calls", "MiB", "", "s", "calls/s", "μs", "μs", "μs", "μs"],
["", "KiB/message", "allocs/message", "calls/s", "μs", "μs", "μs", "μs"],
]
all_benchmarks = [
(workload_smol, 1_000),
Expand Down