Add memory profile utils (#77)

Author: csvance

.gitignore

@@ -4,4 +4,6 @@ __pycache__
 *.pyc
 build
 Manifest.toml
-grpc_test_server
+grpc_test_server
+*.gz 
+*.vscode

utils/gRPCClientUtils.jl/Project.toml

@@ -5,15 +5,21 @@ authors = ["Carroll Vance <cs.vance@icloud.com>"]
 
 [deps]
 BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
-gRPCClient = "aaca4a50-36af-4a1d-b878-4c443f2061ad"
+PProf = "e4faabce-9ead-11e9-39d9-4379958e3056"
 PrettyTables = "08abe8d2-0d0c-5749-adfa-8a2ac140af0d"
+Profile = "9abbd945-dff8-562f-b5e8-e1ebf5ef1b79"
 ProgressBars = "49802e3a-d2f1-5c88-81d8-b72133a6f568"
 ProtoBuf = "3349acd9-ac6a-5e09-bcdb-63829b23a429"
+gRPCClient = "aaca4a50-36af-4a1d-b878-4c443f2061ad"
+
+[sources]
+gRPCClient = {path = "../.."}
 
 [compat]
 BenchmarkTools = "~1.6.0"
-gRPCClient = "~1.0.0"
+PProf = "3.2.0"
 PrettyTables = "~3.1.0"
+Profile = "1.11.0"
 ProgressBars = "~1.5.0"
 ProtoBuf = "~1.2"
-
+gRPCClient = "~1.0.0"

utils/gRPCClientUtils.jl/src/Benchmark.jl

@@ -1,12 +1,14 @@
-benchmark_workload_smol() = @benchmark workload_smol(1_000)
-benchmark_workload_32_224_224_uint8() = @benchmark workload_32_224_224_uint8(100)
-benchmark_workload_streaming_request() = @benchmark workload_streaming_request(1_000)
-benchmark_workload_streaming_response() = @benchmark workload_streaming_response(1_000)
-benchmark_workload_streaming_bidirectional() =
-    @benchmark workload_streaming_bidirectional(1_000)
+benchmark_workload_smol() = @benchmark workload_smol()
+benchmark_workload_32_224_224_uint8() = @benchmark workload_32_224_224_uint8()
+benchmark_workload_streaming_request() = @benchmark workload_streaming_request()
+benchmark_workload_streaming_response() = @benchmark workload_streaming_response()
+benchmark_workload_streaming_bidirectional() = @benchmark workload_streaming_bidirectional()
 
 
-function perform_benchmark(f, N)
+function perform_benchmark(f)
+    # Hack to make the code cleaner (benchmarks return number of messages per workload)
+    N = f()
+
     b = @benchmark $f($N)
     timing = sum(b.times) / 1e9
     timings_us = b.times ./ 1e3
@@ -42,14 +44,14 @@ function benchmark_table()
         ["", "KiB/message", "allocs/message", "messages/s", "μs", "μs", "μs", "μs"],
     ]
     all_benchmarks = [
-        (workload_smol, 1_000),
-        (workload_32_224_224_uint8, 100),
-        (workload_streaming_request, 1_000),
-        (workload_streaming_response, 1_000),
-        (workload_streaming_bidirectional, 1_000),
+        workload_smol,
+        workload_32_224_224_uint8,
+        workload_streaming_request,
+        workload_streaming_response,
+        workload_streaming_bidirectional,
     ]
 
-    all_results = [perform_benchmark(f, N) for (f, N) in ProgressBar(all_benchmarks)]
+    all_results = [perform_benchmark(f) for f in ProgressBar(all_benchmarks)]
 
     pretty_table(
         permutedims(reduce(hcat, all_results));

utils/gRPCClientUtils.jl/src/Profile.jl

@@ -0,0 +1,18 @@
+function profile_memory_fn(f::Function)
+    grpc_init()
+
+    # Warmup
+    _ = f()
+
+    Profile.Allocs.clear()
+    Profile.Allocs.@profile sample_rate=0.1 f()
+    PProf.Allocs.pprof()
+end
+
+profile_memory_workload_smol() = profile_memory_fn(workload_smol)
+profile_memory_workload_32_224_224_uint8() = profile_memory_fn(workload_32_224_224_uint8)
+profile_memory_workload_streaming_request() = profile_memory_fn(workload_streaming_request)
+profile_memory_workload_streaming_response() =
+    profile_memory_fn(workload_streaming_response)
+profile_memory_workload_streaming_bidirectional() =
+    profile_memory_fn(workload_streaming_bidirectional)

utils/gRPCClientUtils.jl/src/Stress.jl

@@ -1,12 +1,12 @@
-function stress_workload(f::Function, n)
+function stress_workload(f::Function)
     while true
-        f(n)
+        f()
     end
 end
 
-stress_workload_smol() = stress_workload(workload_smol, 1_000)
-stress_workload_32_224_224_uint8() = stress_workload(workload_32_224_224_uint8, 100)
-stress_workload_streaming_request() = stress_workload(workload_streaming_request, 1_000)
-stress_workload_streaming_response() = stress_workload(workload_streaming_response, 1_000)
+stress_workload_smol() = stress_workload(workload_smol)
+stress_workload_32_224_224_uint8() = stress_workload(workload_32_224_224_uint8)
+stress_workload_streaming_request() = stress_workload(workload_streaming_request)
+stress_workload_streaming_response() = stress_workload(workload_streaming_response)
 stress_workload_streaming_bidirectional() =
-    stress_workload(workload_streaming_bidirectional, 1_000)
+    stress_workload(workload_streaming_bidirectional)

utils/gRPCClientUtils.jl/src/Workloads.jl

@@ -1,4 +1,4 @@
-function workload_32_224_224_uint8(n)
+function workload_32_224_224_uint8(n = 100)
     client = TestService_TestRPC_Client("localhost", 8001)
 
     reqs = Vector{gRPCRequest}()
@@ -14,9 +14,11 @@ function workload_32_224_224_uint8(n)
     for req in reqs
         grpc_async_await(req)
     end
+
+    n
 end
 
-function workload_smol(n)
+function workload_smol(n = 1_000)
     client = TestService_TestRPC_Client("localhost", 8001)
 
     # Since requests are lightweight, use async / await pattern to avoid creating an extra task per request
@@ -29,9 +31,11 @@ function workload_smol(n)
     for req in reqs
         grpc_async_await(req)
     end
+
+    n
 end
 
-function workload_streaming_request(n)
+function workload_streaming_request(n = 1_000)
     client = TestService_TestClientStreamRPC_Client("localhost", 8001)
     requests_c = Channel{TestRequest}(16)
 
@@ -47,10 +51,10 @@ function workload_streaming_request(n)
         response = grpc_async_await(req)
     end
 
-    nothing
+    n
 end
 
-function workload_streaming_response(n)
+function workload_streaming_response(n = 1_000)
     client = TestService_TestServerStreamRPC_Client("localhost", 8001)
     response_c = Channel{TestResponse}(16)
 
@@ -61,11 +65,11 @@ function workload_streaming_response(n)
     end
     close(response_c)
 
-    nothing
+    n
 end
 
 
-function workload_streaming_bidirectional(n)
+function workload_streaming_bidirectional(n = 1_000)
     client = TestService_TestBidirectionalStreamRPC_Client("localhost", 8001)
     requests_c = Channel{TestRequest}(16)
     response_c = Channel{TestResponse}(16)
@@ -91,4 +95,6 @@ function workload_streaming_bidirectional(n)
 
         nothing
     end
+
+    n
 end

utils/gRPCClientUtils.jl/src/gRPCClientUtils.jl

@@ -5,11 +5,14 @@ using BenchmarkTools
 using PrettyTables
 using ProgressBars
 using ProtoBuf
+using Profile
+using PProf
 
 include("../../../test/gen/test/test_pb.jl")
 include("Workloads.jl")
 include("Benchmark.jl")
 include("Stress.jl")
+include("Profile.jl")
 
 export benchmark_table
 
@@ -25,5 +28,11 @@ export stress_workload_streaming_request
 export stress_workload_streaming_response
 export stress_workload_streaming_bidirectional
 
+export profile_memory_workload_smol
+export profile_memory_workload_32_224_224_uint8
+export profile_memory_workload_streaming_request
+export profile_memory_workload_streaming_response
+export profile_memory_workload_streaming_bidirectional
+
 
 end # module gRPCClientUtils