Skip to content

Commit

Permalink
add tests for prediction API
Browse files Browse the repository at this point in the history
  • Loading branch information
cmelone committed Mar 7, 2024
1 parent 31a0aa1 commit 3c8bc2c
Show file tree
Hide file tree
Showing 3 changed files with 181 additions and 0 deletions.
64 changes: 64 additions & 0 deletions gantry/tests/defs/prediction.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,64 @@
# flake8: noqa
# fmt: off

NORMAL_BUILD = {
"hash": "testing",
"package": {
"name": "py-torch",
"version": "2.2.1",
"variants": "~caffe2+cuda+cudnn~debug+distributed+fbgemm+gloo+kineto~magma~metal+mkldnn+mpi~nccl+nnpack+numa+numpy+onnx_ml+openmp+qnnpack~rocm+tensorpipe~test+valgrind+xnnpack build_system=python_pip cuda_arch=80",
},
"compiler": {
"name": "gcc",
"version": "11.4.0",
},
}

# everything in NORMAL_BUILD["package"]["variants"] except removing build_system=python_pip
# in order to test the expensive variants filter
EXPENSIVE_VARIANT_BUILD = {
"hash": "testing",
"package": {
"name": "py-torch",
"version": "2.2.1",
"variants": "~caffe2+cuda+cudnn~debug+distributed+fbgemm+gloo+kineto~magma~metal+mkldnn+mpi~nccl+nnpack+numa+numpy+onnx_ml+openmp+qnnpack~rocm+tensorpipe~test+valgrind+xnnpack cuda_arch=80",
},
"compiler": {
"name": "gcc",
"version": "11.4.0",
},
}

# no variants should match this, so we expect the default prediction
BAD_VARIANT_BUILD = {
"hash": "testing",
"package": {
"name": "py-torch",
"version": "2.2.1",
"variants": "+no~expensive~variants+match",
},
"compiler": {
"name": "gcc",
"version": "11.4.0",
},
}

# calculated by running the baseline prediction algorithm on the sample data in gantry/tests/sql/insert_prediction.sql
NORMAL_PREDICTION = {
"hash": "testing",
"variables": {
"KUBERNETES_CPU_REQUEST": "12",
"KUBERNETES_MEMORY_REQUEST": "9576M",
},
}


# this is what will get returned when there are no samples in the database
# that match what the client wants
DEFAULT_PREDICTION = {
"hash": "testing",
"variables": {
"KUBERNETES_CPU_REQUEST": "1",
"KUBERNETES_MEMORY_REQUEST": "2000M",
},
}
6 changes: 6 additions & 0 deletions gantry/tests/sql/insert_samples.sql
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
INSERT INTO nodes VALUES(6789,'ec2c47a0-7e9b-cfa3-9ad4-ac227ade598d','ip-192-168-202-150.ec2.internal',32.0,131072000000.0,'amd64','linux','m5.8xlarge');
INSERT INTO jobs VALUES(6781,'runner-2j2ndhxu-project-2-concurrent-0-nbogpypi1',6789,1708919572.983000041,1708924744.811000108,101502092,'success','develop','py-torch','2.2.1','{"caffe2": false, "cuda": true, "cudnn": true, "debug": false, "distributed": true, "fbgemm": true, "gloo": true, "kineto": true, "magma": false, "metal": false, "mkldnn": true, "mpi": true, "nccl": false, "nnpack": true, "numa": true, "numpy": true, "onnx_ml": true, "openmp": true, "qnnpack": true, "rocm": false, "tensorpipe": true, "test": false, "valgrind": true, "xnnpack": true, "build_system": "python_pip", "cuda_arch": "80"}','gcc','11.4.0','linux-ubuntu20.04-x86_64_v3','e4s',12,12.0,NULL,9.77948152336477605,11.98751586519425772,12.00060520666194109,0.3736576704015182604,3.811106184376615414,48000000000.0,64000000000.0,9652098890.24199867,7399608320.0,41186873344.0,85508096.0,8707419891.779100419);
INSERT INTO jobs VALUES(6782,'runner-2j2ndhxu-project-2-concurrent-0-nbogpypi2',6789,1708919572.983000041,1708924744.811000108,101502093,'success','develop','py-torch','2.2.1','{"caffe2": false, "cuda": true, "cudnn": true, "debug": false, "distributed": true, "fbgemm": true, "gloo": true, "kineto": true, "magma": false, "metal": false, "mkldnn": true, "mpi": true, "nccl": false, "nnpack": true, "numa": true, "numpy": true, "onnx_ml": true, "openmp": true, "qnnpack": true, "rocm": false, "tensorpipe": true, "test": false, "valgrind": true, "xnnpack": true, "build_system": "python_pip", "cuda_arch": "80"}','gcc','11.4.0','linux-ubuntu20.04-x86_64_v3','e4s',12,12.0,NULL,10.77948152336477605,11.98751586519425772,12.00060520666194109,0.3736576704015182604,3.811106184376615414,48000000000.0,64000000000.0,9958098890.24199867,7399608320.0,41186873344.0,85508096.0,8707419891.779100419);
INSERT INTO jobs VALUES(6783,'runner-2j2ndhxu-project-2-concurrent-0-nbogpypi3',6789,1708919572.983000041,1708924744.811000108,101502094,'success','develop','py-torch','2.2.1','{"caffe2": false, "cuda": true, "cudnn": true, "debug": false, "distributed": true, "fbgemm": true, "gloo": true, "kineto": true, "magma": false, "metal": false, "mkldnn": true, "mpi": true, "nccl": false, "nnpack": true, "numa": true, "numpy": true, "onnx_ml": true, "openmp": true, "qnnpack": true, "rocm": false, "tensorpipe": true, "test": false, "valgrind": true, "xnnpack": true, "build_system": "python_pip", "cuda_arch": "80"}','gcc','11.4.0','linux-ubuntu20.04-x86_64_v3','e4s',12,12.0,NULL,11.77948152336477605,11.98751586519425772,12.00060520666194109,0.3736576704015182604,3.811106184376615414,48000000000.0,64000000000.0,9158098890.24199867,7399608320.0,41186873344.0,85508096.0,8707419891.779100419);
INSERT INTO jobs VALUES(6784,'runner-2j2ndhxu-project-2-concurrent-0-nbogpypi4',6789,1708919572.983000041,1708924744.811000108,101502095,'success','develop','py-torch','2.2.1','{"caffe2": false, "cuda": true, "cudnn": true, "debug": false, "distributed": true, "fbgemm": true, "gloo": true, "kineto": true, "magma": false, "metal": false, "mkldnn": true, "mpi": true, "nccl": false, "nnpack": true, "numa": true, "numpy": true, "onnx_ml": true, "openmp": true, "qnnpack": true, "rocm": false, "tensorpipe": true, "test": false, "valgrind": true, "xnnpack": true, "build_system": "python_pip", "cuda_arch": "80"}','gcc','11.4.0','linux-ubuntu20.04-x86_64_v3','e4s',12,12.0,NULL,12.77948152336477605,11.98751586519425772,12.00060520666194109,0.3736576704015182604,3.811106184376615414,48000000000.0,64000000000.0,9758098890.24199867,7399608320.0,41186873344.0,85508096.0,8707419891.779100419);
INSERT INTO jobs VALUES(6785,'runner-2j2ndhxu-project-2-concurrent-0-nbogpypi5',6789,1708919572.983000041,1708924744.811000108,101502096,'success','develop','py-torch','2.2.1','{"caffe2": false, "cuda": true, "cudnn": true, "debug": false, "distributed": true, "fbgemm": true, "gloo": true, "kineto": true, "magma": false, "metal": false, "mkldnn": true, "mpi": true, "nccl": false, "nnpack": true, "numa": true, "numpy": true, "onnx_ml": true, "openmp": true, "qnnpack": true, "rocm": false, "tensorpipe": true, "test": false, "valgrind": true, "xnnpack": true, "build_system": "python_pip", "cuda_arch": "80"}','gcc','11.4.0','linux-ubuntu20.04-x86_64_v3','e4s',12,12.0,NULL,13.77948152336477605,11.98751586519425772,12.00060520666194109,0.3736576704015182604,3.811106184376615414,48000000000.0,64000000000.0,9358098890.24199867,7399608320.0,41186873344.0,85508096.0,8707419891.779100419);
111 changes: 111 additions & 0 deletions gantry/tests/test_prediction.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,111 @@
import pytest

from gantry.routes.prediction import prediction
from gantry.tests.defs import prediction as defs
from gantry.util.prediction import validate_payload


@pytest.fixture
async def db_conn_inserted(db_conn):
"""Returns a connection to a database with 5 samples inserted"""

with open("gantry/tests/sql/insert_samples.sql") as f:
await db_conn.executescript(f.read())

return db_conn


async def test_exact_match(db_conn_inserted):
"""All fields are an exact match for 5 samples in the database."""

assert (
await prediction.predict_single(db_conn_inserted, defs.NORMAL_BUILD)
== defs.NORMAL_PREDICTION
)


async def test_expensive_variants(db_conn_inserted):
"""
Tests whether the algorithm filters by expensive variants.
The input has been modified to prevent an exact match with
any of the samples.
"""

assert (
await prediction.predict_single(db_conn_inserted, defs.EXPENSIVE_VARIANT_BUILD)
== defs.NORMAL_PREDICTION
)


async def test_no_variant_match(db_conn_inserted):
"""
All fields match except for variants, expect default predictions with no sample.
"""

assert (
await prediction.predict_single(db_conn_inserted, defs.BAD_VARIANT_BUILD)
== defs.DEFAULT_PREDICTION
)


async def test_partial_match(db_conn_inserted):
"""
Some of the fields match, so the prediction should be based on matching
with other fields. In reality, we're using the same dataset but just
testing that the prediction will be the same with a different compiler name.
"""

# same as NORMAL_BUILD, but with a different compiler name to test partial matching
diff_compiler_build = defs.NORMAL_BUILD.copy()
diff_compiler_build["compiler"]["name"] = "gcc-different"

assert (
await prediction.predict_single(db_conn_inserted, diff_compiler_build)
== defs.NORMAL_PREDICTION
)


async def test_empty_sample(db_conn):
"""No samples in the database, so we expect default predictions."""

assert (
await prediction.predict_single(db_conn, defs.NORMAL_BUILD)
== defs.DEFAULT_PREDICTION
)


# Test validate_payload


def test_valid_payload():
"""Tests that a valid payload returns True"""
assert validate_payload(defs.NORMAL_BUILD) is True


def test_invalid_payloads():
"""Test a series of invalid payloads"""

# non dict
assert validate_payload("hi") is False

build = defs.NORMAL_BUILD.copy()
# missing package
del build["package"]
assert validate_payload(build) is False

build = defs.NORMAL_BUILD.copy()
# missing compiler
del build["compiler"]
assert validate_payload(build) is False

# name and version are strings in the package and compiler
for key in ["name", "version"]:
for field in ["package", "compiler"]:
build = defs.NORMAL_BUILD.copy()
build[field][key] = 123
assert validate_payload(build) is False

# invalid variants
build = defs.NORMAL_BUILD.copy()
build["package"]["variants"] = "+++++"
assert validate_payload(build) is False

0 comments on commit 3c8bc2c

Please sign in to comment.