Skip to content
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

Jump to bottom

[cherry-pick] Fix L0_memory_growth (#5795) #5796

Merged
merged 1 commit into from
May 16, 2023
Merged

[cherry-pick] Fix L0_memory_growth (#5795) #5796

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
41 changes: 27 additions & 14 deletions qa/L0_memory_growth/test.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,8 +92,12 @@ else
fi

# Threshold memory growth in MB
MAX_ALLOWED_ALLOC="150"
export MAX_ALLOWED_ALLOC
# NOTES:
# - Bounded memory growth tests typically show < 70 MB usage
# - Plan/ONNX is typically between 20-40 MB
# - Savedmodel is closer to 50-70 MB
# - Unbounded memory growth test typically shows > 100 MB usage
export MAX_ALLOWED_ALLOC="100"

# Create local model repository
mkdir -p models/
Expand Down Expand Up @@ -164,23 +168,24 @@ for MODEL in $(ls models); do
SECONDS=0
# Run the perf analyzer 'REPETITION' times
for ((i=1; i<=$REPETITION; i++)); do
# [TMA-621] Use --no-stability mode in perf analyzer when available
$PERF_ANALYZER -v -m $MODEL -i grpc --concurrency-range $CONCURRENCY -b $CLIENT_BS > $TEMP_CLIENT_LOG 2>&1
# Only record failure log for unexpected perf analyzer error
# [TMA-625] Currently check failure log as WAR, should check for specific
# code once perf analyzer returns different code for different error
if [ $? -ne 0 ] && [ `grep -c "^No valid requests recorded" $TEMP_CLIENT_LOG` == "0" ]; then
PA_RET=$?
# Success
if [ ${PA_RET} -eq 0 ]; then
continue
# Unstable measurement: OK for this test
elif [ ${PA_RET} -eq 2 ]; then
continue
# Other failures unexpected, report error
else
cat $TEMP_CLIENT_LOG >> $CLIENT_LOG
echo -e "\n***\n*** perf_analyzer for $MODEL failed on iteration $i\n***" >> $CLIENT_LOG
TEMP_RET=1
RET=1
fi
done
TEST_DURATION=$SECONDS

if [ $TEMP_RET -ne 0 ]; then
cat $CLIENT_LOG
RET=1
fi

set -e

# Stop Server
Expand All @@ -200,9 +205,11 @@ for MODEL in $(ls models); do
python $MASSIF_TEST $MASSIF_LOG $MAX_ALLOWED_ALLOC --start-from-middle >> $CLIENT_LOG 2>&1
if [ $? -ne 0 ]; then
cat $CLIENT_LOG
echo -e "\n***\n*** Test for $MODEL Failed\n***"
echo -e "\n***\n*** Test for $MODEL Failed.\n***"
RET=1
fi
# Always output memory usage for easier triage of MAX_ALLOWED_ALLOC settings in the future
grep -i "Change in memory allocation" "${CLIENT_LOG}" || true
set -e
done

Expand Down Expand Up @@ -272,11 +279,17 @@ if [ $SKIP_BUSYOP -ne 1 ]; then
cat ${GRAPH_LOG}
# Check the massif output
python $MASSIF_TEST $MASSIF_LOG $MAX_ALLOWED_ALLOC --start-from-middle >> $CLIENT_LOG 2>&1
# This busyop test is expected to return a non-zero error since it is
# intentionally testing unbounded growth. If it returns success for some
# reason, raise error.
if [ $? -ne 1 ]; then
cat $CLIENT_LOG
echo -e "\n***\n*** Test for graphdef_busyop Failed\n***"
echo -e "\n***\n*** Massif test for graphdef_busyop Failed\n***"
echo -e "\n***\n*** Expected unbounded growth, but found acceptable growth within ${MAX_ALLOWED_ALLOC} MB\n***"
RET=1
fi
# Always output memory usage for easier triage of MAX_ALLOWED_ALLOC settings in the future
grep -i "Change in memory allocation" "${CLIENT_LOG}" || true
fi
set -e

Expand Down