Files

136 lines
5.6 KiB
Bash

#!/bin/sh
set -eu
data_dir=/var/lib/supermemory
sentinel="$data_dir/.stability-gemini-paid-v1.ok"
base_url="${SUPERMEMORY_URL:-http://127.0.0.1:6767}"
container_tag="${SUPERMEMORY_CONTAINER_TAG:-supermemory_stability_paid_v1}"
if [ -f "$sentinel" ]; then
printf '%s\n' '{"stability":"already-passed","suite":"gemini-paid-v1"}'
exit 0
fi
api_key=$(cat "$data_dir/api-key")
auth_header="Authorization: Bearer $api_key"
run_id="paid-v1-$(date +%s)"
submitted=0
add_failures=0
processing_failures=0
documents_failures=0
hybrid_failures=0
metadata_failures=0
case_values() {
case "$1" in
1) content="海棠计划的下一步是核对原始报告并记录出处。"; query="海棠计划下一步是什么?"; expected="核对原始报告" ;;
2) content="Project Juniper rollback owner is Mira Chen."; query="Who owns the Juniper rollback?"; expected="Mira Chen" ;;
3) content="Orion 项目的 maintenance window is Tuesday 03:00 UTC."; query="Orion 的维护窗口是什么时候?"; expected="Tuesday" ;;
4) content="The synthetic 2026 Q4 budget cap is 42000 credits."; query="What is the synthetic Q4 budget cap?"; expected="42000" ;;
5) content="仅在本基准中,状态报告应使用 amber labels。"; query="本基准的状态报告应该用什么标签?"; expected="amber" ;;
esac
}
for case_id in 1 2 3 4 5; do
case_values "$case_id"
marker="supermemory-stability-$run_id-$case_id"
if curl --fail --silent --show-error --max-time 60 \
-H "$auth_header" \
-H 'Content-Type: application/json' \
--data "{\"content\":\"$marker. $content\",\"containerTag\":\"$container_tag\",\"customId\":\"$marker\",\"metadata\":{\"source\":\"poc-stability\",\"suite\":\"gemini-paid-v1\",\"case_id\":\"$case_id\",\"status\":\"temporary\"}}" \
"$base_url/v3/documents" >"/tmp/add-$case_id.json"; then
document_id=$(sed -n 's/.*"id"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' "/tmp/add-$case_id.json" | head -n 1)
if [ -n "$document_id" ]; then
printf '%s\n' "$document_id" >"/tmp/id-$case_id"
submitted=$((submitted + 1))
else
add_failures=$((add_failures + 1))
fi
else
add_failures=$((add_failures + 1))
fi
done
for _ in $(seq 1 300); do
terminal=0
for case_id in 1 2 3 4 5; do
[ -f "/tmp/id-$case_id" ] || continue
if [ -f "/tmp/done-$case_id" ] || [ -f "/tmp/failed-$case_id" ]; then
terminal=$((terminal + 1))
continue
fi
document_id=$(cat "/tmp/id-$case_id")
if curl --fail --silent --show-error --max-time 30 \
-H "$auth_header" \
"$base_url/v3/documents/$document_id" >"/tmp/document-$case_id.json"; then
if grep -Eq '"status"[[:space:]]*:[[:space:]]*"done"' "/tmp/document-$case_id.json"; then
: >"/tmp/done-$case_id"
terminal=$((terminal + 1))
elif grep -Eq '"status"[[:space:]]*:[[:space:]]*"failed"' "/tmp/document-$case_id.json"; then
: >"/tmp/failed-$case_id"
terminal=$((terminal + 1))
processing_failures=$((processing_failures + 1))
fi
fi
done
[ "$terminal" -eq "$submitted" ] && break
sleep 2
done
for case_id in 1 2 3 4 5; do
[ -f "/tmp/id-$case_id" ] || continue
if [ ! -f "/tmp/done-$case_id" ]; then
if [ ! -f "/tmp/failed-$case_id" ]; then
processing_failures=$((processing_failures + 1))
fi
continue
fi
case_values "$case_id"
marker="supermemory-stability-$run_id-$case_id"
if curl --fail --silent --show-error --max-time 30 \
-H "$auth_header" \
-H 'Content-Type: application/json' \
--data "{\"q\":\"$marker\",\"containerTag\":\"$container_tag\",\"searchMode\":\"documents\",\"limit\":5}" \
"$base_url/v4/search" >"/tmp/documents-$case_id.json"; then
if ! grep -q "$marker" "/tmp/documents-$case_id.json"; then
documents_failures=$((documents_failures + 1))
fi
else
documents_failures=$((documents_failures + 1))
fi
if [ -f "/tmp/documents-$case_id.json" ] && ! grep -q 'poc-stability' "/tmp/documents-$case_id.json"; then
metadata_failures=$((metadata_failures + 1))
fi
if curl --fail --silent --show-error --max-time 30 \
-H "$auth_header" \
-H 'Content-Type: application/json' \
--data "{\"q\":\"$query\",\"containerTag\":\"$container_tag\",\"searchMode\":\"hybrid\",\"limit\":5}" \
"$base_url/v4/search" >"/tmp/hybrid-$case_id.json"; then
if ! grep -q "$expected" "/tmp/hybrid-$case_id.json"; then
hybrid_failures=$((hybrid_failures + 1))
fi
else
hybrid_failures=$((hybrid_failures + 1))
fi
done
conversation_http=$(curl --silent --show-error --max-time 60 \
--output /tmp/stability-conversation.json \
--write-out '%{http_code}' \
-H "$auth_header" \
-H 'Content-Type: application/json' \
--data "{\"conversationId\":\"supermemory-stability-$run_id-conversation\",\"containerTags\":[\"$container_tag\"],\"messages\":[{\"role\":\"user\",\"content\":\"Synthetic session question.\"},{\"role\":\"assistant\",\"content\":\"Synthetic session answer.\"}],\"metadata\":{\"source\":\"poc-stability\",\"status\":\"temporary\"}}" \
"$base_url/v4/conversations") || conversation_http=transport-error
printf '{"stability":"result","suite":"gemini-paid-v1","attempted":5,"submitted":%s,"add_failures":%s,"processing_failures":%s,"documents_failures":%s,"hybrid_failures":%s,"metadata_failures":%s,"conversation_http":"%s"}\n' \
"$submitted" "$add_failures" "$processing_failures" "$documents_failures" "$hybrid_failures" "$metadata_failures" "$conversation_http"
if [ "$submitted" -eq 5 ] && [ "$add_failures" -eq 0 ] && [ "$processing_failures" -eq 0 ] && \
[ "$documents_failures" -eq 0 ] && [ "$hybrid_failures" -eq 0 ] && [ "$metadata_failures" -eq 0 ]; then
touch "$sentinel"
exit 0
fi
exit 70