Add CPU stats into monitoring (#244)

tellet-q · web-flow · commit ce8fe7130295 · 2025-04-29T14:46:26.000+02:00
* Add comment line

* Store cpu and telemetry cpu values for default bench
diff --git a/.github/workflows/manual-compare-versions-benchmark.yaml b/.github/workflows/manual-compare-versions-benchmark.yaml
@@ -167,6 +167,7 @@ jobs:
           res_p99_time=$(compare "p99_time" "${{ needs.runBenchmarkForVersion1.outputs.p99_time }}" "${{ needs.runBenchmarkForVersion2.outputs.p99_time }}")
           res_vm_rss_memory_usage=$(compare "vm_rss_memory_usage" "${{ needs.runBenchmarkForVersion1.outputs.vm_rss_memory_usage }}" "${{ needs.runBenchmarkForVersion2.outputs.vm_rss_memory_usage }}")
           res_rss_anon_memory_usage=$(compare "rss_anon_memory_usage" "${{ needs.runBenchmarkForVersion1.outputs.rss_anon_memory_usage }}" "${{ needs.runBenchmarkForVersion2.outputs.rss_anon_memory_usage }}")
+          res_cpu_usage=$(compare "cpu_usage" "${{ needs.runBenchmarkForVersion1.outputs.cpu }}" "${{ needs.runBenchmarkForVersion2.outputs.cpu }}")
           res_upload_time=$(compare "upload_time" "${{ needs.runBenchmarkForVersion1.outputs.upload_time }}" "${{ needs.runBenchmarkForVersion2.outputs.upload_time }}")
           res_indexing_time=$(compare "indexing_time" "${{ needs.runBenchmarkForVersion1.outputs.indexing_time }}" "${{ needs.runBenchmarkForVersion2.outputs.indexing_time }}")
 
@@ -179,6 +180,7 @@ jobs:
           echo "| p95_time              | ${{ needs.runBenchmarkForVersion1.outputs.p95_time }} | ${{ needs.runBenchmarkForVersion2.outputs.p95_time }} | ${res_p95_time} |" >> $GITHUB_STEP_SUMMARY
           echo "| p99_time              | ${{ needs.runBenchmarkForVersion1.outputs.p99_time }} | ${{ needs.runBenchmarkForVersion2.outputs.p99_time }} | ${res_p99_time} |" >> $GITHUB_STEP_SUMMARY
           echo "| vm_rss_memory_usage   | ${{ needs.runBenchmarkForVersion1.outputs.vm_rss_memory_usage }} | ${{ needs.runBenchmarkForVersion2.outputs.vm_rss_memory_usage }} | ${res_vm_rss_memory_usage} |" >> $GITHUB_STEP_SUMMARY
-          echo "| rss_anon_memory_usage  | ${{ needs.runBenchmarkForVersion1.outputs.rss_anon_memory_usage }} | ${{ needs.runBenchmarkForVersion2.outputs.rss_anon_memory_usage }} | ${res_rss_anon_memory_usage} |" >> $GITHUB_STEP_SUMMARY
+          echo "| rss_anon_memory_usage | ${{ needs.runBenchmarkForVersion1.outputs.rss_anon_memory_usage }} | ${{ needs.runBenchmarkForVersion2.outputs.rss_anon_memory_usage }} | ${res_rss_anon_memory_usage} |" >> $GITHUB_STEP_SUMMARY
+          echo "| cpu                   | ${{ needs.runBenchmarkForVersion1.outputs.cpu }} | ${{ needs.runBenchmarkForVersion2.outputs.cpu }} | ${res_cpu_usage} |" >> $GITHUB_STEP_SUMMARY
           echo "| upload_time           | ${{ needs.runBenchmarkForVersion1.outputs.upload_time }} | ${{ needs.runBenchmarkForVersion2.outputs.upload_time }} | ${res_upload_time} |" >> $GITHUB_STEP_SUMMARY
           echo "| indexing_time         | ${{ needs.runBenchmarkForVersion1.outputs.indexing_time }} | ${{ needs.runBenchmarkForVersion2.outputs.indexing_time }} | ${res_indexing_time} |" >> $GITHUB_STEP_SUMMARY
diff --git a/tools/qdrant_collect_cpu_usage.sh b/tools/qdrant_collect_cpu_usage.sh
@@ -0,0 +1,40 @@
+#!/bin/bash
+
+PS4='ts=$(date "+%Y-%m-%dT%H:%M:%SZ") level=DEBUG line=$LINENO file=$BASH_SOURCE '
+set -euo pipefail
+
+# Examples: start or end
+MODE=$1
+
+CLOUD_NAME=${CLOUD_NAME:-"hetzner"}
+SERVER_USERNAME=${SERVER_USERNAME:-"root"}
+
+SCRIPT=$(realpath "$0")
+SCRIPT_PATH=$(dirname "$SCRIPT")
+
+BENCH_SERVER_NAME=${SERVER_NAME:-"benchmark-server-1"}
+
+IP_OF_THE_SERVER=$(bash "${SCRIPT_PATH}/${CLOUD_NAME}/get_public_ip.sh" "$BENCH_SERVER_NAME")
+
+UTIME=$(ssh -tt -o ServerAliveInterval=10 -o ServerAliveCountMax=10 "${SERVER_USERNAME}@${IP_OF_THE_SERVER}" "cat /proc/\$(pidof qdrant)/stat | awk '{print \$14}'")
+# Clean up any whitespace characters
+UTIME=$(echo "$UTIME" | tr -d '[:space:]')
+
+CURRENT_DATE=$(date +%Y-%m-%d-%H-%M-%S)
+
+mkdir -p results/cpu
+
+if [[ "$MODE" == "end" ]]; then
+  echo "Calculate CPU usage (seconds) over period of time"
+  UTIME_FILE=$(ls -t results/cpu/utime-*.txt | head -n 1)
+  UTIME_START=$(cat "$UTIME_FILE" | tr -d '[:space:]')
+  echo "$UTIME" >> "${UTIME_FILE}"
+  CPU=$(echo "scale=2; ($UTIME - $UTIME_START) / 100" | bc)
+  echo "$CPU" > "./results/cpu/cpu-usage-${CURRENT_DATE}.txt"
+elif [[ "$MODE" == "start" ]]; then
+  echo "Store utime start value in ./results/cpu/utime-${CURRENT_DATE}.txt"
+  echo "$UTIME" > "./results/cpu/utime-${CURRENT_DATE}.txt"
+else
+  echo "Unknown mode: $MODE"
+  exit 1
+fi
diff --git a/tools/run_ci.sh b/tools/run_ci.sh
@@ -24,7 +24,8 @@ trap 'handle_term' TERM
 
 # Script, that runs benchmark within the GitHub Actions CI environment
 
-BENCHMARK_STRATEGY=${BENCHMARK_STRATEGY:-"default"}
+# Possible values for BENCHMARK_STRATEGY: default, tenants, parallel and collection-reload
+export BENCHMARK_STRATEGY=${BENCHMARK_STRATEGY:-"default"}
 
 SCRIPT=$(realpath "$0")
 SCRIPT_PATH=$(dirname "$SCRIPT")
@@ -37,9 +38,14 @@ if [[ "$BENCHMARK_STRATEGY" == "collection-reload" ]]; then
   export TELEMETRY_API_RESPONSE_FILE=$(ls -t results/telemetry-api-*.json | head -n 1)
 else
   # any other strategies are considered to have search & upload results
+  export TELEMETRY_API_RESPONSE_FILE=$(ls -t results/telemetry-api-*.json | head -n 1)
   export SEARCH_RESULTS_FILE=$(find results/ -maxdepth 1 -type f -name '*-search-*.json' -printf '%T@ %p\n' | sort -nr | head -n 1 | cut -d' ' -f2-)
   export UPLOAD_RESULTS_FILE=$(find results/ -maxdepth 1 -type f -name '*-upload-*.json' -printf '%T@ %p\n' | sort -nr | head -n 1 | cut -d' ' -f2-)
 
+  if [[ "$BENCHMARK_STRATEGY" == "default" ]]; then
+    export CPU_USAGE_FILE=$(ls -t results/cpu/cpu-usage-*.txt | head -n 1)
+  fi
+
   if [[ "$BENCHMARK_STRATEGY" == "parallel" ]]; then
     export PARALLEL_UPLOAD_RESULTS_FILE=$(ls -t results/parallel/*-upload-*.json | head -n 1)
     export PARALLEL_SEARCH_RESULTS_FILE=$(ls -t results/parallel/*-search-*.json | head -n 1)
diff --git a/tools/run_remote_benchmark.sh b/tools/run_remote_benchmark.sh
@@ -44,8 +44,12 @@ case "$BENCHMARK_STRATEGY" in
 
   bash -x "${SCRIPT_PATH}/run_server_container.sh" "$SERVER_CONTAINER_NAME"
 
+  bash -x "${SCRIPT_PATH}/qdrant_collect_cpu_usage.sh" "start"
+
   bash -x "${SCRIPT_PATH}/run_client_script.sh"
 
+  bash -x "${SCRIPT_PATH}/qdrant_collect_cpu_usage.sh" "end"
+
   bash -x "${SCRIPT_PATH}/qdrant_collect_stats.sh" "$SERVER_CONTAINER_NAME"
   ;;
   "tenants")
diff --git a/tools/upload_parallel_results_postgres.sh b/tools/upload_parallel_results_postgres.sh
@@ -19,6 +19,8 @@
 # 	p99_time real,
 # 	search_time real,
 # 	no_upsert_search_time real,
+# 	cpu real,
+# 	cpu_telemetry real,
 # );
 
 PARALLEL_SEARCH_RESULTS_FILE=${PARALLEL_SEARCH_RESULTS_FILE:-""}
@@ -63,6 +65,13 @@ if [[ -z "$ROOT_API_RESPONSE_FILE" ]]; then
   exit 1
 fi
 
+if [[ "$BENCHMARK_STRATEGY" == "default" ]]; then
+  if [[ -z "$CPU_USAGE_FILE" ]]; then
+    echo "CPU_USAGE_FILE is not set"
+    exit 1
+  fi
+fi
+
 RPS=NULL
 MEAN_PRECISIONS=NULL
 P95_TIME=NULL
@@ -71,6 +80,8 @@ UPLOAD_TIME=NULL
 INDEXING_TIME=NULL
 SEARCH_TIME=NULL
 NO_UPSERT_SEARCH_TIME=NULL
+CPU=NULL
+CPU_TELEMETRY=NULL
 
 RPS=$(jq -r '.results.rps' "$PARALLEL_SEARCH_RESULTS_FILE")
 MEAN_PRECISIONS=$(jq -r '.results.mean_precisions' "$PARALLEL_SEARCH_RESULTS_FILE")
@@ -82,14 +93,20 @@ NO_UPSERT_SEARCH_TIME=$(jq -r '.results.total_time' "$SEARCH_RESULT_FILE")
 UPLOAD_TIME=$(jq -r '.results.upload_time' "$PARALLEL_UPLOAD_RESULTS_FILE")
 INDEXING_TIME=$(jq -r '.results.total_time' "$PARALLEL_UPLOAD_RESULTS_FILE")
 
+if [[ "$BENCHMARK_STRATEGY" == "default" ]]; then
+  # Only this strategy produces cpu usage results files
+  CPU=$(cat "$CPU_USAGE_FILE" | tr -d '[:space:]')
+fi
+CPU_TELEMETRY=$(jq -r '.result.hardware.collection_data.benchmark.cpu' "$TELEMETRY_API_RESPONSE_FILE")
+
 QDRANT_COMMIT=$(jq -r '.commit' "$ROOT_API_RESPONSE_FILE")
 
 MEASURE_TIMESTAMP=${MEASURE_TIMESTAMP:-$(date -u +"%Y-%m-%dT%H:%M:%SZ")}
 
 
 docker run --name "vector-db" --rm jbergknoff/postgresql-client "postgresql://qdrant:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:5432/postgres" -c "
-INSERT INTO ${POSTGRES_TABLE} (engine, branch, commit, dataset, measure_timestamp, upload_time, indexing_time, rps, mean_precisions, p95_time, p99_time, search_time, no_upsert_search_time)
-VALUES ('qdrant-ci', '${QDRANT_VERSION}', '${QDRANT_COMMIT}', '${DATASETS}', '${MEASURE_TIMESTAMP}', ${UPLOAD_TIME}, ${INDEXING_TIME}, ${RPS}, ${MEAN_PRECISIONS}, ${P95_TIME}, ${P99_TIME}, ${SEARCH_TIME}, ${NO_UPSERT_SEARCH_TIME});
+INSERT INTO ${POSTGRES_TABLE} (engine, branch, commit, dataset, measure_timestamp, upload_time, indexing_time, rps, mean_precisions, p95_time, p99_time, search_time, no_upsert_search_time, cpu_telemetry, cpu)
+VALUES ('qdrant-ci', '${QDRANT_VERSION}', '${QDRANT_COMMIT}', '${DATASETS}', '${MEASURE_TIMESTAMP}', ${UPLOAD_TIME}, ${INDEXING_TIME}, ${RPS}, ${MEAN_PRECISIONS}, ${P95_TIME}, ${P99_TIME}, ${SEARCH_TIME}, ${NO_UPSERT_SEARCH_TIME}, ${CPU_TELEMETRY}, ${CPU});
 "
 
 if [[ "$IS_CI_RUN" == "true" ]]; then
@@ -103,4 +120,7 @@ if [[ "$IS_CI_RUN" == "true" ]]; then
 
   echo "upload_time=${UPLOAD_TIME}" >> "$GITHUB_OUTPUT"
   echo "indexing_time=${INDEXING_TIME}" >> "$GITHUB_OUTPUT"
+
+  echo "cpu_telemetry=${CPU_TELEMETRY}" >> "$GITHUB_OUTPUT"
+  echo "cpu=${CPU}" >> "$GITHUB_OUTPUT"
 fi
diff --git a/tools/upload_results_postgres.sh b/tools/upload_results_postgres.sh
@@ -18,8 +18,10 @@
 # 	p95_time real,
 # 	p99_time real,
 # 	vm_rss_mem real,
-# 	rss_anon_mem real
-# 	collection_load_time_ms real
+# 	rss_anon_mem real,
+# 	collection_load_time_ms real,
+# 	cpu real,
+# 	cpu_telemetry real,
 # );
 
 SEARCH_RESULTS_FILE=${SEARCH_RESULTS_FILE:-""}
@@ -69,15 +71,25 @@ if [[ -z "$ROOT_API_RESPONSE_FILE" ]]; then
   exit 1
 fi
 
+if [[ "$BENCHMARK_STRATEGY" == "default" ]]; then
+  if [[ -z "$CPU_USAGE_FILE" ]]; then
+    echo "CPU_USAGE_FILE is not set"
+    exit 1
+  fi
+fi
+
 COLLECTION_LOAD_TIME=NULL
 RPS=NULL
 MEAN_PRECISIONS=NULL
 P95_TIME=NULL
 P99_TIME=NULL
 UPLOAD_TIME=NULL
 INDEXING_TIME=NULL
+CPU=NULL
+CPU_TELEMETRY=NULL
 
 if [[ "$BENCHMARK_STRATEGY" == "collection-reload" ]]; then
+  # this strategy does not produce search & upload results files
   echo "BENCHMARK_STRATEGY is $BENCHMARK_STRATEGY, upload telemetry"
   COLLECTION_LOAD_TIME=$(jq -r '.result.collections.collections[] | select(.id == "benchmark") | .init_time_ms' "$TELEMETRY_API_RESPONSE_FILE")
 else
@@ -94,14 +106,19 @@ fi
 VM_RSS_MEMORY_USAGE=$(cat "$VM_RSS_MEMORY_USAGE_FILE" | tr -d '[:space:]')
 RSS_ANON_MEMORY_USAGE=$(cat "$RSS_ANON_MEMORY_USAGE_FILE" | tr -d '[:space:]')
 
+if [[ "$BENCHMARK_STRATEGY" == "default" ]]; then
+  # Only this strategy produces cpu usage results files
+  CPU=$(cat "$CPU_USAGE_FILE" | tr -d '[:space:]')
+fi
+CPU_TELEMETRY=$(jq -r '.result.hardware.collection_data.benchmark.cpu' "$TELEMETRY_API_RESPONSE_FILE")
+
 QDRANT_COMMIT=$(jq -r '.commit' "$ROOT_API_RESPONSE_FILE")
 
 MEASURE_TIMESTAMP=${MEASURE_TIMESTAMP:-$(date -u +"%Y-%m-%dT%H:%M:%SZ")}
 
-
 docker run --name "vector-db" --rm jbergknoff/postgresql-client "postgresql://qdrant:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:5432/postgres" -c "
-INSERT INTO ${POSTGRES_TABLE} (engine, branch, commit, dataset, measure_timestamp, upload_time, indexing_time, rps, mean_precisions, p95_time, p99_time, vm_rss_mem, rss_anon_mem, collection_load_time_ms)
-VALUES ('qdrant-ci', '${QDRANT_VERSION}', '${QDRANT_COMMIT}', '${DATASETS}', '${MEASURE_TIMESTAMP}', ${UPLOAD_TIME}, ${INDEXING_TIME}, ${RPS}, ${MEAN_PRECISIONS}, ${P95_TIME}, ${P99_TIME}, ${VM_RSS_MEMORY_USAGE}, ${RSS_ANON_MEMORY_USAGE}, ${COLLECTION_LOAD_TIME});
+INSERT INTO ${POSTGRES_TABLE} (engine, branch, commit, dataset, measure_timestamp, upload_time, indexing_time, rps, mean_precisions, p95_time, p99_time, vm_rss_mem, rss_anon_mem, collection_load_time_ms, cpu_telemetry, cpu)
+VALUES ('qdrant-ci', '${QDRANT_VERSION}', '${QDRANT_COMMIT}', '${DATASETS}', '${MEASURE_TIMESTAMP}', ${UPLOAD_TIME}, ${INDEXING_TIME}, ${RPS}, ${MEAN_PRECISIONS}, ${P95_TIME}, ${P99_TIME}, ${VM_RSS_MEMORY_USAGE}, ${RSS_ANON_MEMORY_USAGE}, ${COLLECTION_LOAD_TIME}, ${CPU_TELEMETRY}, ${CPU});
 "
 
 if [[ "$IS_CI_RUN" == "true" ]]; then
@@ -117,4 +134,7 @@ if [[ "$IS_CI_RUN" == "true" ]]; then
 
   echo "upload_time=${UPLOAD_TIME}" >> "$GITHUB_OUTPUT"
   echo "indexing_time=${INDEXING_TIME}" >> "$GITHUB_OUTPUT"
+
+  echo "cpu_telemetry=${CPU_TELEMETRY}" >> "$GITHUB_OUTPUT"
+  echo "cpu=${CPU}" >> "$GITHUB_OUTPUT"
 fi
