Merge pull request #198 from igsilya/aggregate-resource-usage

Add aggregate resource usage graphs

do.sh

@@ -406,13 +406,19 @@ function mine_data() {
     resource_usage_logs=$(find ${out_dir}/logs -name process-stats.json \
                             | grep -E 'ovn-tester|ovn-central-az[0-2]-')
     python3 ${topdir}/utils/process-stats.py \
-        resource-usage-report-central.html ${resource_usage_logs}
+        -o resource-usage-report-central.html ${resource_usage_logs}
 
     # Collecting stats only for 3 workers to avoid bloating the report.
     resource_usage_logs=$(find ${out_dir}/logs -name process-stats.json \
                             | grep ovn-scale | head -3)
     python3 ${topdir}/utils/process-stats.py \
-        resource-usage-report-worker.html ${resource_usage_logs}
+        -o resource-usage-report-worker.html ${resource_usage_logs}
+
+    # Preparing reports for aggregate resource usage.
+    resource_usage_logs=$(find ${out_dir}/logs -name process-stats.json)
+    python3 ${topdir}/utils/process-stats.py --aggregate \
+        -o resource-usage-report-aggregate.html ${resource_usage_logs}
+
     deactivate
 
     popd

ovn-fake-multinode-utils/process-monitor.py

@@ -22,7 +22,7 @@ def monitor(suffix: str, out_file: str, exit_file: str) -> None:
             for p in psutil.process_iter():
                 if any(name in p.name() for name in process_names):
                     processes.add(p)
-                elif any(
+                elif p.name() != 'monitor' and any(
                     name in part
                     for part in p.cmdline()
                     for name in process_names
@@ -36,7 +36,14 @@ def monitor(suffix: str, out_file: str, exit_file: str) -> None:
             tme = time.time()
             for p in processes:
                 try:
-                    name = p.name() + "-" + suffix + "-" + str(p.pid)
+                    name = p.name()
+                    for arg in p.cmdline():
+                        if arg.endswith('.pid') or arg.endswith('.py'):
+                            name = arg.split('/')[-1].split('.')[0]
+                            break
+
+                    name = name + "|" + suffix + "|" + str(p.pid)
+
                     # cpu_percent(seconds) call will block
                     # for the amount of seconds specified.
                     cpu = p.cpu_percent(0.5)

utils/process-stats.py

@@ -1,45 +1,118 @@
+import argparse
 import json
+import logging
 import os
 import pandas as pd
 import plotly.express as px
 import sys
 
-from datetime import datetime
 from typing import Dict, List
 
 
+FORMAT = '%(asctime)s |%(levelname)s| %(message)s'
+logging.basicConfig(stream=sys.stdout, level=logging.INFO, format=FORMAT)
+log = logging.getLogger(__name__)
+
+
 def read_file(filename: str) -> Dict:
     with open(filename, "r") as file:
         return json.load(file)
 
 
-def resource_stats_generate(filename: str, data: Dict) -> None:
+def aggregated(df: pd.DataFrame) -> (pd.DataFrame, int):
+    column_names = list(df.columns)
+    value_name = column_names[2]
+
+    log.info(f'Pivot and interpolate {value_name} ...')
+    df = df.pivot_table(
+        index='Time', columns='Process', values=value_name, aggfunc='mean'
+    ).interpolate(method='time', limit_direction='both')
+
+    result = pd.DataFrame(index=df.index)
+    processes = {p.split('|')[0] for p in df.columns}
+
+    log.info(f'Aggregating {value_name} ...')
+    for p in processes:
+        df_filtered = df.filter(regex='^' + p)
+        result[p + '|sum'] = df_filtered.sum(axis=1)
+        result[p + '|mean'] = df_filtered.mean(axis=1)
+        result[p + '|max'] = df_filtered.max(axis=1)
+        result[p + '|min'] = df_filtered.min(axis=1)
+
+    result['ovn|sum'] = df.filter(regex=r'^ovn.*\|ovn-(central|scale).*').sum(
+        axis=1
+    )
+    ovn_max = result['ovn|sum'].astype('int').max()
+
+    result['ovs|sum'] = df.filter(regex=r'^ovs.*\|ovn-(central|scale).*').sum(
+        axis=1
+    )
+
+    result = result.astype('int').reset_index().melt(id_vars=['Time'])
+    result.columns = column_names
+    result = result.sort_values(['Process', 'Time'])
+
+    return result, ovn_max
+
+
+def resource_stats_generate(
+    filename: str, data: Dict, aggregate: bool
+) -> None:
     rss: List[List] = []
     cpu: List[List] = []
 
+    log.info('Preprocessing ...')
     for ts, time_slice in sorted(data.items()):
+        tme = pd.Timestamp.fromtimestamp(float(ts)).round('1s')
         for name, res in time_slice.items():
-            tme = datetime.fromtimestamp(float(ts))
             rss_mb = int(res['rss']) >> 20
             rss.append([tme, name, rss_mb])
             cpu.append([tme, name, float(res['cpu'])])
 
+    log.info('Creating DataFrame ...')
     df_rss = pd.DataFrame(rss, columns=['Time', 'Process', 'RSS (MB)'])
     df_cpu = pd.DataFrame(cpu, columns=['Time', 'Process', 'CPU (%)'])
 
+    if aggregate:
+        df_rss, max_sum_rss = aggregated(df_rss)
+        df_cpu, max_sum_cpu = aggregated(df_cpu)
+
+    log.info('Creating charts ...')
     rss_chart = px.line(
         df_rss,
         x='Time',
         y='RSS (MB)',
         color='Process',
-        title='Resident Set Size',
+        title=('Aggregate ' if aggregate else '') + 'Resident Set Size',
     )
     cpu_chart = px.line(
-        df_cpu, x='Time', y='CPU (%)', color='Process', title='CPU usage'
+        df_cpu,
+        x='Time',
+        y='CPU (%)',
+        color='Process',
+        title=('Aggregate ' if aggregate else '') + 'CPU usage',
     )
 
+    log.info(f'Writing HTML to {filename} ...')
     with open(filename, 'w') as report_file:
         report_file.write('<html>')
+        if aggregate:
+            report_file.write(
+                f'''
+                <table border="1" class="dataframe">
+                <tbody>
+                    <tr>
+                        <td>Max(Sum(OVN RSS))</td>
+                        <td> {max_sum_rss} MB</td>
+                    </tr>
+                    <tr>
+                        <td>Max(Sum(OVN CPU))</td>
+                        <td> {max_sum_cpu} %</td>
+                    </tr>
+                </tbody>
+                </table>
+            '''
+            )
         report_file.write(
             rss_chart.to_html(
                 full_html=False,
@@ -60,17 +133,36 @@ def resource_stats_generate(filename: str, data: Dict) -> None:
 
 
 if __name__ == '__main__':
-    if len(sys.argv) < 3:
-        print(f'Usage: {sys.argv[0]} output-file input-file [input-file ...]')
-        sys.exit(1)
+    parser = argparse.ArgumentParser(
+        description='Generate resource usage charts.'
+    )
+    parser.add_argument(
+        '--aggregate', action='store_true', help='generate aggregate charts'
+    )
+    parser.add_argument(
+        '-o', '--output', required=True, help='file to write an HTML result'
+    )
+    parser.add_argument(
+        'input_files',
+        metavar='input-file',
+        type=str,
+        nargs='+',
+        help='JSON file with recorded process statistics',
+    )
 
-    if os.path.isfile(sys.argv[1]):
-        print(f'Output file {sys.argv[1]} already exists')
+    args = parser.parse_args()
+
+    if os.path.isfile(args.output):
+        log.fatal(f'Output file {args.output} already exists')
         sys.exit(2)
 
+    log.info(f'Processing stats from {len(args.input_files)} files.')
+
+    log.info('Reading ...')
     data: Dict = {}
-    for f in sys.argv[2:]:
+    for f in args.input_files:
         d = read_file(f)
         data.update(d)
 
-    resource_stats_generate(sys.argv[1], data)
+    resource_stats_generate(args.output, data, args.aggregate)
+    log.info('Done.')