Adapt benchmark result script to new fields. (#10120)

Summary:
Recently merged CI benchmark scripts were failing.

There has clearly been a major revision of the fields of benchmark output. The upload script expects and sanity-checks the existence of some fields (changes date to conform to OpenSearch format)..., so the script needs to change.

Also add a bit more exception checking to make it more obvious when this happens again.

We have deleted the existing report.tsv from the benchmark machine. An existing report.tsv is appended to by default, so that if the fields change, later rows no longer match the header. This makes for an upload that dies half way through the report file, when the format no longer matches the header.

Re-instate the config.yml for running the benchmarks, so we can once again test it in situ.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/10120

Reviewed By: pdillinger

Differential Revision: D37314908

Pulled By: jay-zhuang

fbshipit-source-id: 34f5243fee694b75c6838eb55d3398e4273254b2

.circleci/config.yml

@@ -206,7 +206,11 @@ commands:
           destination: test_logs
       - run:
           name: Send benchmark report to visualisation
-          command: ./build_tools/benchmark_log_tool.py --tsvfile /tmp/benchmark-results/report.tsv --esdocument https://search-rocksdb-bench-k2izhptfeap2hjfxteolsgsynm.us-west-2.es.amazonaws.com/bench_test3/_doc
+          command: |
+            set +e
+            set +o pipefail
+            ./build_tools/benchmark_log_tool.py --tsvfile /tmp/benchmark-results/report.tsv --esdocument https://search-rocksdb-bench-k2izhptfeap2hjfxteolsgsynm.us-west-2.es.amazonaws.com/bench_test3/_doc
+            true
 
 executors:
   windows-2xlarge:
@@ -964,6 +968,16 @@ workflows:
   build-fuzzers:
     jobs:
       - build-fuzzers
+  benchmark-linux:
+    triggers:
+      - schedule:
+          cron: "0 * * * *"
+          filters:
+            branches:
+              only:
+                - main
+    jobs:
+      - benchmark-linux
   nightly:
     triggers:
       - schedule:

build_tools/benchmark_log_tool.py

@@ -30,45 +30,56 @@ class BenchmarkResultException(Exception):
 
 class BenchmarkUtils:
 
-    expected_keys = ['ops_sec', 'mb_sec', 'total_size_gb', 'level0_size_gb', 'sum_gb', 'write_amplification',
+    expected_keys = ['ops_sec', 'mb_sec',  'lsm_sz',  'blob_sz', 'c_wgb',   'w_amp',
-                     'write_mbps', 'usec_op', 'percentile_50', 'percentile_75',
+    'c_mbps',  'c_wsecs', 'c_csecs', 'b_rgb',   'b_wgb',   'usec_op',
-                     'percentile_99', 'percentile_99.9', 'percentile_99.99', 'uptime',
+    'p50',     'p99',     'p99.9',   'p99.99', 'pmax',
-                     'stall_time', 'stall_percent', 'test_name', 'test_date', 'rocksdb_version',
+    'uptime',  'stall%', 'Nstall',  'u_cpu',   's_cpu',   'rss', 'test', 'date', 'version', 'job_id']
-                     'job_id', 'timestamp']
-
-    metric_keys = ['ops_sec', 'mb_sec', 'total_size_gb', 'level0_size_gb', 'sum_gb', 'write_amplification',
-                   'write_mbps', 'usec_op', 'percentile_50', 'percentile_75',
-                   'percentile_99', 'percentile_99.9', 'percentile_99.99', 'uptime',
-                   'stall_time', 'stall_percent']
 
     def sanity_check(row):
-        if not 'test_name' in row:
+        if not 'test' in row:
+            logging.debug(f"not 'test' in row: {row}")
             return False
-        if row['test_name'] == '':
+        if row['test'] == '':
+            logging.debug(f"row['test'] == '': {row}")
             return False
-        if not 'test_date' in row:
+        if not 'date' in row:
+            logging.debug(f"not 'date' in row: {row}")
             return False
         if not 'ops_sec' in row:
+            logging.debug(f"not 'ops_sec' in row: {row}")
             return False
         try:
             v = int(row['ops_sec'])
         except (ValueError, TypeError):
+            logging.debug(f"int(row['ops_sec']): {row}")
+            return False
+        try:
+            (_, _) = parser.parse(row['date'], fuzzy_with_tokens=True)
+        except (parser.ParserError):
+            logging.error(f"parser.parse((row['date']): not a valid format for date in row: {row}")
             return False
         return True
 
     def conform_opensearch(row):
-        (dt, _) = parser.parse(row['test_date'], fuzzy_with_tokens=True)
+        (dt, _) = parser.parse(row['date'], fuzzy_with_tokens=True)
         row['test_date'] = dt.isoformat()
         return dict((key.replace('.', '_'), value)
                     for (key, value) in row.items())
 
 
 class ResultParser:
-    def __init__(self, field="(\w|[+-:.])+", intrafield="(\s)+", separator="\t"):
+    def __init__(self, field="(\w|[+-:.%])+", intrafield="(\s)+", separator="\t"):
         self.field = re.compile(field)
         self.intra = re.compile(intrafield)
         self.sep = re.compile(separator)
 
+    def ignore(self, l_in: str):
+        if len(l_in) == 0:
+            return True
+        if l_in[0:1] == '#':
+            return True
+        return False
+
     def line(self, l_in: str):
         '''Parse a line into items
         Being clever about separators
@@ -102,7 +113,7 @@ class ResultParser:
 
     def parse(self, lines):
         '''Parse something that iterates lines'''
-        rows = [self.line(line) for line in lines]
+        rows = [self.line(line) for line in lines if not self.ignore(line)]
         header = rows[0]
         width = len(header)
         records = [{k: v for (k, v) in itertools.zip_longest(
@@ -123,7 +134,7 @@ def load_report_from_tsv(filename: str):
 def push_report_to_opensearch(report, esdocument):
     sanitized = [BenchmarkUtils.conform_opensearch(row)
                  for row in report if BenchmarkUtils.sanity_check(row)]
-    logging.debug(f"upload {len(sanitized)} benchmarks to opensearch")
+    logging.debug(f"upload {len(sanitized)} sane of {len(report)} benchmarks to opensearch")
     for single_benchmark in sanitized:
         logging.debug(f"upload benchmark: {single_benchmark}")
         response = requests.post(
@@ -133,6 +144,13 @@ def push_report_to_opensearch(report, esdocument):
             f"Sent to OpenSearch, status: {response.status_code}, result: {response.text}")
         response.raise_for_status()
 
+def push_report_to_null(report):
+    
+    for row in report:
+        if BenchmarkUtils.sanity_check(row):
+            logging.debug(f"row {row}")
+            conformed = BenchmarkUtils.conform_opensearch(row)
+            logging.debug(f"conformed row {conformed}")
 
 def main():
     '''Tool for fetching, parsing and uploading benchmark results to OpenSearch / ElasticSearch
@@ -151,11 +169,15 @@ def main():
     parser.add_argument('--tsvfile', default='build_tools/circle_api_scraper_input.txt',
                         help='File from which to read tsv report')
     parser.add_argument('--esdocument', help='ElasticSearch/OpenSearch document URL to upload report into')
+    parser.add_argument('--upload', choices=['opensearch', 'none'], default='opensearch')
 
     args = parser.parse_args()
     logging.debug(f"Arguments: {args}")
     reports = load_report_from_tsv(args.tsvfile)
-    push_report_to_opensearch(reports, args.esdocument)
+    if (args.upload == 'opensearch'):
+        push_report_to_opensearch(reports, args.esdocument)
+    else:
+        push_report_to_null(reports)
 
 if __name__ == '__main__':
     sys.exit(main())