Merge bitcoin/bitcoin#23799: test: Let test_runner.py start multiple jobs per timeslot

975097f Let test_runner.py start multiple jobs per timeslot (Pieter Wuille)

Pull request description:

  test_runner.py currently only checks every 0.5s whether any job has finished, and if so, starts at most one new job. At higher parallellism it becomes increasingly likely that multiple jobs have finished at the same time. Fix this by always noticing *all* finished jobs every timeslot, and starting as many new ones.

ACKs for top commit:
  laanwj:
    Code review and lightly tested ACK 975097f
  prayank23:
    ACK bitcoin/bitcoin@975097f

Tree-SHA512: b70c51f05efcde9bc25475c192b86e86b4c399495b42dee20576af3e6b99e8298be8b9e82146abdabbaedb24a13ee158a7c8947518b16fc4f33a3b434935b550

Author: laanwj

test/functional/test_runner.py

@@ -530,33 +530,35 @@ def run_tests(*, test_list, src_dir, build_dir, tmpdir, jobs=1, enable_coverage=
 
     max_len_name = len(max(test_list, key=len))
     test_count = len(test_list)
-    for i in range(test_count):
-        test_result, testdir, stdout, stderr = job_queue.get_next()
-        test_results.append(test_result)
-        done_str = "{}/{} - {}{}{}".format(i + 1, test_count, BOLD[1], test_result.name, BOLD[0])
-        if test_result.status == "Passed":
-            logging.debug("%s passed, Duration: %s s" % (done_str, test_result.time))
-        elif test_result.status == "Skipped":
-            logging.debug("%s skipped" % (done_str))
-        else:
-            print("%s failed, Duration: %s s\n" % (done_str, test_result.time))
-            print(BOLD[1] + 'stdout:\n' + BOLD[0] + stdout + '\n')
-            print(BOLD[1] + 'stderr:\n' + BOLD[0] + stderr + '\n')
-            if combined_logs_len and os.path.isdir(testdir):
-                # Print the final `combinedlogslen` lines of the combined logs
-                print('{}Combine the logs and print the last {} lines ...{}'.format(BOLD[1], combined_logs_len, BOLD[0]))
-                print('\n============')
-                print('{}Combined log for {}:{}'.format(BOLD[1], testdir, BOLD[0]))
-                print('============\n')
-                combined_logs_args = [sys.executable, os.path.join(tests_dir, 'combine_logs.py'), testdir]
-                if BOLD[0]:
-                    combined_logs_args += ['--color']
-                combined_logs, _ = subprocess.Popen(combined_logs_args, universal_newlines=True, stdout=subprocess.PIPE).communicate()
-                print("\n".join(deque(combined_logs.splitlines(), combined_logs_len)))
-
-            if failfast:
-                logging.debug("Early exiting after test failure")
-                break
+    i = 0
+    while i < test_count:
+        for test_result, testdir, stdout, stderr in job_queue.get_next():
+            test_results.append(test_result)
+            i += 1
+            done_str = "{}/{} - {}{}{}".format(i, test_count, BOLD[1], test_result.name, BOLD[0])
+            if test_result.status == "Passed":
+                logging.debug("%s passed, Duration: %s s" % (done_str, test_result.time))
+            elif test_result.status == "Skipped":
+                logging.debug("%s skipped" % (done_str))
+            else:
+                print("%s failed, Duration: %s s\n" % (done_str, test_result.time))
+                print(BOLD[1] + 'stdout:\n' + BOLD[0] + stdout + '\n')
+                print(BOLD[1] + 'stderr:\n' + BOLD[0] + stderr + '\n')
+                if combined_logs_len and os.path.isdir(testdir):
+                    # Print the final `combinedlogslen` lines of the combined logs
+                    print('{}Combine the logs and print the last {} lines ...{}'.format(BOLD[1], combined_logs_len, BOLD[0]))
+                    print('\n============')
+                    print('{}Combined log for {}:{}'.format(BOLD[1], testdir, BOLD[0]))
+                    print('============\n')
+                    combined_logs_args = [sys.executable, os.path.join(tests_dir, 'combine_logs.py'), testdir]
+                    if BOLD[0]:
+                        combined_logs_args += ['--color']
+                    combined_logs, _ = subprocess.Popen(combined_logs_args, universal_newlines=True, stdout=subprocess.PIPE).communicate()
+                    print("\n".join(deque(combined_logs.splitlines(), combined_logs_len)))
+
+                if failfast:
+                    logging.debug("Early exiting after test failure")
+                    break
 
     print_results(test_results, max_len_name, (int(time.time() - start_time)))
 
@@ -650,8 +652,9 @@ def get_next(self):
 
         dot_count = 0
         while True:
-            # Return first proc that finishes
+            # Return all procs that have finished, if any. Otherwise sleep until there is one.
             time.sleep(.5)
+            ret = []
             for job in self.jobs:
                 (name, start_time, proc, testdir, log_out, log_err) = job
                 if proc.poll() is not None:
@@ -670,7 +673,9 @@ def get_next(self):
                         clearline = '\r' + (' ' * dot_count) + '\r'
                         print(clearline, end='', flush=True)
                     dot_count = 0
-                    return TestResult(name, status, int(time.time() - start_time)), testdir, stdout, stderr
+                    ret.append((TestResult(name, status, int(time.time() - start_time)), testdir, stdout, stderr))
+            if ret:
+                return ret
             if self.use_term_control:
                 print('.', end='', flush=True)
             dot_count += 1