[CI] Enable retry logic for flaky CI tests (#14983)

This commit is contained in:
Alison Shao
2025-12-22 22:23:42 -08:00
committed by GitHub
parent 883747ced1
commit ac42797cf7
5 changed files with 392 additions and 68 deletions
+33 -1
View File
@@ -168,10 +168,18 @@ def run_a_suite(args):
pretty_print_tests(args, ci_tests, skipped_tests)
# Add extra timeout when retry is enabled
timeout = args.timeout_per_file
if args.enable_retry:
timeout += args.retry_timeout_increase
return run_unittest_files(
ci_tests,
timeout_per_file=args.timeout_per_file,
timeout_per_file=timeout,
continue_on_error=args.continue_on_error,
enable_retry=args.enable_retry,
max_attempts=args.max_attempts,
retry_wait_seconds=args.retry_wait_seconds,
)
@@ -214,6 +222,30 @@ def main():
type=int,
help="Use auto load balancing. The number of parts.",
)
parser.add_argument(
"--enable-retry",
action="store_true",
default=False,
help="Enable smart retry for accuracy/performance assertion failures (not code errors)",
)
parser.add_argument(
"--max-attempts",
type=int,
default=2,
help="Maximum number of attempts per file including initial run (default: 2)",
)
parser.add_argument(
"--retry-wait-seconds",
type=int,
default=60,
help="Seconds to wait between retries (default: 60)",
)
parser.add_argument(
"--retry-timeout-increase",
type=int,
default=600,
help="Additional timeout in seconds when retry is enabled (default: 600)",
)
args = parser.parse_args()
# Validate auto-partition arguments
+37 -1
View File
@@ -491,6 +491,30 @@ def main():
default=False,
help="Continue running remaining tests even if one fails (useful for nightly tests)",
)
arg_parser.add_argument(
"--enable-retry",
action="store_true",
default=False,
help="Enable smart retry for accuracy/performance assertion failures (not code errors)",
)
arg_parser.add_argument(
"--max-attempts",
type=int,
default=2,
help="Maximum number of attempts per file including initial run (default: 2)",
)
arg_parser.add_argument(
"--retry-wait-seconds",
type=int,
default=60,
help="Seconds to wait between retries (default: 60)",
)
arg_parser.add_argument(
"--retry-timeout-increase",
type=int,
default=600,
help="Additional timeout in seconds when retry is enabled (default: 600)",
)
args = arg_parser.parse_args()
print(f"{args=}")
@@ -506,7 +530,19 @@ def main():
print("The running tests are ", [f.name for f in files])
exit_code = run_unittest_files(files, args.timeout_per_file, args.continue_on_error)
# Add extra timeout when retry is enabled
timeout = args.timeout_per_file
if args.enable_retry:
timeout += args.retry_timeout_increase
exit_code = run_unittest_files(
files,
timeout,
args.continue_on_error,
args.enable_retry,
args.max_attempts,
args.retry_wait_seconds,
)
exit(exit_code)