ci/lava: Detect a6xx gpu recovery failures

Sporadically a6xx gpu will fail to recover causing the lava job
a660_vk_full to loop on error messages for three hours before timing
out.

A few sporadic error messages may still be recoverable, but when multiple
errors occur over a short period, successful recovery is unlikely. Parse
the logs to look for repeated error messages within a short time period.
If found, cancel the lava job and rerun it.

Also add unit tests for this behaviour.

cc: mesa-stable

Reported-by: Valentine Burley <valentine.burley@gmail.com>
Acked-by: Daniel Stone <daniel.stone@collabora.com>
Reviewed-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Signed-off-by: Deborah Brouwer <deborah.brouwer@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30032>
This commit is contained in:
Deborah Brouwer
2024-07-05 15:28:24 -07:00
committed by Marge Bot
parent 53f249b921
commit 72c182f873
3 changed files with 83 additions and 2 deletions

View File

@@ -23,3 +23,10 @@ KNOWN_ISSUE_R8152_PATTERNS: tuple[str, ...] = (
# This is considered noise, since LAVA produces this log after receiving a package of feedback
# messages.
LOG_DEBUG_FEEDBACK_NOISE = "Listened to connection for namespace 'dut' done"
A6XX_GPU_RECOVERY_WATCH_PERIOD_MIN = 3
A6XX_GPU_RECOVERY_FAILURE_MAX_COUNT = 30
A6XX_GPU_RECOVERY_FAILURE_MESSAGE = (
"cx gdsc didn't collapse",
"Timeout waiting for GMU OOB",
)

View File

@@ -1,8 +1,9 @@
from __future__ import annotations
import re
from datetime import datetime, timedelta
from dataclasses import dataclass, field
from typing import TYPE_CHECKING, Any, Sequence
from typing import TYPE_CHECKING, Any, Optional, Sequence
if TYPE_CHECKING:
from lava.utils import LogFollower
@@ -13,6 +14,9 @@ from lava.utils.constants import (
KNOWN_ISSUE_R8152_MAX_CONSECUTIVE_COUNTER,
LOG_DEBUG_FEEDBACK_NOISE,
KNOWN_ISSUE_R8152_PATTERNS,
A6XX_GPU_RECOVERY_WATCH_PERIOD_MIN,
A6XX_GPU_RECOVERY_FAILURE_MESSAGE,
A6XX_GPU_RECOVERY_FAILURE_MAX_COUNT,
)
from lava.utils.log_section import LogSectionType
@@ -29,6 +33,8 @@ class LAVALogHints:
log_follower: LogFollower
r8152_issue_consecutive_counter: int = field(default=0, init=False)
reboot_counter: int = field(default=0, init=False)
a6xx_gpu_recovery_fail_counter: int = field(default=0, init=False)
a6xx_gpu_first_fail_time: Optional[datetime] = field(default=None, init=False)
def raise_known_issue(self, message) -> None:
raise MesaCIKnownIssueException(
@@ -44,6 +50,7 @@ class LAVALogHints:
continue
self.detect_r8152_issue(line)
self.detect_forced_reboot(line)
self.detect_a6xx_gpu_recovery_failure(line)
def detect_r8152_issue(self, line):
if self.log_follower.phase in (
@@ -77,3 +84,23 @@ class LAVALogHints:
self.raise_known_issue(
"Forced reboot detected during test phase, failing the job..."
)
# If the a6xx gpu repeatedly fails to recover over a short period of time,
# then successful recovery is unlikely so cancel the job preemptively.
def detect_a6xx_gpu_recovery_failure(self, line: dict[str, Any]) -> None:
if search_known_issue_patterns(A6XX_GPU_RECOVERY_FAILURE_MESSAGE, line["msg"]):
time_of_failure = datetime.fromisoformat(line["dt"])
self.a6xx_gpu_recovery_fail_counter += 1
if self.a6xx_gpu_first_fail_time is None:
self.a6xx_gpu_first_fail_time = time_of_failure
if self.a6xx_gpu_recovery_fail_counter == A6XX_GPU_RECOVERY_FAILURE_MAX_COUNT:
time_since_first_fail = time_of_failure - self.a6xx_gpu_first_fail_time
if time_since_first_fail <= timedelta(minutes=A6XX_GPU_RECOVERY_WATCH_PERIOD_MIN):
self.raise_known_issue(
"Repeated GPU recovery failure detected: cancelling the job"
)
else:
self.a6xx_gpu_first_fail_time = None
self.a6xx_gpu_recovery_fail_counter = 0

View File

@@ -16,7 +16,13 @@ from lava.utils import (
fix_lava_gitlab_section_log,
hide_sensitive_data,
)
from lava.utils.constants import KNOWN_ISSUE_R8152_MAX_CONSECUTIVE_COUNTER
from lava.utils.constants import (
KNOWN_ISSUE_R8152_MAX_CONSECUTIVE_COUNTER,
A6XX_GPU_RECOVERY_WATCH_PERIOD_MIN,
A6XX_GPU_RECOVERY_FAILURE_MESSAGE,
A6XX_GPU_RECOVERY_FAILURE_MAX_COUNT,
)
from lava.utils.lava_log_hints import LAVALogHints
from ..lava.helpers import (
create_lava_yaml_msg,
@@ -390,3 +396,44 @@ def test_detect_failure(messages, expectation):
lf = LogFollower(starting_section=boot_section)
with expectation:
lf.feed(messages)
def test_detect_a6xx_gpu_recovery_failure(frozen_time):
log_follower = LogFollower()
lava_log_hints = LAVALogHints(log_follower=log_follower)
failure_message = {
"dt": datetime.now().isoformat(),
"msg": A6XX_GPU_RECOVERY_FAILURE_MESSAGE[0],
"lvl": "feedback",
}
with pytest.raises(MesaCIKnownIssueException):
for _ in range(A6XX_GPU_RECOVERY_FAILURE_MAX_COUNT):
lava_log_hints.detect_a6xx_gpu_recovery_failure(failure_message)
# Simulate the passage of time within the watch period
frozen_time.tick(1)
failure_message["dt"] = datetime.now().isoformat()
def test_detect_a6xx_gpu_recovery_success(frozen_time):
log_follower = LogFollower()
lava_log_hints = LAVALogHints(log_follower=log_follower)
failure_message = {
"dt": datetime.now().isoformat(),
"msg": A6XX_GPU_RECOVERY_FAILURE_MESSAGE[0],
"lvl": "feedback",
}
# Simulate sending a tolerable number of failure messages
for _ in range(A6XX_GPU_RECOVERY_FAILURE_MAX_COUNT - 1):
lava_log_hints.detect_a6xx_gpu_recovery_failure(failure_message)
frozen_time.tick(1)
failure_message["dt"] = datetime.now().isoformat()
# Simulate the passage of time outside of the watch period
frozen_time.tick(60 * A6XX_GPU_RECOVERY_WATCH_PERIOD_MIN + 1)
failure_message = {
"dt": datetime.now().isoformat(),
"msg": A6XX_GPU_RECOVERY_FAILURE_MESSAGE[1],
"lvl": "feedback",
}
with does_not_raise():
lava_log_hints.detect_a6xx_gpu_recovery_failure(failure_message)
assert lava_log_hints.a6xx_gpu_first_fail_time is None, "a6xx_gpu_first_fail_time is not None"
assert lava_log_hints.a6xx_gpu_recovery_fail_counter == 0, "a6xx_gpu_recovery_fail_counter is not 0"