[201811] Check platform reboot cause to see if any reset happened during fast/warm-reboot (#8912)

[201811] Check platform reboot cause to see if any reset happened during fast/warm-reboot

Why I did it
To recover syncd and swss from any cold reset during fast/warm-reboot

How I did it
Check platform reboot-cause to see if any cold reset happens for fast-reboot power up

How to verify it
Manual test
This commit is contained in:
Sujin Kang 2021-12-01 10:50:55 -08:00 committed by GitHub
parent 9e75e856eb
commit a80319e2d0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
7 changed files with 290 additions and 140 deletions

View File

@ -2,6 +2,7 @@
Description=Database container Description=Database container
Requires=docker.service Requires=docker.service
After=docker.service After=docker.service
After=rc-local.service
[Service] [Service]
User=root User=root

View File

@ -5,6 +5,17 @@ function getMountPoint()
echo $1 | python -c "import sys, json, os; mnts = [x for x in json.load(sys.stdin)[0]['Mounts'] if x['Destination'] == '/usr/share/sonic/hwsku']; print '' if len(mnts) == 0 else os.path.basename(mnts[0]['Source'])" 2>/dev/null echo $1 | python -c "import sys, json, os; mnts = [x for x in json.load(sys.stdin)[0]['Mounts'] if x['Destination'] == '/usr/share/sonic/hwsku']; print '' if len(mnts) == 0 else os.path.basename(mnts[0]['Source'])" 2>/dev/null
} }
function isPlatformReset()
{
output=$(python -c "import sonic_platform.platform; p = sonic_platform.platform.Platform(); c = p.get_chassis(); hw_rc_major, hw_rc_minor = c.get_reboot_cause(); print(hw_rc_major)" 2>/dev/null)
if [[ "${output}" != "Non-Hardware" ]]; then
return 0
else
return 1
fi
}
function getBootType() function getBootType()
{ {
# same code snippet in files/scripts/syncd.sh # same code snippet in files/scripts/syncd.sh
@ -21,6 +32,9 @@ function getBootType()
*) *)
TYPE='cold' TYPE='cold'
esac esac
if isPlatformReset; then
TYPE='cold'
fi
echo "${TYPE}" echo "${TYPE}"
} }
@ -59,7 +73,7 @@ function postStartAction()
fi fi
if [[ "$BOOT_TYPE" == "fast" ]]; then if [[ "$BOOT_TYPE" == "fast" ]]; then
# set the key to expire in 3 minutes # set the FAST_REBOOT|system key to expire in 3 minutes
redis-cli -n 6 SET "FAST_REBOOT|system" "1" "EX" "180" redis-cli -n 6 SET "FAST_REBOOT|system" "1" "EX" "180"
fi fi

View File

@ -268,6 +268,11 @@ sudo cp $IMAGE_CONFIGS/process-reboot-cause/process-reboot-cause.service $FILES
sudo LANG=C chroot $FILESYSTEM_ROOT systemctl enable process-reboot-cause.service sudo LANG=C chroot $FILESYSTEM_ROOT systemctl enable process-reboot-cause.service
sudo cp $IMAGE_CONFIGS/process-reboot-cause/process-reboot-cause $FILESYSTEM_ROOT/usr/bin/ sudo cp $IMAGE_CONFIGS/process-reboot-cause/process-reboot-cause $FILESYSTEM_ROOT/usr/bin/
# Copy determine-reboot-cause service files
sudo cp $IMAGE_CONFIGS/process-reboot-cause/determine-reboot-cause.service $FILESYSTEM_ROOT/etc/systemd/system/
sudo LANG=C chroot $FILESYSTEM_ROOT systemctl enable determine-reboot-cause.service
sudo cp $IMAGE_CONFIGS/process-reboot-cause/determine-reboot-cause $FILESYSTEM_ROOT/usr/bin/
## Install package without starting service ## Install package without starting service
## ref: https://wiki.debian.org/chroot ## ref: https://wiki.debian.org/chroot
sudo tee -a $FILESYSTEM_ROOT/usr/sbin/policy-rc.d > /dev/null <<EOF sudo tee -a $FILESYSTEM_ROOT/usr/sbin/policy-rc.d > /dev/null <<EOF

View File

@ -8,6 +8,7 @@ Requires=nps-modules-4.9.0-14-amd64.service
{% endif %} {% endif %}
After=database.service updategraph.service After=database.service updategraph.service
After=interfaces-config.service After=interfaces-config.service
After=pmon.service
Before=ntp-config.service Before=ntp-config.service
StartLimitIntervalSec=1200 StartLimitIntervalSec=1200
StartLimitBurst=3 StartLimitBurst=3

View File

@ -0,0 +1,248 @@
#!/usr/bin/env python
#
# determine-reboot-cause
#
# Program designed to run once, soon after system boot which will
# determine the cause of the previous reboot and store it to the disk,
#
try:
import datetime
import json
import os
import pwd
import re
import sys
import syslog
import sonic_device_util
except ImportError as err:
raise ImportError("%s - required module not found" % str(err))
VERSION = "1.0"
SYSLOG_IDENTIFIER = "determine-reboot-cause"
REBOOT_CAUSE_DIR = "/host/reboot-cause/"
REBOOT_CAUSE_FILE = REBOOT_CAUSE_DIR + "reboot-cause.txt"
PREVIOUS_REBOOT_CAUSE_FILE = REBOOT_CAUSE_DIR + "previous-reboot-cause.txt"
PREVIOUS_REBOOT_CAUSE_FILE_JSON = REBOOT_CAUSE_DIR + "previous-reboot-cause.json"
FIRST_BOOT_PLATFORM_FILE = "/tmp/notify_firstboot_to_platform"
REBOOT_TYPE_KEXEC_FILE = "/proc/cmdline"
# The following SONIC_BOOT_TYPEs come from the warm/fast reboot script which is in sonic-utilities
# Because the system can be rebooted from some old versions, we have to take all possible BOOT options into consideration.
# On 201803, 201807 we have
# BOOT_OPTIONS="$(echo $KERNEL_OPTIONS | sed -e 's/\s*linux\s*/BOOT_IMAGE=/') fast-reboot"
# On 201811 and later we have
# BOOT_OPTIONS="$(echo $KERNEL_OPTIONS | sed -e 's/\s*linux\s*/BOOT_IMAGE=/') SONIC_BOOT_TYPE=${BOOT_TYPE_ARG}" where BOOT_TYPE_ARG can be warm, fastfast or fast
# To extract the commom part of them, we should have the following PATTERN
REBOOT_TYPE_KEXEC_PATTERN_WARM = ".*SONIC_BOOT_TYPE=(warm|fastfast).*"
REBOOT_TYPE_KEXEC_PATTERN_FAST = ".*SONIC_BOOT_TYPE=(fast|fast-reboot).*"
REBOOT_CAUSE_UNKNOWN = "Unknown"
# ========================== Syslog wrappers ==========================
def log_info(msg):
syslog.openlog(SYSLOG_IDENTIFIER)
syslog.syslog(syslog.LOG_INFO, msg)
syslog.closelog()
def log_warning(msg):
syslog.openlog(SYSLOG_IDENTIFIER)
syslog.syslog(syslog.LOG_WARNING, msg)
syslog.closelog()
def log_error(msg):
syslog.openlog(SYSLOG_IDENTIFIER)
syslog.syslog(syslog.LOG_ERR, msg)
syslog.closelog()
# ============================= Functions =============================
def parse_warmfast_reboot_from_proc_cmdline():
if os.path.isfile(REBOOT_TYPE_KEXEC_FILE):
with open(REBOOT_TYPE_KEXEC_FILE, "r") as cause_file:
cause_file_kexec = cause_file.readline()
m = re.search(REBOOT_TYPE_KEXEC_PATTERN_WARM, cause_file_kexec)
if m and m.group(1):
return 'warm-reboot'
m = re.search(REBOOT_TYPE_KEXEC_PATTERN_FAST, cause_file_kexec)
if m and m.group(1):
return 'fast-reboot'
return None
def find_software_reboot_cause():
software_reboot_cause = REBOOT_CAUSE_UNKNOWN
if os.path.isfile(REBOOT_CAUSE_FILE):
with open(REBOOT_CAUSE_FILE, "r") as cause_file:
software_reboot_cause = cause_file.readline().rstrip('\n')
log_info("{} indicates the reboot cause: {}".format(REBOOT_CAUSE_FILE, software_reboot_cause))
else:
log_info("Reboot cause file {} not found".format(REBOOT_CAUSE_FILE))
if os.path.isfile(FIRST_BOOT_PLATFORM_FILE):
if software_reboot_cause == REBOOT_CAUSE_UNKNOWN:
version_info = sonic_device_util.get_sonic_version_info()
build_version = version_info['build_version'] if version_info else "unknown"
software_reboot_cause += " (First boot of SONiC version {})".format(build_version)
os.remove(FIRST_BOOT_PLATFORM_FILE)
return software_reboot_cause
def find_proc_cmdline_reboot_cause():
proc_cmdline_reboot_cause = parse_warmfast_reboot_from_proc_cmdline()
if proc_cmdline_reboot_cause:
log_info("/proc/cmdline indicates reboot type: {}".format(proc_cmdline_reboot_cause))
else:
log_info("No reboot cause found from /proc/cmdline")
return proc_cmdline_reboot_cause
def find_hardware_reboot_cause():
hardware_reboot_cause = None
# Until all platform vendors have provided sonic_platform packages,
# if there is no sonic_platform package installed, we only provide
# software-related reboot causes.
try:
import sonic_platform
platform = sonic_platform.platform.Platform()
chassis = platform.get_chassis()
hardware_reboot_cause_major, hardware_reboot_cause_minor = chassis.get_reboot_cause()
if hardware_reboot_cause_major == chassis.REBOOT_CAUSE_NON_HARDWARE:
# The reboot was not caused by hardware. If there is a REBOOT_CAUSE_FILE, it will
# contain any software-related reboot info. We will use it as the previous cause.
pass
elif hardware_reboot_cause_major == chassis.REBOOT_CAUSE_HARDWARE_OTHER:
hardware_reboot_cause = "{} ({})".format(hardware_reboot_cause_major, hardware_reboot_cause_minor)
else:
hardware_reboot_cause = hardware_reboot_cause_major
except ImportError as err:
log_warning("sonic_platform package not installed. Unable to detect hardware reboot causes.")
if hardware_reboot_cause:
log_info("Platform api indicates reboot cause {}".format(hardware_reboot_cause))
else:
log_info("No reboot cause found from platform api")
return hardware_reboot_cause
def get_reboot_cause_dict(previous_reboot_cause, comment, gen_time):
"""Store the key infomation of device reboot into a dictionary by parsing the string in
previous_reboot_cause.
If user issused a command to reboot device, then user, command and time will be
stored into a dictionary.
If device was rebooted due to the kernel panic, then the string `Kernel Panic`
and time will be stored into a dictionary.
"""
reboot_cause_dict = {}
reboot_cause_dict['gen_time'] = gen_time
reboot_cause_dict['cause'] = previous_reboot_cause
reboot_cause_dict['user'] = "N/A"
reboot_cause_dict['time'] = "N/A"
reboot_cause_dict['comment'] = comment if comment is not None else "N/A"
if re.search(r'User issued', previous_reboot_cause):
# Match with "User issued '{}' command [User: {}, Time: {}]"
match = re.search(r'User issued \'(.*)\' command \[User: (.*), Time: (.*)\]', previous_reboot_cause)
if match is not None:
reboot_cause_dict['cause'] = match.group(1)
reboot_cause_dict['user'] = match.group(2)
reboot_cause_dict['time'] = match.group(3)
elif re.search(r'Kernel Panic', previous_reboot_cause):
match = re.search(r'Kernel Panic \[Time: (.*)\]', previous_reboot_cause)
if match is not None:
reboot_cause_dict['cause'] = "Kernel Panic"
reboot_cause_dict['time'] = match.group(1)
return reboot_cause_dict
def main():
log_info("Starting up...")
if not os.geteuid() == 0:
log_error("User {} does not have permission to execute".format(pwd.getpwuid(os.getuid()).pw_name))
sys.exit("This utility must be run as root")
# Create REBOOT_CAUSE_DIR if it doesn't exist
if not os.path.exists(REBOOT_CAUSE_DIR):
os.makedirs(REBOOT_CAUSE_DIR)
# Remove stale PREVIOUS_REBOOT_CAUSE_FILE if it exists
if os.path.exists(PREVIOUS_REBOOT_CAUSE_FILE):
os.remove(PREVIOUS_REBOOT_CAUSE_FILE)
# Set a default previous reboot cause
previous_reboot_cause = REBOOT_CAUSE_UNKNOWN
# 1. Check if the previous reboot was warm/fast reboot by testing whether there is "fast|fastfast|warm" in /proc/cmdline
proc_cmdline_reboot_cause = find_proc_cmdline_reboot_cause()
# 2. Check if the previous reboot was caused by hardware
# If yes, the hardware reboot cause will be treated as the reboot cause
hardware_reboot_cause = find_hardware_reboot_cause()
# 3. If there is a REBOOT_CAUSE_FILE, it will contain any software-related
# reboot info. We will use it as the previous cause.
software_reboot_cause = find_software_reboot_cause()
additional_reboot_info = None
# The main decision logic of the reboot cause:
# If there is a reboot cause indicated by /proc/cmdline, it should be warmreboot/fastreboot
# the software_reboot_cause which is the content of /hosts/reboot-cause/reboot-cause.txt
# will be treated as the reboot cause
# Elif there is a reboot cause indicated by platform API,
# the hardware_reboot_cause will be treated as the reboot cause
# Else the software_reboot_cause will be treated as the reboot cause
if proc_cmdline_reboot_cause is not None:
previous_reboot_cause = software_reboot_cause
if not hardware_reboot_cause.startswith(REBOOT_CAUSE_NON_HARDWARE):
# Add the hardware_reboot_cause into additional_reboot_info
additional_reboot_info = hardware_reboot_cause
elif hardware_reboot_cause is not None:
previous_reboot_cause = hardware_reboot_cause
# Check if any software reboot was issued before this hardware reboot happened
if software_reboot_cause is not REBOOT_CAUSE_UNKNOWN:
additional_reboot_info = software_reboot_cause
else:
previous_reboot_cause = software_reboot_cause
# Current time
reboot_cause_gen_time = str(datetime.datetime.now().strftime('%Y_%m_%d_%H_%M_%S'))
# Save the previous cause info into its history file as json format
reboot_cause_dict = get_reboot_cause_dict(previous_reboot_cause, additional_reboot_info, reboot_cause_gen_time)
# Write the previous reboot cause to REBOOT_CAUSE_HISTORY_FILE_JSON as a JSON format
with open(PREVIOUS_REBOOT_CAUSE_FILE_JSON, "w") as reboot_cause_history_file:
json.dump(reboot_cause_dict, reboot_cause_history_file)
# Write the previous reboot cause to PREVIOUS_REBOOT_CAUSE_FILE
with open(PREVIOUS_REBOOT_CAUSE_FILE, "w") as prev_cause_file:
prev_cause_file.write(previous_reboot_cause)
# Remove the old REBOOT_CAUSE_FILE
if os.path.exists(REBOOT_CAUSE_FILE):
os.remove(REBOOT_CAUSE_FILE)
# Write a new default reboot cause file for the next reboot
with open(REBOOT_CAUSE_FILE, "w") as cause_file:
cause_file.write(REBOOT_CAUSE_UNKNOWN)
if __name__ == "__main__":
main()

View File

@ -0,0 +1,7 @@
[Unit]
Description=Reboot cause determination service
After=rc-local.service
[Service]
Type=simple
ExecStart=/usr/bin/determine-reboot-cause

View File

@ -7,11 +7,12 @@
# #
try: try:
import json
import os import os
import pwd import pwd
import re
import sys import sys
import syslog import syslog
import re
import sonic_device_util import sonic_device_util
except ImportError as err: except ImportError as err:
@ -22,19 +23,7 @@ VERSION = "1.0"
SYSLOG_IDENTIFIER = "process-reboot-cause" SYSLOG_IDENTIFIER = "process-reboot-cause"
REBOOT_CAUSE_DIR = "/host/reboot-cause/" REBOOT_CAUSE_DIR = "/host/reboot-cause/"
REBOOT_CAUSE_FILE = REBOOT_CAUSE_DIR + "reboot-cause.txt" PREVIOUS_REBOOT_CAUSE_FILE = REBOOT_CAUSE_DIR + "previous-reboot-cause.json"
PREVIOUS_REBOOT_CAUSE_FILE = REBOOT_CAUSE_DIR + "previous-reboot-cause.txt"
FIRST_BOOT_PLATFORM_FILE = "/tmp/notify_firstboot_to_platform"
REBOOT_TYPE_KEXEC_FILE = "/proc/cmdline"
# The following SONIC_BOOT_TYPEs come from the warm/fast reboot script which is in sonic-utilities
# Because the system can be rebooted from some old versions, we have to take all possible BOOT options into consideration.
# On 201803, 201807 we have
# BOOT_OPTIONS="$(echo $KERNEL_OPTIONS | sed -e 's/\s*linux\s*/BOOT_IMAGE=/') fast-reboot"
# On 201811 and later we have
# BOOT_OPTIONS="$(echo $KERNEL_OPTIONS | sed -e 's/\s*linux\s*/BOOT_IMAGE=/') SONIC_BOOT_TYPE=${BOOT_TYPE_ARG}" where BOOT_TYPE_ARG can be warm, fastfast or fast
# To extract the commom part of them, we should have the following PATTERN
REBOOT_TYPE_KEXEC_PATTERN_WARM = ".*SONIC_BOOT_TYPE=(warm|fastfast).*"
REBOOT_TYPE_KEXEC_PATTERN_FAST = ".*SONIC_BOOT_TYPE=(fast|fast-reboot).*"
REBOOT_CAUSE_UNKNOWN = "Unknown" REBOOT_CAUSE_UNKNOWN = "Unknown"
@ -60,142 +49,27 @@ def log_error(msg):
# ============================= Functions ============================= # ============================= Functions =============================
def parse_warmfast_reboot_from_proc_cmdline():
if os.path.isfile(REBOOT_TYPE_KEXEC_FILE):
with open(REBOOT_TYPE_KEXEC_FILE, "r") as cause_file:
cause_file_kexec = cause_file.readline()
m = re.search(REBOOT_TYPE_KEXEC_PATTERN_WARM, cause_file_kexec)
if m and m.group(1):
return 'warm-reboot'
m = re.search(REBOOT_TYPE_KEXEC_PATTERN_FAST, cause_file_kexec)
if m and m.group(1):
return 'fast-reboot'
return None
def find_software_reboot_cause():
software_reboot_cause = REBOOT_CAUSE_UNKNOWN
if os.path.isfile(REBOOT_CAUSE_FILE):
with open(REBOOT_CAUSE_FILE, "r") as cause_file:
software_reboot_cause = cause_file.readline().rstrip('\n')
log_info("{} indicates the reboot cause: {}".format(REBOOT_CAUSE_FILE, software_reboot_cause))
else:
log_info("Reboot cause file {} not found".format(REBOOT_CAUSE_FILE))
if os.path.isfile(FIRST_BOOT_PLATFORM_FILE):
if software_reboot_cause == REBOOT_CAUSE_UNKNOWN:
version_info = sonic_device_util.get_sonic_version_info()
build_version = version_info['build_version'] if version_info else "unknown"
software_reboot_cause += " (First boot of SONiC version {})".format(build_version)
os.remove(FIRST_BOOT_PLATFORM_FILE)
return software_reboot_cause
def find_proc_cmdline_reboot_cause():
proc_cmdline_reboot_cause = parse_warmfast_reboot_from_proc_cmdline()
if proc_cmdline_reboot_cause:
log_info("/proc/cmdline indicates reboot type: {}".format(proc_cmdline_reboot_cause))
else:
log_info("No reboot cause found from /proc/cmdline")
return proc_cmdline_reboot_cause
def find_hardware_reboot_cause():
hardware_reboot_cause = None
# Until all platform vendors have provided sonic_platform packages,
# if there is no sonic_platform package installed, we only provide
# software-related reboot causes.
try:
import sonic_platform
platform = sonic_platform.platform.Platform()
chassis = platform.get_chassis()
hardware_reboot_cause_major, hardware_reboot_cause_minor = chassis.get_reboot_cause()
if hardware_reboot_cause_major == chassis.REBOOT_CAUSE_NON_HARDWARE:
# The reboot was not caused by hardware. If there is a REBOOT_CAUSE_FILE, it will
# contain any software-related reboot info. We will use it as the previous cause.
pass
elif hardware_reboot_cause_major == chassis.REBOOT_CAUSE_HARDWARE_OTHER:
hardware_reboot_cause = "{} ({})".format(hardware_reboot_cause_major, hardware_reboot_cause_minor)
else:
hardware_reboot_cause = hardware_reboot_cause_major
except ImportError as err:
log_warning("sonic_platform package not installed. Unable to detect hardware reboot causes.")
if hardware_reboot_cause:
log_info("Platform api indicates reboot cause {}".format(hardware_reboot_cause))
else:
log_info("No reboot cause found from platform api")
return hardware_reboot_cause
def main(): def main():
log_info("Starting up...") log_info("Logging reboot-cause...")
if not os.geteuid() == 0: if not os.geteuid() == 0:
log_error("User {} does not have permission to execute".format(pwd.getpwuid(os.getuid()).pw_name)) log_error("User {} does not have permission to execute".format(pwd.getpwuid(os.getuid()).pw_name))
sys.exit("This utility must be run as root") sys.exit("This utility must be run as root")
# Create REBOOT_CAUSE_DIR if it doesn't exist
if not os.path.exists(REBOOT_CAUSE_DIR):
os.makedirs(REBOOT_CAUSE_DIR)
# Remove stale PREVIOUS_REBOOT_CAUSE_FILE if it exists
if os.path.exists(PREVIOUS_REBOOT_CAUSE_FILE):
os.remove(PREVIOUS_REBOOT_CAUSE_FILE)
# Set a default previous reboot cause # Set a default previous reboot cause
previous_reboot_cause = REBOOT_CAUSE_UNKNOWN previous_reboot_cause = REBOOT_CAUSE_UNKNOWN
# 1. Check if the previous reboot was warm/fast reboot by testing whether there is "fast|fastfast|warm" in /proc/cmdline # Read the most recent reboot cause file and log data to syslog
proc_cmdline_reboot_cause = find_proc_cmdline_reboot_cause() if os.path.exists(PREVIOUS_REBOOT_CAUSE_FILE):
with open(PREVIOUS_REBOOT_CAUSE_FILE, "r") as last_cause_file:
data = json.load(last_cause_file)
if data['user']:
previous_reboot_cause = USER_ISSUED_REBOOT_CAUSE_REGEX.format(data['cause'], data['user'], data['time'])
else:
previous_reboot_cause = "{}".format(data['cause'])
# 2. Check if the previous reboot was caused by hardware # Log the last reboot cause to the syslog
# If yes, the hardware reboot cause will be treated as the reboot cause
hardware_reboot_cause = find_hardware_reboot_cause()
# 3. If there is a REBOOT_CAUSE_FILE, it will contain any software-related
# reboot info. We will use it as the previous cause.
software_reboot_cause = find_software_reboot_cause()
# The main decision logic of the reboot cause:
# If there is a reboot cause indicated by /proc/cmdline, it should be warmreboot/fastreboot
# the software_reboot_cause which is the content of /hosts/reboot-cause/reboot-cause.txt
# will be treated as the reboot cause
# Elif there is a reboot cause indicated by platform API,
# the hardware_reboot_cause will be treated as the reboot cause
# Else the software_reboot_cause will be treated as the reboot cause
if proc_cmdline_reboot_cause is not None:
previous_reboot_cause = software_reboot_cause
elif hardware_reboot_cause is not None:
previous_reboot_cause = hardware_reboot_cause
else:
previous_reboot_cause = software_reboot_cause
# Write the previous reboot cause to PREVIOUS_REBOOT_CAUSE_FILE
with open(PREVIOUS_REBOOT_CAUSE_FILE, "w") as prev_cause_file:
prev_cause_file.write(previous_reboot_cause)
# Also log the previous reboot cause to the syslog
log_info("Previous reboot cause: {}".format(previous_reboot_cause)) log_info("Previous reboot cause: {}".format(previous_reboot_cause))
# Remove the old REBOOT_CAUSE_FILE
if os.path.exists(REBOOT_CAUSE_FILE):
os.remove(REBOOT_CAUSE_FILE)
# Write a new default reboot cause file for the next reboot
with open(REBOOT_CAUSE_FILE, "w") as cause_file:
cause_file.write(REBOOT_CAUSE_UNKNOWN)
if __name__ == "__main__": if __name__ == "__main__":
main() main()