From 726fecaf8b1dd95df7d77e71b165c147c47f4f65 Mon Sep 17 00:00:00 2001 From: Stephen Sun <5379172+stephenxs@users.noreply.github.com> Date: Wed, 12 Feb 2020 01:54:12 +0800 Subject: [PATCH] [process-reboot-cause]Clean up the process-reboot-cause as reqired in issue 3927 (#4128) --- .../process-reboot-cause/process-reboot-cause | 115 +++++++++++------- 1 file changed, 71 insertions(+), 44 deletions(-) diff --git a/files/image_config/process-reboot-cause/process-reboot-cause b/files/image_config/process-reboot-cause/process-reboot-cause index e5d228b4c6..409deb7d68 100755 --- a/files/image_config/process-reboot-cause/process-reboot-cause +++ b/files/image_config/process-reboot-cause/process-reboot-cause @@ -62,10 +62,10 @@ 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.match(REBOOT_TYPE_KEXEC_PATTERN_WARM, cause_file_kexec) + m = re.search(REBOOT_TYPE_KEXEC_PATTERN_WARM, cause_file_kexec) if m and m.group(1): return 'warm-reboot' - m = re.match(REBOOT_TYPE_KEXEC_PATTERN_FAST, cause_file_kexec) + m = re.search(REBOOT_TYPE_KEXEC_PATTERN_FAST, cause_file_kexec) if m and m.group(1): return 'fast-reboot' return None @@ -77,13 +77,61 @@ def find_software_reboot_cause(): 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): 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(): log_info("Starting up...") @@ -99,54 +147,33 @@ def main(): if os.path.exists(PREVIOUS_REBOOT_CAUSE_FILE): os.remove(PREVIOUS_REBOOT_CAUSE_FILE) - # Set a default previous reboot cause previous_reboot_cause = UNKNOWN_REBOOT_CAUSE - # 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 + # 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() - # 1. Check if the previous reboot was warm/fast reboot by testing whether there is "fast|fastfast|warm" in /proc/cmdline - # If yes, the content of /hosts/reboot-cause/reboot-cause.txt will be treated as the 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)) - if os.path.isfile(REBOOT_CAUSE_FILE): - with open(REBOOT_CAUSE_FILE, "r") as cause_file: - proc_cmdline_reboot_cause = cause_file.readline().rstrip('\n') - else: - # /proc/cmdline says it's a warm/fast reboot but /host/reboot-cause.txt doesn't exist. - # This could happen when upgrading from a version doesn't support reboot cause. - log_info("Reboot cause file {} doesn't exist".format(REBOOT_CAUSE_DIR)) + # 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() - if proc_cmdline_reboot_cause is not None: - previous_reboot_cause = proc_cmdline_reboot_cause - else: - # 2. Check if the previous reboot was caused by hardware - # If yes, the hardware reboot cause will be treated as the reboot cause - platform = sonic_platform.platform.Platform() + # 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() - chassis = platform.get_chassis() - - hardware_reboot_cause, optional_details = chassis.get_reboot_cause() - - if hardware_reboot_cause == 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. - previous_reboot_cause = find_software_reboot_cause() - elif hardware_reboot_cause == chassis.REBOOT_CAUSE_HARDWARE_OTHER: - previous_reboot_cause = "{} ({})".format(hardware_reboot_cause, optional_details) - else: - previous_reboot_cause = hardware_reboot_cause - except ImportError as err: - log_warning("sonic_platform package not installed. Unable to detect hardware reboot causes.") - - # If there is a REBOOT_CAUSE_FILE, it will contain any software-related - # reboot info. We will use it as the previous cause. - previous_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: