Add SSD Health CLI utility (sonic-net#587)

* Add SSD Health CLI utility

Signed-off-by: Andriy Moroz <c_andriym@mellanox.com>

* Add SSD Health command reference

* Rename ssdutility entrypoint to ssdutil

Signed-off-by: Andriy Moroz <c_andriym@mellanox.com>

* Fix - do not add units if value is N/A

Signed-off-by: Andriy Moroz <c_andriym@mellanox.com>

doc/Command-Reference.md

@@ -598,7 +598,19 @@ Couple of example outputs are given below.
 	(checksum valid)
   ```
 
+**show platform ssdhealth**  
+This command displays health parameters of the device's SSD
 
+- Usage:  
+  show platform ssdhealth [--verbose, --vendor]
+
+- Example:
+  ```
+  root@arc-switch1029:/home/admin# show platform ssdhealth
+  Device Model : M.2 (S42) 3IE3
+  Health       : 99.665%
+  Temperature  : 30C
+  ```
 
 **show platform psustatus**
 This command displays the status of the device's power supply units

setup.py

@@ -38,6 +38,7 @@
         'debug',
         'pfcwd',
         'sfputil',
+        'ssdutil',
         'pfc',
         'psuutil',
         'show',
@@ -100,6 +101,7 @@
             'debug = debug.main:cli',
             'pfcwd = pfcwd.main:cli',
             'sfputil = sfputil.main:cli',
+            'ssdutil = ssdutil.main:ssdutil',
             'pfc = pfc.main:cli',
             'psuutil = psuutil.main:cli',
             'show = show.main:cli',

show/main.py

@@ -1258,6 +1258,20 @@ def psustatus(index, verbose):
 
     run_command(cmd, display_cmd=verbose)
 
+# 'ssdhealth' subcommand ("show platform ssdhealth [--verbose/--vendor]")
+@platform.command()
+@click.argument('device', required=False)
+@click.option('--verbose', is_flag=True, help="Enable verbose output")
+@click.option('--vendor', is_flag=True, help="Enable vendor specific output")
+def ssdhealth(device, verbose, vendor):
+    """Show SSD Health information"""
+    if not device:
+        device = os.popen("lsblk -o NAME,TYPE -p | grep disk").readline().strip().split()[0]
+    cmd = "ssdutil -d " + device
+    options = " -v" if verbose else ""
+    options += " -e" if vendor else ""
+    run_command(cmd + options, display_cmd=verbose)
+
 #
 # 'logging' command ("show logging")
 #

ssdutil/main.py

@@ -0,0 +1,130 @@
+#!/usr/bin/env python
+#
+# main.py
+#
+# Command-line utility to check SSD health and parameters
+#
+
+try:
+    import sys
+    import os
+    import subprocess
+    import argparse
+    import syslog
+except ImportError as e:
+    raise ImportError("%s - required module not found" % str(e))
+
+DEFAULT_DEVICE="/dev/sda"
+SYSLOG_IDENTIFIER = "ssdutil"
+
+PLATFORM_ROOT_PATH = '/usr/share/sonic/device'
+SONIC_CFGGEN_PATH = '/usr/local/bin/sonic-cfggen'
+HWSKU_KEY = 'DEVICE_METADATA.localhost.hwsku'
+PLATFORM_KEY = 'DEVICE_METADATA.localhost.platform'
+
+def syslog_msg(severity, msg, stdout=False):
+    """
+    Prints to syslog (and stdout if needed) message with specified severity
+
+    Args:
+        severity : message severity
+        msg      : message
+        stdout   : also primt message to stdout
+
+    """
+    syslog.openlog(SYSLOG_IDENTIFIER)
+    syslog.syslog(severity, msg)
+    syslog.closelog()
+
+    if stdout:
+        print msg
+
+def get_platform_and_hwsku():
+    """
+    Retrieves current platform name and hwsku
+    Raises an OSError exception when failed to fetch
+
+    Returns:
+        tuple of strings platform and hwsku
+        e.g. ("x86_64-mlnx_msn2700-r0", "ACS-MSN2700")
+    """
+    try:
+        proc = subprocess.Popen([SONIC_CFGGEN_PATH, '-H', '-v', PLATFORM_KEY],
+                                stdout=subprocess.PIPE,
+                                shell=False,
+                                stderr=subprocess.STDOUT)
+        stdout = proc.communicate()[0]
+        proc.wait()
+        platform = stdout.rstrip('\n')
+
+        proc = subprocess.Popen([SONIC_CFGGEN_PATH, '-d', '-v', HWSKU_KEY],
+                                stdout=subprocess.PIPE,
+                                shell=False,
+                                stderr=subprocess.STDOUT)
+        stdout = proc.communicate()[0]
+        proc.wait()
+        hwsku = stdout.rstrip('\n')
+    except OSError, e:
+        raise OSError("Cannot detect platform")
+
+    return (platform, hwsku)
+
+def import_ssd_api(diskdev):
+    """
+    Loads platform specific or generic ssd_util module from source
+    Raises an ImportError exception if none of above available
+
+    Returns:
+        Instance of the class with SSD API implementation (vendor or generic)
+    """
+
+    # Get platform and hwsku
+    (platform, hwsku) = get_platform_and_hwsku()
+
+    # try to load platform specific module
+    try:
+        hwsku_plugins_path = "/".join([PLATFORM_ROOT_PATH, platform, "plugins"])
+        sys.path.append(os.path.abspath(hwsku_plugins_path))
+        from ssd_util import SsdUtil
+    except ImportError as e:
+        syslog_msg(syslog.LOG_WARNING, "Platform specific SsdUtil module not found. Falling down to the generic implementation")
+        try:
+            from sonic_platform_base.sonic_ssd.ssd_generic import SsdUtil
+        except ImportError as e:
+            syslog_msg(syslog.LOG_ERR, "Failed to import default SsdUtil. Error: {}".format(str(e)), True)
+            raise e
+
+    return SsdUtil(diskdev)
+
+def is_number(s):
+    try:
+        float(s)
+        return True
+    except ValueError:
+        return False
+
+# ==================== Entry point ====================
+def ssdutil():
+    if os.geteuid() != 0:
+        print "Root privileges are required for this operation"
+        sys.exit(1)
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("-d", "--device", help="Device name to show health info", default=DEFAULT_DEVICE)
+    parser.add_argument("-v", "--verbose", action="store_true", default=False, help="Show verbose output (some additional parameters)")
+    parser.add_argument("-e", "--vendor", action="store_true", default=False, help="Show vendor output (extended output if provided by platform vendor)")
+    args = parser.parse_args()
+
+    ssd = import_ssd_api(args.device)
+
+    print "Device Model : {}".format(ssd.get_model())
+    if args.verbose:
+        print "Firmware     : {}".format(ssd.get_firmware())
+        print "Serial       : {}".format(ssd.get_serial())
+    print "Health       : {}{}".format(ssd.get_health(),      "%" if is_number(ssd.get_health()) else "")
+    print "Temperature  : {}{}".format(ssd.get_temperature(), "C" if is_number(ssd.get_temperature()) else "")
+    if args.vendor:
+        print ssd.get_vendor_output()
+
+if __name__ == '__main__':
+    ssdutil()