WIP: mgr/devicehealth: added life_expectancy_response()

src/common/blkdev.cc

@@ -384,12 +384,10 @@ std::string get_device_id(const std::string& devname)
 
   udev = udev_new();
   if (!udev) {
-    //derr << "failed to run udev_new(), when calling for device " << devname << dendl;
     return {};
   }
   dev = udev_device_new_from_subsystem_sysname(udev, "block", devname.c_str());
   if (!dev) {
-    //derr << "failed to run udev_device_new_from_subsystem_sysname() for " << devname << dendl;
     udev_unref(udev);
     return {};
   }
@@ -405,15 +403,13 @@ std::string get_device_id(const std::string& devname)
   udev_unref(udev);
 
   if (!device_id.empty()) {
-    //dout << devname << " serial number: " << data << dendl;
     std::replace(device_id.begin(), device_id.end(), ' ', '_');
     return device_id;
   }
 
   // either udev_device_get_property_value() failed, or succeeded but
   // returned nothing; trying to read from files.  note that the 'vendor'
   // file rarely contains the actual vendor; it's usually 'ATA'.
-  //derr << "udev could not retrieve serial number of " << devname << dendl;
   std::string model, serial;
   model = get_block_device_string_property_wrap(devname, "device/model");
   serial = get_block_device_string_property_wrap(devname, "device/serial");
@@ -434,7 +430,6 @@ std::string get_block_device_string_property_wrap(const std::string &devname,
   std::string prop_val;
   int ret = get_block_device_string_property(devname.c_str(), property.c_str(), buff, sizeof(buff));
   if (ret < 0) {
-    //derr << "Could not retrieve content of " << property << " file of " << devname << dendl;
     return {};
   }
   prop_val = buff;

src/common/options.cc

@@ -4749,7 +4749,7 @@ std::vector<Option> get_global_options() {
     .set_description("Filesystem path to manager modules."),
 
     Option("mgr_initial_modules", Option::TYPE_STR, Option::LEVEL_BASIC)
-    .set_default("restful status balancer iostat")
+    .set_default("restful status balancer iostat devicehealth")
     .set_flag(Option::FLAG_NO_MON_UPDATE)
     .set_flag(Option::FLAG_CLUSTER_CREATE)
     .add_service("mon")

src/mgr/ActivePyModules.cc

@@ -281,10 +281,10 @@ PyObject *ActivePyModules::get_python(const std::string &what)
     return f.get();
   } else if (what == "pg_dump") {
     PyFormatter f;
-        cluster_state.with_pgmap(
-        [&f](const PGMap &pg_map) {
-	  pg_map.dump(&f);
-        }
+    cluster_state.with_pgmap(
+      [&f](const PGMap &pg_map) {
+	pg_map.dump(&f);
+      }
     );
     return f.get();
   } else if (what == "devices") {
@@ -295,6 +295,14 @@ PyObject *ActivePyModules::get_python(const std::string &what)
       });
     f.close_section();
     return f.get();
+  } else if (what.size() > 7 &&
+	     what.substr(0, 7) == "device ") {
+    string devid = what.substr(7);
+    PyFormatter f;
+    daemon_state.with_device(devid, [&f] (const DeviceState& dev) {
+	f.dump_object("device", dev);
+      });
+    return f.get();
   } else if (what == "io_rate") {
     PyFormatter f;
     cluster_state.with_pgmap(

src/osd/OSD.cc

@@ -2243,7 +2243,9 @@ will start to track new ops received afterwards.";
     }
     f->close_section();
   } else if (admin_command == "smart") {
-    probe_smart(ss);
+    string devid;
+    cmd_getval(cct, cmdmap, "devid", devid);
+    probe_smart(devid, ss);
   } else if (admin_command == "list_devices") {
     set<string> devnames;
     store->get_devices(&devnames);
@@ -6171,7 +6173,9 @@ void OSD::do_command(Connection *con, ceph_tid_t tid, vector<string>& cmd, buffe
   }
 
   else if (prefix == "smart") {
-    probe_smart(ds);
+    string devid;
+    cmd_getval(cct, cmdmap, "devid", devid);
+    probe_smart(devid, ds);
   }
 
   else {
@@ -6192,14 +6196,15 @@ void OSD::do_command(Connection *con, ceph_tid_t tid, vector<string>& cmd, buffe
   }
 }
 
-void OSD::probe_smart(ostream& ss)
+void OSD::probe_smart(const string& only_devid, ostream& ss)
 {
   set<string> devnames;
   store->get_devices(&devnames);
-  uint64_t smart_timeout = cct->_conf->get_val<uint64_t>("osd_smart_report_timeout");
-  std::string result;
+  uint64_t smart_timeout = cct->_conf->get_val<uint64_t>(
+    "osd_smart_report_timeout");
 
-  json_spirit::mObject json_map; // == typedef std::map<std::string, mValue> mObject;
+  // == typedef std::map<std::string, mValue> mObject;
+  json_spirit::mObject json_map;
   json_spirit::mValue smart_json;
 
   for (auto dev : devnames) {
@@ -6208,16 +6213,28 @@ void OSD::probe_smart(ostream& ss)
       continue;
     }
 
+    string devid = get_device_id(dev);
+    if (devid.size() == 0) {
+      dout(10) << __func__ << " no unique id for dev " << dev << ", skipping"
+	       << dendl;
+      continue;
+    }
+    if (only_devid.size() && devid != only_devid) {
+      continue;
+    }
+
+    std::string result;
     if (probe_smart_device(("/dev/" + dev).c_str(), smart_timeout, &result)) {
       dout(10) << "probe_smart_device failed for /dev/" << dev << dendl;
-      continue;
+      //continue;
+      result = "{\"error\": \"smartctl failed\", \"dev\": \"" + dev + "\"}";
     }
 
     // TODO: change to read_or_throw?
     if (!json_spirit::read(result, smart_json)) {
       derr << "smartctl JSON output of /dev/" + dev + " is invalid" << dendl;
     } else { //json is valid, assigning
-      json_map[dev] = smart_json;
+      json_map[devid] = smart_json;
     }
     // no need to result.clear() or clear smart_json
   }
@@ -6227,8 +6244,9 @@ void OSD::probe_smart(ostream& ss)
 int OSD::probe_smart_device(const char *device, int timeout, std::string *result)
 {
   // when using --json, smartctl will report its errors in JSON format to stdout 
-  SubProcessTimed smartctl("sudo", SubProcess::CLOSE, SubProcess::PIPE, SubProcess::CLOSE,
-			   timeout);
+  SubProcessTimed smartctl(
+    "sudo", SubProcess::CLOSE, SubProcess::PIPE, SubProcess::CLOSE,
+    timeout);
   smartctl.add_cmd_args(
     "smartctl",
     "-a",

src/osd/OSD.h

@@ -2208,7 +2208,7 @@ class OSD : public Dispatcher,
 
   float get_osd_recovery_sleep();
 
-  void probe_smart(ostream& ss);
+  void probe_smart(const string& devid, ostream& ss);
   int probe_smart_device(const char *device, int timeout, std::string *result);
 
 public:

src/pybind/mgr/devicehealth/__init__.py

@@ -0,0 +1,2 @@
+
+from .module import Module