Add support for drives connected by MegaRAID (#47)

* Add support for drives connected by MegaRAID

* Fixed strict handling of errors generated by MegaRAID

* Separate MegaRAID-related logic to megaraid.py

* Add get_megaraid_device_type

* Added to README an example of a device connected by MegaRAID

Dockerfile

@@ -8,7 +8,7 @@ RUN apk add --no-cache smartmontools \
     && rm -rf /root/.cache/ \
     && find / -name '*.pyc' -delete
 
-COPY ./smartprom.py /smartprom.py
+COPY ./smartprom.py /megaraid.py /
 
 EXPOSE 9902
 ENTRYPOINT ["/usr/local/bin/python", "-u", "/smartprom.py"]

README.md

@@ -13,7 +13,7 @@ _Note: You don't have to do this if you use the Docker image._
 1. Copy the `smartprom.service` file into `/etc/systemd/system` folder.
 2. Copy the `smartprom.py` file anywhere into your system.
 3. Modify `ExecStart=` in the `smartprom.service` so that it points to `smartprom.py` in your system.
-4. Run `chmod +x smartprom.py` 
+4. Run `chmod +x smartprom.py`
 5. Install `prometheus_client` for the root user, example: `sudo -H python3 -m pip install prometheus_client`
 6. Run `systemctl enable smartprom` and `systemctl start smartprom`
 7. Your metrics will now be available at `http://localhost:9902`
@@ -56,6 +56,13 @@ smartprom_airflow_temperature_cel_raw{drive="/dev/sda",model_family="Seagate Bar
 ...
 ```
 
+If you are using a MegaRAID card to connect the drives, the metrics will export look like these:
+
+```shell
+smartprom_power_on_hours_raw{drive="megaraid,0",model_family="Western Digital Ultrastar He10/12",model_name="WDC WD80EMAZ-00M9AA0",serial_number="XXXXXXXX",type="sat"} 28522.0
+smartprom_power_on_time_hours{drive="megaraid,1",model_family="Unknown",model_name="HGST HUH728080AL5200",serial_number="XXXXXXXX",type="scsi"} 37341.0
+```
+
 ## Configuration
 
 All configuration is done with environment variables.

megaraid.py

@@ -0,0 +1,89 @@
+import json
+import re
+
+import smartprom
+
+MEGARAID_TYPE_PATTERN = r"(sat\+)?(megaraid,\d+)"
+
+
+def get_megaraid_device_info(dev: str, typ: str) -> dict:
+    """
+    Get device information connected with MegaRAID,
+    and process the information into get_device_info compatible format.
+    """
+    megaraid_id = get_megaraid_device_id(typ)
+    if megaraid_id is None:
+        return {}
+
+    results, _ = smartprom.run_smartctl_cmd(
+        ["smartctl", "-i", "--json=c", "-d", megaraid_id, dev]
+    )
+    results = json.loads(results)
+    serial_number = results.get("serial_number", "Unknown")
+    model_family = results.get("model_family", "Unknown")
+
+    # When using SAS drive and smartmontools r5286 and later,
+    # scsi_ prefix is added to model_name field.
+    # https://sourceforge.net/p/smartmontools/code/5286/
+    model_name = results.get(
+        "scsi_model_name",
+        results.get("model_name", "Unknown"),
+    )
+
+    return {
+        "model_family": model_family,
+        "model_name": model_name,
+        "serial_number": serial_number,
+    }
+
+
+def get_megaraid_device_type(dev: str, typ: str) -> str:
+    megaraid_id = get_megaraid_device_id(typ)
+    if megaraid_id is None:
+        return "unknown"
+
+    results, _ = smartprom.run_smartctl_cmd(
+        ["smartctl", "-i", "--json=c", "-d", megaraid_id, dev]
+    )
+    results = json.loads(results)
+
+    if "device" not in results or "protocol" not in results["device"]:
+        return "unknown"
+    return "sat" if results["device"]["protocol"] == "ATA" else "scsi"
+
+
+def get_megaraid_device_id(typ: str) -> str | None:
+    """
+    Returns the device ID on the MegaRAID from the typ string
+    """
+    megaraid_match = re.search(MEGARAID_TYPE_PATTERN, typ)
+    if not megaraid_match:
+        return None
+
+    return megaraid_match.group(2)
+
+
+def smart_megaraid(dev: str, megaraid_id: str) -> dict:
+    """
+    Runs the smartctl command on device connected by MegaRAID
+    and processes its attributes
+    """
+    results, exit_code = smartprom.run_smartctl_cmd(
+        ["smartctl", "-A", "-H", "-d", megaraid_id, "--json=c", dev]
+    )
+    results = json.loads(results)
+
+    if results["device"]["protocol"] == "ATA":
+        # SATA device on MegaRAID
+        data = results["ata_smart_attributes"]["table"]
+        attributes = smartprom.table_to_attributes_sat(data)
+        attributes["smart_passed"] = (0, smartprom.get_smart_status(results))
+        attributes["exit_code"] = (0, exit_code)
+        return attributes
+    elif results["device"]["protocol"] == "SCSI":
+        # SAS device on MegaRAID
+        attributes = smartprom.results_to_attributes_scsi(results)
+        attributes["smart_passed"] = smartprom.get_smart_status(results)
+        attributes["exit_code"] = exit_code
+        return attributes
+    return {}

smartprom.py

@@ -3,10 +3,13 @@
 import os
 import subprocess
 import time
+import re
 from typing import Tuple
 
 import prometheus_client
 
+import megaraid
+
 LABELS = ['drive', 'type', 'model_family', 'model_name', 'serial_number']
 DRIVES = {}
 METRICS = {}
@@ -42,12 +45,34 @@ def get_drives() -> dict:
     disks = {}
     result, _ = run_smartctl_cmd(['smartctl', '--scan-open', '--json=c'])
     result_json = json.loads(result)
+
+    # Ignore devices that fail on open, such as Virtual Drives created by MegaRAID.
+    result_json["devices"] = list(
+        filter(
+            lambda x: (
+                x.get("open_error", "")
+                != "DELL or MegaRaid controller, please try adding '-d megaraid,N'"
+            ),
+            result_json["devices"],
+        )
+    )
+
     if 'devices' in result_json:
         devices = result_json['devices']
         for device in devices:
             dev = device["name"]
-            disk_attrs = get_device_info(dev)
-            disk_attrs["type"] = device["type"]
+            if re.match(megaraid.MEGARAID_TYPE_PATTERN, device["type"]):
+                # If drive is connected by MegaRAID, dev has a bus name like "/dev/bus/0".
+                # After retrieving the disk information using the bus name,
+                # replace dev with a disk ID such as "megaraid,0".
+                disk_attrs = megaraid.get_megaraid_device_info(dev, device["type"])
+                disk_attrs["type"] = megaraid.get_megaraid_device_type(dev, device["type"])
+                disk_attrs["bus_device"] = dev
+                disk_attrs["megaraid_id"] = megaraid.get_megaraid_device_id(device["type"])
+                dev = disk_attrs["megaraid_id"]
+            else:
+                disk_attrs = get_device_info(dev)
+                disk_attrs["type"] = device["type"]
             disks[dev] = disk_attrs
             print("Discovered device", dev, "with attributes", disk_attrs)
     else:
@@ -85,11 +110,18 @@ def smart_sat(dev: str) -> dict:
     results, exit_code = run_smartctl_cmd(['smartctl', '-A', '-H', '-d', 'sat', '--json=c', dev])
     results = json.loads(results)
 
-    attributes = {
-        'smart_passed': (0, get_smart_status(results)),
-        'exit_code': (0, exit_code)
-    }
-    data = results['ata_smart_attributes']['table']
+    attributes = table_to_attributes_sat(results["ata_smart_attributes"]["table"])
+    attributes["smart_passed"] = (0, get_smart_status(results))
+    attributes["exit_code"] = (0, exit_code)
+    return attributes
+
+
+def table_to_attributes_sat(data: dict) -> dict:
+    """
+    Returns a results["ata_smart_attributes"]["table"]
+    processed into an attributes dict
+    """
+    attributes = {}
     for metric in data:
         code = metric['id']
         name = metric['name']
@@ -146,11 +178,19 @@ def smart_scsi(dev: str) -> dict:
     results, exit_code = run_smartctl_cmd(['smartctl', '-A', '-H', '-d', 'scsi', '--json=c', dev])
     results = json.loads(results)
 
-    attributes = {
-        'smart_passed': get_smart_status(results),
-        'exit_code': exit_code
-    }
-    for key, value in results.items():
+    attributes = results_to_attributes_scsi(results)
+    attributes["smart_passed"] = get_smart_status(results)
+    attributes["exit_code"] = exit_code
+    return attributes
+
+
+def results_to_attributes_scsi(data: dict) -> dict:
+    """
+    Returns the result of smartctl -i on the SCSI device
+    processed into an attributes dict
+    """
+    attributes = {}
+    for key, value in data.items():
         if type(value) == dict:
             for _label, _value in value.items():
                 if type(_value) == int:
@@ -169,7 +209,11 @@ def collect():
     for drive, drive_attrs in DRIVES.items():
         typ = drive_attrs['type']
         try:
-            if typ in SAT_TYPES:
+            if "megaraid_id" in drive_attrs:
+                attrs = megaraid.smart_megaraid(
+                    drive_attrs["bus_device"], drive_attrs["megaraid_id"]
+                )
+            elif typ in SAT_TYPES:
                 attrs = smart_sat(drive)
             elif typ in NVME_TYPES:
                 attrs = smart_nvme(drive)