SNMP discovery updates continued

worker/app/collectors/snmp.py

@@ -58,6 +58,10 @@ ENT_PHY_SENSOR_SCALE = (1, 3, 6, 1, 2, 1, 99, 1, 1, 1, 2)
 ENT_PHY_SENSOR_PRECISION = (1, 3, 6, 1, 2, 1, 99, 1, 1, 1, 3)
 ENT_PHY_SENSOR_VALUE = (1, 3, 6, 1, 2, 1, 99, 1, 1, 1, 4)
 ENT_PHY_SENSOR_OPER_STATUS = (1, 3, 6, 1, 2, 1, 99, 1, 1, 1, 5)
+UCD_LA_LOAD_INT = (1, 3, 6, 1, 4, 1, 2021, 10, 1, 5)
+UCD_MEM_TOTAL_REAL = (1, 3, 6, 1, 4, 1, 2021, 4, 5, 0)
+UCD_MEM_AVAIL_REAL = (1, 3, 6, 1, 4, 1, 2021, 4, 6, 0)
+UCD_DSK_PERCENT = (1, 3, 6, 1, 4, 1, 2021, 9, 1, 9)
 
 STATUS_LABELS = {
     1: "up",
@@ -128,6 +132,61 @@ def _run_snmp_check_sync(config: SnmpCheckConfig) -> SnmpCheckResult:
             metrics=[SnmpMetricValue(name="load_percent", value=float(value), unit="%")],
         )
 
+    if config.item_type == "linux_load_average":
+        load_index = _item_index(config.item_id, "linux", position=2)
+        if load_index is None:
+            return SnmpCheckResult(status="down", response_time_ms=0, message="SNMP Linux load item was not valid")
+        oid = _with_index(UCD_LA_LOAD_INT, load_index)
+        value = _int_value(client.get_many([oid]).get(oid))
+        response_time_ms = int((perf_counter() - started) * 1000)
+        if value is None:
+            return SnmpCheckResult(status="down", response_time_ms=response_time_ms, message="Linux load average was not reported")
+        load_average = value / 100
+        label = config.label or "Load average"
+        return SnmpCheckResult(
+            status="up",
+            response_time_ms=response_time_ms,
+            message=f"{label} is {load_average:.2f}",
+            metrics=[SnmpMetricValue(name="load_average", value=load_average)],
+        )
+
+    if config.item_type == "linux_memory_usage":
+        values = client.get_many([UCD_MEM_TOTAL_REAL, UCD_MEM_AVAIL_REAL])
+        response_time_ms = int((perf_counter() - started) * 1000)
+        total_kb = _int_value(values.get(UCD_MEM_TOTAL_REAL))
+        available_kb = _int_value(values.get(UCD_MEM_AVAIL_REAL))
+        if not total_kb or available_kb is None:
+            return SnmpCheckResult(status="down", response_time_ms=response_time_ms, message="Linux memory was not reported")
+        used_kb = total_kb - available_kb
+        used_percent = (used_kb / total_kb) * 100
+        return SnmpCheckResult(
+            status="up",
+            response_time_ms=response_time_ms,
+            message=f"Memory is {used_percent:.1f}% used",
+            metrics=[
+                SnmpMetricValue(name="used_percent", value=used_percent, unit="%"),
+                SnmpMetricValue(name="used_bytes", value=float(used_kb * 1024), unit="bytes"),
+                SnmpMetricValue(name="total_bytes", value=float(total_kb * 1024), unit="bytes"),
+            ],
+        )
+
+    if config.item_type == "linux_disk_usage":
+        disk_index = _item_index(config.item_id, "linux", position=2)
+        if disk_index is None:
+            return SnmpCheckResult(status="down", response_time_ms=0, message="SNMP Linux disk item was not valid")
+        oid = _with_index(UCD_DSK_PERCENT, disk_index)
+        used_percent = _int_value(client.get_many([oid]).get(oid))
+        response_time_ms = int((perf_counter() - started) * 1000)
+        if used_percent is None:
+            return SnmpCheckResult(status="down", response_time_ms=response_time_ms, message="Linux disk usage was not reported")
+        label = config.label or "Disk"
+        return SnmpCheckResult(
+            status="up",
+            response_time_ms=response_time_ms,
+            message=f"{label} is {used_percent}% used",
+            metrics=[SnmpMetricValue(name="used_percent", value=float(used_percent), unit="%")],
+        )
+
     if config.item_type in {"memory_usage", "storage_usage"}:
         storage_index = _item_index(config.item_id, "storage")
         if storage_index is None:
@@ -276,12 +335,12 @@ def _interface_index(item_id: str) -> int | None:
     return _item_index(item_id, "interface")
 
 
-def _item_index(item_id: str, expected_prefix: str) -> int | None:
+def _item_index(item_id: str, expected_prefix: str, position: int = 1) -> int | None:
     parts = item_id.split(".")
-    if len(parts) < 3 or parts[0] != expected_prefix:
+    if len(parts) <= position or parts[0] != expected_prefix:
         return None
     try:
-        return int(parts[1])
+        return int(parts[position])
     except ValueError:
         return None
 

worker/tests/test_snmp_collector.py

@@ -20,6 +20,10 @@ from app.collectors.snmp import (
     HR_STORAGE_SIZE,
     HR_STORAGE_USED,
     SYS_UPTIME,
+    UCD_DSK_PERCENT,
+    UCD_LA_LOAD_INT,
+    UCD_MEM_AVAIL_REAL,
+    UCD_MEM_TOTAL_REAL,
     SnmpCheckConfig,
     _with_index,
     run_snmp_check,
@@ -66,6 +70,72 @@ class SnmpCollectorTestCase(unittest.IsolatedAsyncioTestCase):
             ("load_percent", 42.0, "%")
         ]
 
+    async def test_collects_linux_load_average(self) -> None:
+        oid = _with_index(UCD_LA_LOAD_INT, 1)
+        with patch("app.collectors.snmp.SnmpV2Client") as client_class:
+            client_class.return_value.get_many.return_value = {oid: 123}
+
+            result = await run_snmp_check(
+                SnmpCheckConfig(
+                    host="192.0.2.10",
+                    community="private-community",
+                    item_id="linux.load.1",
+                    item_type="linux_load_average",
+                    label="Load average 1 minute",
+                )
+            )
+
+        assert result.status == "up"
+        assert result.message == "Load average 1 minute is 1.23"
+        assert [(metric.name, metric.value, metric.unit) for metric in result.metrics] == [
+            ("load_average", 1.23, None)
+        ]
+
+    async def test_collects_linux_memory_usage(self) -> None:
+        with patch("app.collectors.snmp.SnmpV2Client") as client_class:
+            client_class.return_value.get_many.return_value = {
+                UCD_MEM_TOTAL_REAL: 1000,
+                UCD_MEM_AVAIL_REAL: 250,
+            }
+
+            result = await run_snmp_check(
+                SnmpCheckConfig(
+                    host="192.0.2.10",
+                    community="private-community",
+                    item_id="linux.memory.real",
+                    item_type="linux_memory_usage",
+                )
+            )
+
+        assert result.status == "up"
+        assert result.message == "Memory is 75.0% used"
+        assert [(metric.name, metric.value, metric.unit) for metric in result.metrics] == [
+            ("used_percent", 75.0, "%"),
+            ("used_bytes", 768000.0, "bytes"),
+            ("total_bytes", 1024000.0, "bytes"),
+        ]
+
+    async def test_collects_linux_disk_usage(self) -> None:
+        oid = _with_index(UCD_DSK_PERCENT, 31)
+        with patch("app.collectors.snmp.SnmpV2Client") as client_class:
+            client_class.return_value.get_many.return_value = {oid: 81}
+
+            result = await run_snmp_check(
+                SnmpCheckConfig(
+                    host="192.0.2.10",
+                    community="private-community",
+                    item_id="linux.disk.31",
+                    item_type="linux_disk_usage",
+                    label="Disk / usage",
+                )
+            )
+
+        assert result.status == "up"
+        assert result.message == "Disk / usage is 81% used"
+        assert [(metric.name, metric.value, metric.unit) for metric in result.metrics] == [
+            ("used_percent", 81.0, "%")
+        ]
+
     async def test_collects_storage_usage(self) -> None:
         oids = [
             _with_index(HR_STORAGE_ALLOCATION_UNITS, 31),