Add state and phys_state metrics to InfiniBand Integration (#20070)

* add state and phys_state parsing

* add test and tweak check

* changelog

* lint

* fix unit

Author: steveny91

infiniband/changelog.d/20070.added

@@ -0,0 +1 @@
+Add state and phys_state metrics

infiniband/datadog_checks/infiniband/check.py

@@ -7,7 +7,7 @@
 
 from datadog_checks.base import AgentCheck  # noqa: F401
 
-from .metrics import IB_COUNTERS, RDMA_COUNTERS
+from .metrics import IB_COUNTERS, RDMA_COUNTERS, STATUS_COUNTERS
 
 
 class InfinibandCheck(AgentCheck):
@@ -27,6 +27,7 @@ def __init__(self, name, init_config, instances):
         # Allow for specific counters to be excluded if configured
         self.exclude_counters = set(self.instance.get('exclude_counters', []))
         self.exclude_hw_counters = set(self.instance.get('exclude_hw_counters', []))
+        self.exclude_status_counters = set(self.instance.get('exclude_status_counters', []))
 
         # Allow for specific devices to be excluded if configured
         self.exclude_devices = set(self.instance.get('exclude_devices', []))
@@ -67,6 +68,7 @@ def _collect_counters(self, device, port):
 
         self._collect_counter_metrics(port_path, tags)
         self._collect_hw_counter_metrics(port_path, tags)
+        self._collect_status_metrics(port_path, tags)
 
     def _collect_counter_metrics(self, port_path, tags):
         counters_path = os.path.join(port_path, "counters")
@@ -94,6 +96,31 @@ def _collect_hw_counter_metrics(self, port_path, tags):
             ) and filename not in self.exclude_hw_counters:
                 self._submit_counter_metric(file, f"rdma.{filename}", tags)
 
+    def _collect_status_metrics(self, port_path, tags):
+        for status_file in STATUS_COUNTERS:
+            if status_file in self.exclude_status_counters:
+                self.log.debug("Skipping status counter %s as it is in the exclude list", status_file)
+                continue
+            file_path = os.path.join(port_path, status_file)
+            if os.path.exists(file_path):
+                with open(file_path, "r") as f:
+                    content = f.read().strip()
+                    # "4: ACTIVE" - split to get value and state
+                    parts = content.split(":", 1)
+                    value = int(parts[0].strip())
+                    metric_tags = list(tags)
+
+                    # Add state as a tag if it exists
+                    if len(parts) > 1:
+                        state = parts[1].strip()
+                        metric_tags.append(f"port_{status_file}:{state}")
+
+                    if self.collection_type in {'gauge', 'both'}:
+                        self.gauge(f"port_{status_file}", value, metric_tags)
+
+                    if self.collection_type in {'monotonic_count', 'both'}:
+                        self.monotonic_count(f"port_{status_file}.count", value, metric_tags)
+
     def _submit_counter_metric(self, file_path, metric_name, tags):
         with open(file_path, "r") as f:
             value = int(f.read().strip())

infiniband/datadog_checks/infiniband/metrics.py

@@ -99,3 +99,5 @@
     "rdma_read_resp_bytes",
     "rdma_read_bytes",
 }
+
+STATUS_COUNTERS = {"state", "phys_state"}  # "4: ACTIVE"  # "5: LinkUp"

infiniband/metadata.csv

@@ -13,6 +13,7 @@ infiniband.multicast_rcv_packets,gauge,,packet,,"Number of multicast packets,inc
 infiniband.multicast_rcv_packets.count,count,,packet,,Number of new multicast packets received since the last metric submission (legacy),0,infiniband,,,
 infiniband.multicast_xmit_packets,gauge,,packet,,Number of multicast packets transmitted on all VLs from the port (legacy),0,infiniband,,,
 infiniband.multicast_xmit_packets.count,count,,packet,,Number of new multicast packets transmitted since the last metric submission (legacy),0,infiniband,,,
+infiniband.phys_state,gauge,,,,Physical link state,0,infiniband,,,
 infiniband.port_multicast_rcv_packets,gauge,,packet,,Number of multicast packets received,0,infiniband,,,
 infiniband.port_multicast_rcv_packets.count,count,,packet,,Number of new multicast packets received since the last metric submission,0,infiniband,,,
 infiniband.port_multicast_xmit_packets,gauge,,packet,,Number of multicast packets transmitted on all VLs from the port,0,infiniband,,,
@@ -175,6 +176,7 @@ infiniband.rdma.tx_vport_unicast_bytes,gauge,,byte,,Number of unicast bytes tran
 infiniband.rdma.tx_vport_unicast_bytes.count,count,,byte,,Number of new unicast bytes transmitted on virtual port since the last metric submission,0,infiniband,,,
 infiniband.rdma.tx_vport_unicast_packets,gauge,,packet,,Number of unicast packets transmitted on virtual port,0,infiniband,,,
 infiniband.rdma.tx_vport_unicast_packets.count,count,,packet,,Number of new unicast packets transmitted on virtual port since the last metric submission,0,infiniband,,,
+infiniband.state,gauge,,,,Port state,0,infiniband,,,
 infiniband.symbol_error,gauge,,error,,Number of minor link errors detected on one or more physical lanes,0,infiniband,,,
 infiniband.symbol_error.count,count,,error,,Number of new minor link errors detected since the last metric submission,0,infiniband,,,
 infiniband.unicast_rcv_packets,gauge,,packet,,"Number of unicast packets,including unicast packets containing errors (legacy)",0,infiniband,,,

infiniband/tests/common.py

@@ -14,5 +14,10 @@
     'rx_bytes': '3000',
 }
 
+MOCK_STATUS_DATA = {
+    'state': '4: ACTIVE',
+    'phys_state': '5: LinkUp',
+}
+
 MOCK_DEVICE = 'mlx5_0'
 MOCK_PORT = '1'

infiniband/tests/conftest.py

@@ -11,6 +11,7 @@
     MOCK_IB_COUNTER_DATA,
     MOCK_PORT,
     MOCK_RDMA_COUNTER_DATA,
+    MOCK_STATUS_DATA,
 )
 
 
@@ -56,6 +57,8 @@ def _get_file_content(filename):
         return MOCK_IB_COUNTER_DATA[counter]
     elif counter in MOCK_RDMA_COUNTER_DATA:
         return MOCK_RDMA_COUNTER_DATA[counter]
+    elif counter in MOCK_STATUS_DATA:
+        return MOCK_STATUS_DATA[counter]
     return '0'
 
 

infiniband/tests/test_unit.py

@@ -11,6 +11,7 @@
     MOCK_IB_COUNTER_DATA,
     MOCK_PORT,
     MOCK_RDMA_COUNTER_DATA,
+    MOCK_STATUS_DATA,
 )
 
 
@@ -44,6 +45,20 @@ def test_check(aggregator, instance, mock_fs):
     _assert_metrics(aggregator, MOCK_IB_COUNTER_DATA, 'infiniband', tags)
     _assert_metrics(aggregator, MOCK_RDMA_COUNTER_DATA, 'infiniband.rdma', tags)
 
+    for status_name, status_value in MOCK_STATUS_DATA.items():
+        value, state_name = status_value.split(':', 1)
+        value = int(value.strip())
+        state_name = state_name.strip()
+
+        expected_tags = tags + [f'port_{status_name}:{state_name}']
+        aggregator.assert_metric(
+            f'infiniband.port_{status_name}',
+            metric_type=aggregator.GAUGE,
+            value=value,
+            tags=expected_tags,
+            count=1,
+        )
+
 
 @pytest.mark.parametrize(
     "collection_type,m_type,count",