From 2965ddf0eecf1c654472b0254c2437f30fd179a8 Mon Sep 17 00:00:00 2001
From: lnhanks <67074258+lnhanks@users.noreply.github.com>
Date: Wed, 13 Sep 2023 07:35:03 -0700
Subject: [PATCH] Only metrics (#2557)

Prometheus metrics for capturing ENI IP usage and no available IP address errors

Co-authored-by: Lindsay Hanks <lnhanks@dev-dsk-lnhanks-2a-167bac85.us-west-2.amazon.com>
---
 pkg/ipamd/datastore/data_store.go | 31 +++++++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)

diff --git a/pkg/ipamd/datastore/data_store.go b/pkg/ipamd/datastore/data_store.go
index 410ba9f3c6..8adc2965ab 100644
--- a/pkg/ipamd/datastore/data_store.go
+++ b/pkg/ipamd/datastore/data_store.go
@@ -122,6 +122,19 @@ var (
 		},
 		[]string{"cidr"},
 	)
+	noAvailableIPAddrs = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Name: "awscni_err_no_avail_addrs",
+			Help: "The number of pod IP assignments that fail due to no available IP addresses",
+		},
+	)
+	eniIPsInUse = prometheus.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Name: "awscni_eni_util",
+			Help: "The number of allocated ips partitioned by eni",
+		},
+		[]string{"eni"},
+	)
 	prometheusRegistered = false
 )
 
@@ -344,6 +357,8 @@ func prometheusRegister() {
 		prometheus.MustRegister(forceRemovedIPs)
 		prometheus.MustRegister(totalPrefixes)
 		prometheus.MustRegister(ipsPerCidr)
+		prometheus.MustRegister(noAvailableIPAddrs)
+		prometheus.MustRegister(eniIPsInUse)
 		prometheusRegistered = true
 	}
 }
@@ -437,6 +452,8 @@ func (ds *DataStore) ReadBackingStore(isv6Enabled bool) error {
 					cidr.IPAddresses[ipAddr.String()] = addr
 					ds.assignPodIPAddressUnsafe(addr, allocation.IPAMKey, allocation.Metadata, time.Unix(0, allocation.AllocationTimestamp))
 					ds.log.Debugf("Recovered %s => %s/%s", allocation.IPAMKey, eni.ID, addr.Address)
+					// Increment ENI IP usage upon finding assigned ips
+					eniIPsInUse.WithLabelValues(eni.ID).Inc()
 					// Update prometheus for ips per cidr
 					// Secondary IP mode will have /32:1 and Prefix mode will have /28:<number of /32s>
 					ipsPerCidr.With(prometheus.Labels{"cidr": cidr.Cidr.String()}).Inc()
@@ -522,6 +539,8 @@ func (ds *DataStore) AddENI(eniID string, deviceNumber int, isPrimary, isTrunk,
 		AvailableIPv4Cidrs: make(map[string]*CidrInfo)}
 
 	enis.Set(float64(len(ds.eniPool)))
+	// Initialize ENI IPs In Use to 0 when an ENI is created
+	eniIPsInUse.WithLabelValues(eniID).Set(0)
 	return nil
 }
 
@@ -711,9 +730,12 @@ func (ds *DataStore) AssignPodIPv6Address(ipamKey IPAMKey, ipamMetadata IPAMMeta
 				delete(V6Cidr.IPAddresses, addr.Address)
 				return "", -1, err
 			}
+			// Increment ENI IP usage on pod IPv6 allocation
+			eniIPsInUse.WithLabelValues(eni.ID).Inc()
 			return addr.Address, eni.DeviceNumber, nil
 		}
 	}
+	noAvailableIPAddrs.Inc()
 	return "", -1, errors.New("assignPodIPv6AddressUnsafe: no available IP addresses")
 }
 
@@ -776,11 +798,14 @@ func (ds *DataStore) AssignPodIPv4Address(ipamKey IPAMKey, ipamMetadata IPAMMeta
 				ipsPerCidr.With(prometheus.Labels{"cidr": availableCidr.Cidr.String()}).Dec()
 				return "", -1, err
 			}
+			// Increment ENI IP usage on pod IPv4 allocation
+			eniIPsInUse.WithLabelValues(eni.ID).Inc()
 			return addr.Address, eni.DeviceNumber, nil
 		}
 		ds.log.Debugf("AssignPodIPv4Address: ENI %s does not have available addresses", eni.ID)
 	}
 
+	noAvailableIPAddrs.Inc()
 	ds.log.Errorf("DataStore has no available IP/Prefix addresses")
 	return "", -1, errors.New("assignPodIPv4AddressUnsafe: no available IP/Prefix addresses")
 }
@@ -1072,6 +1097,8 @@ func (ds *DataStore) RemoveUnusedENIFromStore(warmIPTarget, minimumIPTarget, war
 
 	// Prometheus update
 	enis.Set(float64(len(ds.eniPool)))
+	// Delete ENI IPs In Use when ENI is removed
+	eniIPsInUse.DeleteLabelValues(removableENI)
 	totalIPs.Set(float64(ds.total))
 	return removableENI
 }
@@ -1126,6 +1153,8 @@ func (ds *DataStore) RemoveENIFromDataStore(eniID string, force bool) error {
 
 	// Prometheus gauge
 	enis.Set(float64(len(ds.eniPool)))
+	// Delete ENI IPs In Use when ENI is removed
+	eniIPsInUse.DeleteLabelValues(eniID)
 	return nil
 }
 
@@ -1166,6 +1195,8 @@ func (ds *DataStore) UnassignPodIPAddress(ipamKey IPAMKey) (e *ENI, ip string, d
 	ipsPerCidr.With(prometheus.Labels{"cidr": availableCidr.Cidr.String()}).Dec()
 	ds.log.Infof("UnassignPodIPAddress: sandbox %s's ipAddr %s, DeviceNumber %d",
 		ipamKey, addr.Address, eni.DeviceNumber)
+	// Decrement ENI IP usage when a pod is deallocated
+	eniIPsInUse.WithLabelValues(eni.ID).Dec()
 	return eni, addr.Address, eni.DeviceNumber, nil
 }