Handle eventually-consistent PrivateDnsName on 1.26+

files/bin/private-dns-name

@@ -0,0 +1,44 @@
+#!/usr/bin/env bash
+
+set -o errexit
+set -o nounset
+set -o xtrace
+
+# Retrieves the PrivateDnsName from EC2 for this instance, waiting until
+# it is available if necessary (due to eventual consistency).
+
+function log {
+  echo >&2 "$(date '+%Y-%m-%dT%H:%M:%S%z')" "[private-dns-name]" "$@"
+}
+
+INSTANCE_ID=$(imds /latest/meta-data/instance-id)
+
+# the AWS CLI currently constructs the wrong endpoint URL on localzones (the availability zone group will be used instead of the parent region)
+# more info: https://github.com/aws/aws-cli/issues/7043
+REGION=$(imds /latest/meta-data/placement/region)
+
+# by default, wait for 120 seconds
+PRIVATE_DNS_NAME_MAX_ATTEMPTS=${PRIVATE_DNS_NAME_MAX_ATTEMPTS:-20}
+PRIVATE_DNS_NAME_ATTEMPT_INTERVAL=${PRIVATE_DNS_NAME_ATTEMPT_INTERVAL:-6}
+
+log "will make up to ${PRIVATE_DNS_NAME_MAX_ATTEMPTS} attempt(s) every ${PRIVATE_DNS_NAME_ATTEMPT_INTERVAL} second(s)"
+
+ATTEMPT=0
+while true; do
+  PRIVATE_DNS_NAME=$(aws ec2 describe-instances --region $REGION --instance-ids $INSTANCE_ID | jq -r '.Reservations[].Instances[].PrivateDnsName')
+  if [ ! "${PRIVATE_DNS_NAME}" = "" ] || [ ${ATTEMPT} -ge ${PRIVATE_DNS_NAME_MAX_ATTEMPTS} ]; then
+    break
+  fi
+  ATTEMPT=$((ATTEMPT + 1))
+  log "WARN: PrivateDnsName is not available, waiting for ${PRIVATE_DNS_NAME_ATTEMPT_INTERVAL} seconds..."
+  sleep ${PRIVATE_DNS_NAME_ATTEMPT_INTERVAL}
+done
+
+if [ "${PRIVATE_DNS_NAME}" = "" ]; then
+  log "ERROR: failed to retrieve PrivateDnsName after ${ATTEMPT} attempts!"
+  exit 1
+else
+  log "INFO: retrieved PrivateDnsName: ${PRIVATE_DNS_NAME}"
+  echo "${PRIVATE_DNS_NAME}"
+  exit 0
+fi

files/bootstrap.sh

@@ -531,12 +531,7 @@ else
   # If the VPC has a custom `domain-name` in its DHCP options set, and the VPC has `enableDnsHostnames` set to `true`,
   # then /etc/hostname is not the same as EC2's PrivateDnsName.
   # The name of the Node object must be equal to EC2's PrivateDnsName for the aws-iam-authenticator to allow this kubelet to manage it.
-  INSTANCE_ID=$(imds /latest/meta-data/instance-id)
-  # the AWS CLI currently constructs the wrong endpoint URL on localzones (the availability zone group will be used instead of the parent region)
-  # more info: https://github.com/aws/aws-cli/issues/7043
-  REGION=$(imds /latest/meta-data/placement/region)
-  PRIVATE_DNS_NAME=$(AWS_RETRY_MODE=standard AWS_MAX_ATTEMPTS=10 aws ec2 describe-instances --region $REGION --instance-ids $INSTANCE_ID --query 'Reservations[].Instances[].PrivateDnsName' --output text)
-  KUBELET_ARGS="$KUBELET_ARGS --hostname-override=$PRIVATE_DNS_NAME"
+  KUBELET_ARGS="$KUBELET_ARGS --hostname-override=$(private-dns-name)"
 fi
 
 KUBELET_ARGS="$KUBELET_ARGS --cloud-provider=$KUBELET_CLOUD_PROVIDER"

test/cases/private-dns-name.sh

@@ -0,0 +1,31 @@
+#!/usr/bin/env bash
+
+set -o nounset
+set -o errexit
+set -o pipefail
+
+echo "--> Should fetch PrivateDnsName correctly"
+EXPECTED_PRIVATE_DNS_NAME="ip-10-0-0-157.us-east-2.compute.internal"
+PRIVATE_DNS_NAME=$(private-dns-name)
+if [ ! "$PRIVATE_DNS_NAME" = "$EXPECTED_PRIVATE_DNS_NAME" ]; then
+  echo "❌ Test Failed: expected private-dns-name=$EXPECTED_PRIVATE_DNS_NAME but got '${PRIVATE_DNS_NAME}'"
+  exit 1
+fi
+
+echo "--> Should try to fetch PrivateDnsName until timeout is reached"
+export PRIVATE_DNS_NAME_ATTEMPT_INTERVAL=3
+export PRIVATE_DNS_NAME_MAX_ATTEMPTS=2
+export AWS_MOCK_FAIL=true
+START_TIME=$(date '+%s')
+EXIT_CODE=0
+private-dns-name || EXIT_CODE=$?
+STOP_TIME=$(date '+%s')
+if [[ ${EXIT_CODE} -eq 0 ]]; then
+  echo "❌ Test Failed: expected a non-zero exit code"
+  exit 1
+fi
+ELAPSED_TIME=$((STOP_TIME - START_TIME))
+if [[ "$ELAPSED_TIME" -lt 6 ]]; then
+  echo "❌ Test Failed: expected 6 seconds to elapse, but got: $ELAPSED_TIME"
+  exit 1
+fi

test/mocks/aws

@@ -7,15 +7,30 @@ SCRIPTPATH="$(
 
 echo >&2 "mocking 'aws $@'"
 
-if [[ $1 == "ec2" ]]; then
+AWS_MOCK_FAIL=${AWS_MOCK_FAIL:-false}
+if [ "$AWS_MOCK_FAIL" = "true" ]; then
+  echo >&2 "failing mocked 'aws $@'"
+  exit 1
+fi
 
+if [[ $1 == "ec2" ]]; then
   if [[ $2 == "describe-instance-types" ]]; then
     instance_type=$(echo "${@}" | grep -o '[a-z]\+[0-9]\+[a-z]*\.[0-9a-z]\+' | tr '.' '-')
     if [[ -f "${SCRIPTPATH}/describe-instance-types/${instance_type}.json" ]]; then
       cat "${SCRIPTPATH}/describe-instance-types/${instance_type}.json"
       exit 0
     fi
-    echo "instance type not found"
+    echo >&2 "instance type not found"
+    exit 1
+  fi
+  if [[ $2 == "describe-instances" ]]; then
+    instance_id=$(echo "${@}" | grep -o 'i\-[a-z0-9]\+')
+    echo >&2 "instance-id: $instance_id"
+    if [[ -f "${SCRIPTPATH}/describe-instances/${instance_id}.json" ]]; then
+      cat "${SCRIPTPATH}/describe-instances/${instance_id}.json"
+      exit 0
+    fi
+    echo >&2 "instance not found"
     exit 1
   fi
 fi

test/mocks/describe-instances/i-1234567890abcdef0.json

@@ -0,0 +1,154 @@
+{
+    "Reservations": [
+        {
+            "Groups": [],
+            "Instances": [
+                {
+                    "AmiLaunchIndex": 0,
+                    "ImageId": "ami-0abcdef1234567890",
+                    "InstanceId": "i-1234567890abcdef0",
+                    "InstanceType": "t3.nano",
+                    "KeyName": "my-key-pair",
+                    "LaunchTime": "2022-11-15T10:48:59+00:00",
+                    "Monitoring": {
+                        "State": "disabled"
+                    },
+                    "Placement": {
+                        "AvailabilityZone": "us-east-2a",
+                        "GroupName": "",
+                        "Tenancy": "default"
+                    },
+                    "PrivateDnsName": "ip-10-0-0-157.us-east-2.compute.internal",
+                    "PrivateIpAddress": "10-0-0-157",
+                    "ProductCodes": [],
+                    "PublicDnsName": "ec2-34-253-223-13.us-east-2.compute.amazonaws.com",
+                    "PublicIpAddress": "34.253.223.13",
+                    "State": {
+                        "Code": 16,
+                        "Name": "running"
+                    },
+                    "StateTransitionReason": "",
+                    "SubnetId": "subnet-04a636d18e83cfacb",
+                    "VpcId": "vpc-1234567890abcdef0",
+                    "Architecture": "x86_64",
+                    "BlockDeviceMappings": [
+                        {
+                            "DeviceName": "/dev/xvda",
+                            "Ebs": {
+                                "AttachTime": "2022-11-15T10:49:00+00:00",
+                                "DeleteOnTermination": true,
+                                "Status": "attached",
+                                "VolumeId": "vol-02e6ccdca7de29cf2"
+                            }
+                        }
+                    ],
+                    "ClientToken": "1234abcd-1234-abcd-1234-d46a8903e9bc",
+                    "EbsOptimized": true,
+                    "EnaSupport": true,
+                    "Hypervisor": "xen",
+                    "IamInstanceProfile": {
+                        "Arn": "arn:aws:iam::111111111111:instance-profile/AmazonSSMRoleForInstancesQuickSetup",
+                        "Id": "111111111111111111111"
+                    },
+                    "NetworkInterfaces": [
+                        {
+                            "Association": {
+                                "IpOwnerId": "amazon",
+                                "PublicDnsName": "ec2-34-253-223-13.us-east-2.compute.amazonaws.com",
+                                "PublicIp": "34.253.223.13"
+                            },
+                            "Attachment": {
+                                "AttachTime": "2022-11-15T10:48:59+00:00",
+                                "AttachmentId": "eni-attach-1234567890abcdefg",
+                                "DeleteOnTermination": true,
+                                "DeviceIndex": 0,
+                                "Status": "attached",
+                                "NetworkCardIndex": 0
+                            },
+                            "Description": "",
+                            "Groups": [
+                                {
+                                    "GroupName": "launch-wizard-146",
+                                    "GroupId": "sg-1234567890abcdefg"
+                                }
+                            ],
+                            "Ipv6Addresses": [],
+                            "MacAddress": "00:11:22:33:44:55",
+                            "NetworkInterfaceId": "eni-1234567890abcdefg",
+                            "OwnerId": "104024344472",
+                            "PrivateDnsName": "ip-10-0-0-157.us-east-2.compute.internal",
+                            "PrivateIpAddress": "10-0-0-157",
+                            "PrivateIpAddresses": [
+                                {
+                                    "Association": {
+                                        "IpOwnerId": "amazon",
+                                        "PublicDnsName": "ec2-34-253-223-13.us-east-2.compute.amazonaws.com",
+                                        "PublicIp": "34.253.223.13"
+                                    },
+                                    "Primary": true,
+                                    "PrivateDnsName": "ip-10-0-0-157.us-east-2.compute.internal",
+                                    "PrivateIpAddress": "10-0-0-157"
+                                }
+                            ],
+                            "SourceDestCheck": true,
+                            "Status": "in-use",
+                            "SubnetId": "subnet-1234567890abcdefg",
+                            "VpcId": "vpc-1234567890abcdefg",
+                            "InterfaceType": "interface"
+                        }
+                    ],
+                    "RootDeviceName": "/dev/xvda",
+                    "RootDeviceType": "ebs",
+                    "SecurityGroups": [
+                        {
+                            "GroupName": "launch-wizard-146",
+                            "GroupId": "sg-1234567890abcdefg"
+                        }
+                    ],
+                    "SourceDestCheck": true,
+                    "Tags": [
+                        {
+                            "Key": "Name",
+                            "Value": "my-instance"
+                        }
+                    ],
+                    "VirtualizationType": "hvm",
+                    "CpuOptions": {
+                        "CoreCount": 1,
+                        "ThreadsPerCore": 2
+                    },
+                    "CapacityReservationSpecification": {
+                        "CapacityReservationPreference": "open"
+                    },
+                    "HibernationOptions": {
+                        "Configured": false
+                    },
+                    "MetadataOptions": {
+                        "State": "applied",
+                        "HttpTokens": "optional",
+                        "HttpPutResponseHopLimit": 1,
+                        "HttpEndpoint": "enabled",
+                        "HttpProtocolIpv6": "disabled",
+                        "InstanceMetadataTags": "enabled"
+                    },
+                    "EnclaveOptions": {
+                        "Enabled": false
+                    },
+                    "PlatformDetails": "Linux/UNIX",
+                    "UsageOperation": "RunInstances",
+                    "UsageOperationUpdateTime": "2022-11-15T10:48:59+00:00",
+                    "PrivateDnsNameOptions": {
+                        "HostnameType": "ip-name",
+                        "EnableResourceNameDnsARecord": true,
+                        "EnableResourceNameDnsAAAARecord": false
+                    },
+                    "MaintenanceOptions": {
+                        "AutoRecovery": "default"
+                    }
+                }
+            ],
+            "OwnerId": "111111111111",
+            "ReservationId": "r-1234567890abcdefg"
+        }
+    ]
+}