Abhishek Chennaka created KUDU-3529:
---------------------------------------
Summary: Multiple HTTP 200 response when determining cloud
instance types could crash Kudu
Key: KUDU-3529
URL: https://issues.apache.org/jira/browse/KUDU-3529
Project: Kudu
Issue Type: Improvement
Reporter: Abhishek Chennaka
During server startup to determine the cloud instance type, we query the
link-local address 169.254.169.254. By default the flag values are as below:
cloud_aws_instance_id_url http://169.254.169.254/latest/meta-data/instance-id
cloud_azure_instance_id_url
http://169.254.169.254/metadata/instance/compute/vmId?api-version=2018-10-01&format=text
cloud_gce_instance_id_url -vs
http://metadata.google.internal/computeMetadata/v1/instance/id
cloud_openstack_instance_id_url
http://169.254.169.254/openstack/latest/meta_data.json
We rely on a HTTP 200 OK response from these URL to determine the instance
type. I came across a situation where the call to this URL (except
metadata.google.internal) was intercepted by a Firewall and resulted in a 200
response for three of the URLs causing the servers to fail to start with the
below response:
{code:java}
F20231121 11:48:39.135685 2094206 instance_detector.cc:109] Check failed:
kNoIdx == result_detector_idx_ (18446744073709551615 vs. 1) conflicting cloud
instance types
{code}
In order to avoid running into this issue we could parse the URL for the
contents instead of relying on just the 200 response.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
