List Connections API endpoint returns duplicate records when using pagination

[GideonStowell]

My team and I are troubleshooting an issue with our self hosted airbyte instances (v1.7) related to the airbyte api.

We have external systems that use the airbyte api to create new connections (and sources/destinations if needed). The process uses the GET v1/connections endpoint to first fetch all connections and check if the target connection already exists. It's important to note that the api calls do use the pagination to fetch all records using the limit and offset query parameters. If the target connection exists then it uses the PUT v1/connections/{connectionId} endpoint to assert the connection matches our stored config. If the connection doesn't exist it creates the connection using the POST v1/connections endpoint.

The behavior we have noticed is that the GET v1/connections endpoint returns inconsistent results. We have run a few tests and routinely the airbyte API will return the same connection twice in the result set.

For example, we know through the UI there are 200 connections, we can make some api calls like this:

curl --request GET --url 'https://airbyte.domain.com/v1/connections?limit=100&offset=0' --header 'accept: application/json'
curl --request GET --url 'https://airbyte.domain.com/v1/connections?limit=100&offset=100' --header 'accept: application/json'

This would return the correct number of connections (200) but when I search for my target connection (that I know exists) it is not found. And upon further inspection of the returned results, I see that several connections are duplicates of each other.

We did attempt an upgrade to version 1.8 on our development cluster but this behavior has persisted.

Has anyone run into a similar thing before?

It feels like this is a bug in the airbyte api but perhaps there is a gap in our troubleshooting process.

Environment details:
OSS Self hosted on Kubernetes EKS
v1.7 (also tested on v1.8)