Unique id for Repository on /repos/ API endpoint

[humitos]

Hi! We use the /repos/ (https://docs.github.com/en/rest/repos/repos) endpoint from the API to save a local cache to speed up some queries for our application without hitting GitHub API.

After some time we found some weird behavior where some repositories were duplicated in our local cache. In the last months, it started to happen more frequently and while debugging it I found that the id returned by this endpoint was not unique and could change over time --even if all the rest of its attribute are the same.

As an example, we detected that on August 16, 2022 the repository extratone/wtf changes its id from 422691395 to 515187740:

$ gh api -H "Accept: application/vnd.github+json" /repos/extratone/wtf | jq | head      
{
  "id": 515187740,             // this is the `id` we are using as unique
  "node_id": "R_kgDOHrUkHA",
  "name": "wtf",
  "full_name": "extratone/wtf",
  "private": false,
  "owner": {
    "login": "extratone",
    "id": 43663476,
    "node_id": "MDQ6VXNlcjQzNjYzNDc2",

What is the reason behind this id change? What is the field from the /repos/ endpoint we should consider unique?

Note that we are tracking it by id and not by full_name because the later could change if the repository is renamed or transferred to a different organization.

[adelinn]

I am also interested whether the repository id is unique and immutable or not.

[typed-sigterm]

So guys let's use node_id.

Bump ⬆️ also interested in this.

[marckhair]

Didn't test it for repos, but the node_id of branches changes on branch rename. Might be globally unique, but it does not seem immutable.

[Miggleness]

I believe it's because behind the scenes that deletes the existing branch and creates a new one.

[heronsanches]

We may want to use "full_name" as the unique id.

[typed-sigterm]

No full_name will change when the repo is renamed, transferred or the owner is renamed