[Question][GitHub] PR records remain duplicated and stale after a repository is moved between GitHub connections

[sobanjom]

Question

We are investigating how Apache DevLake handles repositories that are moved from one GitHub connection to another.

Our DevLake instance contains the same GitHub repository and pull request under multiple connection-scoped identities. For example, GitHub PR #137 from an example repo

is present in the pull_requests table as follows:

github:GithubPullRequest:1:3375195798   OPEN
github:GithubPullRequest:35:3375195798  MERGED
github:GithubPullRequest:36:3375195798  MERGED
github:GithubPullRequest:37:3375195798  MERGED

GitHub confirms that PR #137 was merged:

created_at: 2026-03-09T20:00:14Z
updated_at: 2026-03-10T03:38:33Z
closed_at: 2026-03-10T03:38:31Z
merged_at: 2026-03-10T03:38:31Z

The most recent pipeline for the project collected the repository through connection 35:

{
  "plugin": "github_graphql",
  "options": {
    "connectionId": 35,
    "fullName": "",
    "githubId":
  }
}

That pipeline correctly created or updated the connection-35 record to MERGED. However, the older connection-1 record remains OPEN and is not updated by subsequent pipeline runs.

Our reporting queries join pull requests to repositories by repository name and therefore include both the active connection-35 record and the stale connection-1 record. This results in incorrect open-PR counts unless we explicitly filter by the currently active connection-specific repository ID.

Is this behavior expected by design, or should DevLake reconcile records across GitHub connections when the same GitHub repository and pull request are collected through a different connection?

In particular:

1. Is the connection ID intentionally part of the identity of a GitHub repository and pull request?
2. When a repository is removed from connection 1 and added to connection 35, should historical records under connection 1 remain permanently queryable?
3. Should a future full-sync or backfill pipeline update the old connection-1 record, even though connection 1 is no longer present in the project blueprint?
4. Is there a supported procedure for migrating repository data between connections without creating duplicate tool-layer and domain-layer entities?
5. Is there a recommended way to mark or remove obsolete connection-scoped records?
6. What is the recommended query or dashboard strategy for counting PRs and identifying stale open PRs when the same repository exists under multiple connection IDs?
7. Is there a supported cleanup or deduplication task that preserves related PR comments, reviews, commits, deployments, and DORA metrics?

Additional context

The project uses the GitHub GraphQL plugin and has skipOnFail enabled. The affected pipeline also experienced a separate network timeout while preparing a GitHub GraphQL task:

unable to get github API client instance
Failed to connect
dial tcp 20.26.156.210:443: i/o timeout

However, for PR #137, the active connection-35 record was successfully updated to MERGED. The issue appears specifically related to the older connection-1 record remaining stale and being included in cross-connection reporting.

The project’s sync policy is currently:

Data time range: 2026-05-07 01:00 to Now
Sync frequency: Custom
Skip failed tasks: Enabled

We would appreciate clarification on whether this is:

• expected connection-scoped identity behavior,
• a limitation of connection migration,
• a data cleanup/configuration issue,
• or a defect in the GitHub GraphQL collector, extractor, converter, or project mapping logic.

Expected outcome

We would like to understand the supported DevLake behavior and obtain guidance for:

• migrating repositories between GitHub connections,
• preventing duplicate records,
• excluding obsolete connection data from project metrics,
• and safely cleaning up stale records without corrupting related DevLake data.