devabhishekpal opened a new pull request, #1370: URL: https://github.com/apache/ratis/pull/1370
## What changes were proposed in this pull request? ### Problem In the current implementation, the leader provides the snapshot. However this causes tasks to get paused until the snapshot installation is completed and also puts unnecessary pressure on the leader. ### Goal Allow a lagging follower to install a snapshot from another follower without making that source follower act as the leader. This will let the lagging follower stay either in sync or "catch-up" to the point where it can append new entries without a complete snapshot. ### High Level Flow Diagram <img width="369" height="438" alt="image" src="https://github.com/user-attachments/assets/59583533-edb1-4ba9-bbc2-6b9f8bae277c"; /> ### How do we select the follower source? For selecting the follower source we consider the following metrics / conditions. #### Inputs - Target follower (`T`) which needs snapshot - Leader log state - specifically `lastEntry`, `logStartIndex` and `firstAvailableTermIndex` - Current follower progresses - specifically `matchIndex`, `commitIndex`, `snapshotIndex`, `lastRespondedAppendEntriesSendTime` and `lastRpcResponseTime`. #### Eligibility A follower (`F`) is considered a source only if: - `F` is recently responsive on the append path, for this we can use `lastRespondedAppendEntriesSendTime` as a check and fallback to `lastRpcResponseTime`. - `F.matchIndex >= requiredSnapshotIndex` where `requiredSnapshotIndex = firstAvailableTermIndex.index - 1` This is because: - `requiredSnapshotIndex` is the minimum snapshot index that still lets the target resume normal AppendEntries from the leader after install. - If `F.matchIndex < requiredSnapshotIndex`, that follower is too far behind to bridge the leader's log gap for this target, so the leader should not choose it. #### Ranking Rank eligible followers by this lexicographic order: - Exact sync with leader i.e. it is fully caught up to the leader - Highest `matchIndex` - Highest `commitIndex` in case match index is tied - Freshest `lastRespondedAppendEntriesSendTime` ##### IMPORTANT: If no follower satisfies `matchIndex >= requiredSnapshotIndex`, do not attempt follower-sourced install. Fall back immediately to the existing leader path because otherwise the target follower will need to perform another snapshot install to catchup anyway. ## What is the link to the Apache JIRA https://issues.apache.org/jira/browse/RATIS-2428 ## How was this patch tested? Patch was tested using the unit tests. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
