Merged upstreamFeatureSelected contributionMerged Sep 15, 2026
Add checkpoint-aware append-only LocalFile tailing
Extended continuous LocalFile discovery to read complete appended text rows with committed offsets and checkpoint restoration.
apache/seatunnel · #11819
Problem
A user reported that log collection stopped after reading the initial file and ignored subsequent appends.
Approach
Emits delimiter-complete byte ranges, checkpoints committed offsets, restores in-flight ranges, supports earliest/latest starts, follows rename by filesystem identity, detects replacement/copy-truncate, and isolates discovery failures. Default bounded reads and binary updates remain unchanged.
Impact and scope
- Enables ongoing local log ingestion instead of treating the initial file contents as the end of the source.
- Retains committed progress on checkpoint recovery and handles supported rotation without rereading completed ranges.
- Limited to uncompressed UTF-8 text with stable file keys. Rotated files must remain under the configured path; copy-truncate restarts at the header boundary, and other in-place edits are not guaranteed.
Validation
- The PR reports 272 file-base tests passing on JDK 11, including partial rows, checkpoints, identity changes, failure isolation, retries, and serialization compatibility.
- GitHub Build passed and the linked issue is closed. The verified merge commit credits Goutam Adwant; this does not establish deployed log-ingestion outcomes.