raft: storage append error leaves batches in log for replication

[dotnwat]

A bug related to snappy (https://github.com/vectorizedio/redpanda/issues/2390) is triggered on append when a large batch is decompressed in a compacted topic. In the storage appender a batch is first appended to the log and second the decompression occurs.

This decompression throws an exception, but the data would likely still be in the log.

The bug we appear to be observing is that despite this exception during append, a topic still replicated the batch to some replicas. At some point this issue results in a segfault but as nodes crash and recover the same batch is re-replicated during recovery.

In principle, in a acks=all situation, this should never happen because those failed append should not be considered to have succeeded and the logs shoudl be truncated at the point of the problematic batch.

JIRA Link: CORE-768

[github-actions[bot]]

This issue hasn't seen activity in 3 months. If you want to keep it open, post a comment or remove the stale label – otherwise this will be closed in two weeks.

[github-actions[bot]]

This issue was closed due to lack of activity. Feel free to reopen if it's still relevant.