Skip to content

Cluster test: a replica keeps a row written after an expiring row in the same transaction #1399

Closed

Description

@DavidCockerill

Summary

HarperFast/harper#4390: when one origin transaction writes a row to a table with an expiration and then a row to a table without one, the replica stores the second row with the first row's expiry. After that time the replica hides the row and can evict it, while the origin keeps it. The core fix is HarperFast/harper#4391.

Core's regression test drives the shared apply loop through a subscribing source. It cannot reach the replication path: the origin's audit entry, the receiver copying the entry's expiry onto the event, and the receiver batching a message's writes into one transaction. Only a two-node test covers that.

Proposal

Add a cluster integration test under integrationTests/cluster/ with a small fixture component deployed on both nodes:

  • Exp (@table(expiration: <short>)) and Plain (no expiration).
  • One request writes Exp, then Plain, in one transaction; another writes them in the opposite order.
  • B joins A with add_node. After the Exp expiration passes, both Plain rows are readable on B and Exp is gone on both nodes.

The test fails with core before harper#4391 and passes with it. It lands with a core bump to that fix.

🤖 Generated by Claude Code (Anthropic Claude Opus); posted via @DavidCockerill.

Metadata

Metadata

Labels

No labels
No labels

Type

Fields

Priority

P2

Projects

No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions