[CURATOR-545] PersistentTtlNode can leave dangling containers behind

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Won't Do
• Affects Version/s: 4.2.0
• Component/s: Recipes
• Assignee: Unassigned
• Reporter: Sylvain Wallez
• Created: 09/Oct/19 16:51
• Updated: 25/Nov/19 14:44
• Resolved: 25/Nov/19 14:44

Description

PersistentTtlNode creates a container as a PersistentNode and then adds a TTL node in it named touch that is refreshed periodically.

We encountered an issue were the application crashed (or was disconnected from ZooKeeper) between the creation of the container and the creation of the touch child.

ZooKeeper doesn't delete containers that never had any children. So this result in this container node to stay around forever and never expire.

A fix for that is to create both container and touch child as part of a single transaction.

Since PersistentTtlNode delegates to PersistentNode this may require quite some refactoring of the recipe. But I actually don't understand why the container has to be a PersistentNode and is not a regular container.

Issue Links

• relates to: ZOOKEEPER-3546 Containers that never have children stay forever