[SLING-11450] Partially started instance suppression can lead to unwanted leader loss

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Discovery Base 2.0.12, Discovery Oak 1.2.36
• Fix Version/s: Discovery Base 2.0.14, Discovery Oak 1.2.38
• Component/s: Discovery
• Assignee: Stefan Egli
• Reporter: Stefan Egli
• Created: 13/Jul/22 14:37
• Updated: 03/Aug/22 09:49
• Resolved: 21/Jul/22 14:08

Description

SLING-10489 introduced "partial startup suppression" sometimes also referred to as "joinerdelay" (even though the latter is actually a subfeature of the former).

With this suppression enabled (it is disabled by default), upon a topology change the leader instance can loose its leader status even though it did not actually leave the topology or crash. This is against the discovery API contract, which says that the leader stays leader until it crashes.