Resolve AEM 6.5 Author slowness and failed publish operations caused by stuck Test Connection threads

If your AEM 6.5 Author instance is slow or publish operations fail, stuck replication Test Connection threads can be the cause. Perform a graceful restart of the Author instance, avoid repeated clicks on Test Connection, and apply AEM 6.5 Service Pack 2 (SP2) or later.

Description description

Environment

Adobe Experience Manager (AEM) 6.5 Author instance

Symptoms

  • High system load and heavy I/O wait on the Author instance
  • Author UI and page requests are slow or unresponsive
  • Publishing operations from Author to Publisher fail or are delayed
  • Thread dumps show multiple stuck threads related to replication Test Connection actions, each consuming significant CPU and holding open repository sessions
  • Request logs indicate that Test Connection requests never complete, and repository session statistics show accumulating unreleased sessions

Cause

A platform-level defect in AEM 6.5 (Oak 1.78.1) and replication modules can cause Test Connection actions on replication agents to enter a non-terminating loop. Each stuck request consumes CPU and holds open repository sessions, leading to resource exhaustion and degraded performance. This issue is resolved in AEM 6.5 Service Pack 2 (SP2) and later.

Resolution resolution

Follow these steps to resolve the issue.

  1. Perform a graceful restart of the affected AEM Author instance to clear all stuck Test Connection threads and release associated repository sessions.

  2. After restart, verify that system load and I/O wait have returned to normal levels and that the Author UI is responsive.

  3. Prevent recurrence by avoiding repeated clicks on the Test Connection button for replication agents, as each click can create a thread that may become stuck and consume resources.

  4. To check publisher health, use a direct curl command to the /bin/receive endpoint on the publisher:

    1. Run: curl -I https://<publisher-host>/bin/receive
    2. A healthy endpoint returns HTTP/1.1 401 Unauthorized, indicating it is reachable and functioning. (A 200 OK is returned only when valid credentials are provided.)
  5. Ensure the Author host has adequate memory and is not using swap, as low memory can worsen performance issues caused by stuck threads.

  6. Apply AEM 6.5 Service Pack 2 (SP2) or later, which includes fixes for the underlying repository and replication components. After updating, confirm that Test Connection completes normally and no stuck threads or unreleased sessions are observed.

Replication queue issues

recommendation-more-help
experience-cloud-kcs-help-kbarticles