Skip to main content

Version 1.18.1.7

Release Date: September 10, 2026 Release Type: Stable Previously published as: 1.18.12 (image tag staging-1.18.12)

Failed messages are recorded as ERROR instead of remaining in DELIVERING, credentials used by Azure and AWS Start nodes are protected from deletion, list search and sorting stay consistent, and the Health Check external API explains what causes an unhealthy status.

Backend Server​

Bug Fixes​

  • Failed messages are recorded as ERROR instead of staying in DELIVERING: In some cases a failed message was answered to the caller but its monitoring record was never switched to ERROR. The record stayed in DELIVERING until the cleanup of stuck messages closed it about fifteen minutes later with a generic 504 error in place of the real one. The error is now recorded when it happens; the fix covers the general flow error path, Kafka subscriptions and the OData sender connector.
  • The record of a failed run is consistent in both stores: The end date and status recorded for a failed run now match what actually happened to it. Node log entries that PostgreSQL rejected, or that replaced an earlier entry on Elasticsearch without any error, are now written correctly, so they can be found and updated afterwards.

Frontend Server​

Bug Fixes​

  • Credentials used by Azure Blob Storage and AWS Start nodes can no longer be deleted while in use: The check that protects a credential from deletion did not recognize flows whose Start node is an Azure Blob Storage or AWS connector, so a credential that a deployed flow depends on was reported as unused and could be deleted; the flow then failed the next time it tried to connect. Such credentials are now reported as in use and cannot be deleted.

Frontend Web​

New Features​

  • Redesigned Component Health table on System Health: The Component Health table listed every store instance as a flat, unpaged list of raw UP and DOWN texts. It is now a paged table that shows each instance with the health states the platform distinguishes.

Bug Fixes​

  • Search no longer discards the active sort and filters: On list pages with a search box, changing the sort order shortly after typing a search term could bring back rows in the default order while the column header still showed the selected sort; on several pages the active filters were dropped as well. The search now always uses the sort and filters currently on screen. The fix covers the Alerts list and twenty other list pages, including Message Monitoring and its detail page, System Logs, AI Monitoring and Performance Monitoring.

Health Check​

New Features​

  • External API states the reason for the overall status: The overall section of the external API now carries reason, one sentence that describes the worst current issue with its threshold and notes how many further issues exist, and issues[], a list of every component and resource graded WARNING, CRITICAL or DOWN, worst first. An external monitoring tool can therefore see directly that, for example, disk usage is the cause, without reading every section. The thresholds used in the wording come from the external-api.severity.* settings.
  • Stale backend pod names are removed automatically: The list of known backend pods only ever grew, so pods that no longer existed, for example after a container was recreated, kept being listed, and the external API reported them with no history. Health Check now removes a pod name once it has been observed for six hours without a single successful check. Nothing is removed during the first fifteen minutes after Health Check starts, and the pods are never all removed at once, so a backend outage does not empty the list. These are the default values; the behaviour is configured through the healthwatch.pod-reaper.* settings.

Bug Fixes​

  • A disabled Elasticsearch is no longer shown as a failed component: When Elasticsearch is switched off for an installation, the component list showed it as permanently DOWN, an unhealthy entry that no action could clear. It is no longer listed among the components. Resync suggestions for recorded store outages are also withheld while Elasticsearch is disabled, so the outage banner no longer asks for a repair that cannot run; the outage history itself remains visible.

1.18.1.7 is a stable release. Previous: 1.18.1.6