17 lines
4.0 KiB
Markdown
17 lines
4.0 KiB
Markdown
# OT11 Fleet and Hosts refresh checkpoint
|
|
|
|
Fleet now catches read failures without sending the page to its boundary or replacing the last successful reading. A first failure says unavailable rather than loading forever or reporting an empty fleet. It displays the acquisition date/time in UTC and a separate first failure time; retry restores the existing view. Per-node applied/degraded/failed/waiting meanings, query and cluster union are unchanged.
|
|
|
|
Hosts keeps membership and configuration acquisition times separate. A failed configuration read preserves both prior node and instance enrichment lists. Both are installed only after the existing two queries and deployed-set lookup return; an unsuccessful query cannot trigger snapshot eviction. The deployed/authored retention rule, health snapshot source, dirty coalescing and existing status mappings are retained. Initial loading now shares the timer/manual overlap guard, and retirement prevents late subscriptions/timer creation/results. The Controls modal remains outside this checkpoint, owned by the later modal migration.
|
|
|
|
Both views preserve the Refresh button's focus with aria-disabled plus the existing server overlap guard, expose aria-busy, and show manual-only polite pending/success feedback. Error first-occurrence timestamps remain stable across repeated failures so automatic polling does not continually rewrite live error announcements. Successful acquisition clocks are outside live regions. Internal exception details are logged, not rendered.
|
|
|
|
## Verification
|
|
|
|
- Six actual bUnit page cases: Fleet unavailable/retry, last-good rows/time, overlap/disposal cancellation; Hosts retained rendered human name/fault and cached config, initial unavailable/recovery, and initial/manual overlap/disposal cancellation. Initial four cases retained three failures/one overlap positive before the fix. Final cases use bUnit2.7.2 and existing authorization/test facilities.
|
|
- One private SQL provider regression directly invokes the actual Hosts configuration read: the real ClusterNode query completes, a command interceptor rejects the subsequent DriverInstance query, both distinct cached lists and their acquisition time remain intact, and deployed/eviction dependencies are not reached. It seeds no application rows and uses the existing uniquely provisioned test database lifecycle. Initial interceptor used plural table names, missed both singular tables and failed its observation assertion; that harness error is retained in aggregate-with-sql.log. The corrected test passes.
|
|
- Final combined40passed/0skips in acceptance.log: seven new cases plus33 existing Hosts projection/coalescing/deployed-set/panel-wiring contracts. All test discovery and execution uses the owned-loopback private SQL environment runner. State/password remain outside Git. Earlier missing assets/compiler failures are retained as harness setup, not behavioral failures.
|
|
- Twelve actual browser cases across Fleet/Hosts, initial failure versus later stale failure,1440/390 and native200% at720CSSpx/DPR2. Native Enter retries preserve focused Refresh, retained rows and acquisition time; correct unavailable/last-good feedback disappears on recovery. Two desktop stale cases wait through the actual polling interval and verify stable error content. No page errors or document overflow. Two representative captures were visually inspected. Browser evidence/script and owned RefreshPilot live under scadaproj/components/ui-theme/validation/ot-ui/ot11-root. First readiness click arrived before hydration; the harmless fixture readiness toggle now waits through hydration, with original failure retained.
|
|
|
|
The browser host uses memory-only records, a seedless loopback Akka membership shell and operation-refusing services. Its empty membership view is synthetic, not fleet certification. No actual deployment, external probe, cluster operation or production SQL write is performed. Full OT suite, other live-tail pages, remaining modal callers and human screen-reader listening are separate gates; this does not close OT11 or OT14.
|