kubernetes/apps/otbr/deployment.yaml

@@ -115,12 +115,53 @@ spec:
               # dataset is persisted yet (clean install), the seed Job
               # (otbr-dataset-bootstrap) will populate it and activate
               # Thread itself.
-              if ot-ctl dataset active 2>&1 | grep -qE "Active Timestamp|Network Name"; then
-                echo "otbr: persisted dataset found; activating Thread"
+              #
+              # The dataset check MUST be polled, not sampled once.
+              # otbr-agent answers `ot-ctl state` as soon as its control
+              # socket exists, which is *before* it has restored the
+              # persisted dataset from /var/lib/thread. A single check
+              # right after the socket appears therefore races the restore
+              # and loses: on 2026-06-27 this pod booted, saw no dataset,
+              # took the "awaiting bootstrap Job" branch, and sat with
+              # Thread `disabled` for twelve days while `ot-ctl dataset
+              # active` returned NEST-PAN-339C the whole time. The seed Job
+              # had completed 61 days earlier, so nothing ever activated it.
+              # Reproduced on a 2026-07-09 restart before this fix landed.
+              DATASET_FOUND=""
+              for i in $(seq 1 30); do
+                if ot-ctl dataset active 2>&1 | grep -qE "Active Timestamp|Network Name"; then
+                  DATASET_FOUND=1
+                  break
+                fi
+                sleep 2
+              done
+              if [ -n "$DATASET_FOUND" ]; then
+                echo "otbr: persisted dataset found after ${i} attempt(s); activating Thread"
                 ot-ctl ifconfig up || true
                 ot-ctl thread start || true
+                # Surface the outcome. `thread start` returns before MLE has
+                # attached, so poll rather than trusting the exit code — a
+                # silent `disabled` here is exactly the failure this block
+                # exists to prevent. Note the trailing `|| true`: this script
+                # runs under `set -e`, so a bare failing test would kill the
+                # container rather than log a warning.
+                STATE=""
+                for i in $(seq 1 30); do
+                  STATE=$(ot-ctl state 2>/dev/null | head -1 || true)
+                  case "$STATE" in
+                    leader|router|child)
+                      echo "otbr: Thread attached as ${STATE}"
+                      break
+                      ;;
+                  esac
+                  sleep 2
+                done
+                case "$STATE" in
+                  leader|router|child) ;;
+                  *) echo "otbr: WARNING Thread not attached after 60s (state=${STATE:-unknown})" ;;
+                esac
               else
-                echo "otbr: no persisted dataset; awaiting bootstrap Job"
+                echo "otbr: no persisted dataset after 60s; awaiting bootstrap Job"
               fi
               wait "$ENTRYPOINT_PID"
           env: