fix: improve langbot rollout diagnostics

config/platform-infra/langbot.yaml

@@ -13,7 +13,7 @@ image:
   pullPolicy: IfNotPresent
 
 dependencyImages:
-  postgresClient: docker.m.daocloud.io/library/postgres:16-alpine
+  postgresClient: docker.io/library/postgres:16-alpine
 
 targets:
   - id: G14
@@ -32,7 +32,7 @@ targets:
         deploymentName: langbot-frpc
         secretName: langbot-frpc-secrets
         secretKey: frpc.toml
-        image: 127.0.0.1:5000/hwlab/frpc:v0.68.1
+        image: fatedier/frpc:v0.68.1
         serverAddr: 82.156.23.220
         serverPort: 22000
         proxyName: platform-infra-langbot-g14-web

scripts/src/platform-infra-langbot.ts

@@ -1273,6 +1273,7 @@ capture networkpolicy kubectl -n ${target.namespace} get networkpolicy allow-all
 capture ingress kubectl -n ${target.namespace} get ingress
 capture quota kubectl -n ${target.namespace} get resourcequota
 capture limitrange kubectl -n ${target.namespace} get limitrange
+capture events kubectl -n ${target.namespace} get events --sort-by=.lastTimestamp
 python3 - "$tmp" <<'PY'
 import json, os, sys
 tmp = sys.argv[1]
@@ -1290,6 +1291,7 @@ deployments = load("deployments") or {"items": []}
 pods = load("pods") or {"items": []}
 services = load("services") or {"items": []}
 pvc = load("pvc") or {"items": []}
+events = load("events") or {"items": []}
 def deployment_summary(name):
     for item in deployments.get("items", []):
         if item.get("metadata", {}).get("name") == name:
@@ -1314,6 +1316,66 @@ deployment_ready = (
     and plugin_deploy.get("readyReplicas", 0) >= 1
     and frpc_deploy.get("readyReplicas", 0) >= 1
 )
+watched_names = {"${serviceName}", "${pluginRuntimeServiceName}", "${target.publicExposure.frpc.deploymentName}"}
+watched_pod_names = {
+    item.get("metadata", {}).get("name")
+    for item in pods.get("items", [])
+    if item.get("metadata", {}).get("labels", {}).get("app.kubernetes.io/name") in watched_names
+}
+watched_pvc_names = {
+    item.get("metadata", {}).get("name")
+    for item in pvc.get("items", [])
+    if item.get("metadata", {}).get("labels", {}).get("app.kubernetes.io/part-of") == "platform-infra"
+       and str(item.get("metadata", {}).get("name", "")).startswith("langbot-")
+}
+def compact_message(value, limit=500):
+    if not value:
+        return None
+    value = str(value).replace("\\n", " ")
+    return value if len(value) <= limit else value[:limit] + "...(truncated)"
+def state_summary(status):
+    state = status.get("state") or {}
+    if not state:
+        return {"type": "unknown"}
+    state_type = next(iter(state.keys()))
+    detail = state.get(state_type) or {}
+    return {
+        "type": state_type,
+        "reason": detail.get("reason"),
+        "message": compact_message(detail.get("message")),
+    }
+def condition_summary(condition):
+    return {
+        "type": condition.get("type"),
+        "status": condition.get("status"),
+        "reason": condition.get("reason"),
+        "message": compact_message(condition.get("message")),
+    }
+def event_summary(event):
+    involved = event.get("involvedObject", {})
+    return {
+        "type": event.get("type"),
+        "reason": event.get("reason"),
+        "message": compact_message(event.get("message"), 700),
+        "count": event.get("count"),
+        "lastTimestamp": event.get("lastTimestamp") or event.get("eventTime"),
+        "involvedObject": {
+            "kind": involved.get("kind"),
+            "name": involved.get("name"),
+        },
+    }
+relevant_events = []
+for event in events.get("items", []):
+    involved = event.get("involvedObject", {})
+    kind = involved.get("kind")
+    name = involved.get("name")
+    if (
+        name in watched_pod_names
+        or name in watched_pvc_names
+        or name in watched_names
+        or (kind == "ReplicaSet" and any(str(name or "").startswith(prefix + "-") for prefix in watched_names))
+    ):
+        relevant_events.append(event_summary(event))
 payload = {
     "ok": rc("ns") == 0 and rc("deployments") == 0 and rc("pods") == 0 and deployment_ready,
     "namespace": "${target.namespace}",
@@ -1334,18 +1396,34 @@ payload = {
             "name": item.get("metadata", {}).get("name"),
             "phase": item.get("status", {}).get("phase"),
             "ready": sum(1 for condition in item.get("status", {}).get("conditions", []) if condition.get("type") == "Ready" and condition.get("status") == "True"),
+            "conditions": [
+                condition_summary(condition)
+                for condition in item.get("status", {}).get("conditions", [])
+                if condition.get("status") != "True" or condition.get("type") in ("Ready", "ContainersReady", "PodScheduled")
+            ],
+            "initContainers": [
+                {
+                    "name": c.get("name"),
+                    "ready": c.get("ready"),
+                    "restartCount": c.get("restartCount"),
+                    "state": state_summary(c),
+                }
+                for c in item.get("status", {}).get("initContainerStatuses", [])
+            ],
             "containers": [
                 {
                     "name": c.get("name"),
                     "ready": c.get("ready"),
                     "restartCount": c.get("restartCount"),
-                    "state": list((c.get("state") or {}).keys()),
+                    "state": state_summary(c),
                 }
                 for c in item.get("status", {}).get("containerStatuses", [])
             ],
         }
         for item in pods.get("items", [])
+        if item.get("metadata", {}).get("name") in watched_pod_names
     ],
+    "events": relevant_events[-20:],
     "services": [item.get("metadata", {}).get("name") for item in services.get("items", [])],
     "pvcs": [item.get("metadata", {}).get("name") for item in pvc.get("items", [])],
     "secrets": {