Fix stale logic in mgr status command

src/ai/backend/manager/cli/__main__.py

@@ -343,12 +343,19 @@ async def inspect_node_status(cli_ctx: CLIContext) -> None:
     headers = ["ENDPOINT", "NODE ID", "IS LEADER", "RAFT TERM", "RAFT APPLIED INDEX"]
 
     if raft_configs is not None:
+        raft_cluster_configs = cli_ctx.raft_cluster_config
+        assert raft_cluster_configs is not None
+
+        other_peers = [{**peer, "myself": False} for peer in raft_cluster_configs["peers"]["other"]]
+        my_peers = [{**peer, "myself": True} for peer in raft_cluster_configs["peers"]["myself"]]
+        all_peers = sorted([*other_peers, *my_peers], key=lambda x: x["node-id"])
+
         initial_peers = Peers({
             int(peer_config["node-id"]): Peer(
                 addr=f"{peer_config['host']}:{peer_config['port']}",
                 role=InitialRole.from_str(peer_config["role"]),
             )
-            for peer_config in raft_configs["peers"]
+            for peer_config in all_peers
         })
 
         peers: dict[str, Any] | None = None

src/ai/backend/manager/cli/context.py

@@ -22,18 +22,20 @@
 from ai.backend.common.logging import AbstractLogger, LocalLogger
 from ai.backend.common.types import LogSeverity, RedisConnectionInfo
 
-from ..config import LocalConfig, SharedConfig
+from ..config import LocalConfig, SharedConfig, load_raft_cluster_config
 from ..config import load as load_config
 
 
 class CLIContext:
     _local_config: LocalConfig | None
+    _raft_cluster_config: LocalConfig | None
     _logger: AbstractLogger
 
     def __init__(self, config_path: Path, log_level: LogSeverity) -> None:
         self.config_path = config_path
         self.log_level = log_level
         self._local_config = None
+        self._raft_cluster_config = None
 
     @property
     def local_config(self) -> LocalConfig:
@@ -50,6 +52,23 @@ def local_config(self) -> LocalConfig:
             raise click.Abort()
         return self._local_config
 
+    @property
+    def raft_cluster_config(self) -> LocalConfig | None:
+        # Lazy-load the configuration only when requested.
+        try:
+            if self._raft_cluster_config is None:
+                self._raft_cluster_config = load_raft_cluster_config(
+                    self.config_path, self.log_level
+                )
+        except ConfigurationError as e:
+            print(
+                "ConfigurationError: Could not read or validate the manager raft cluster config:",
+                file=sys.stderr,
+            )
+            print(pformat(e.invalid_data), file=sys.stderr)
+            raise click.Abort()
+        return self._raft_cluster_config
+
     def __enter__(self) -> Self:
         # The "start-server" command is injected by ai.backend.cli from the entrypoint
         # and it has its own multi-process-aware logging initialization.

src/ai/backend/manager/config.py

@@ -615,8 +615,8 @@ def load(
 
 
 def load_raft_cluster_config(
-    debug_enabled: bool = False,
     raft_cluster_config_path: Optional[Path] = None,
+    log_level: LogSeverity = LogSeverity.INFO,
 ) -> Optional[LocalConfig]:
     try:
         raw_cfg, _ = config.read_from_file(raft_cluster_config_path, "raft-cluster-config")
@@ -625,7 +625,7 @@ def load_raft_cluster_config(
 
     try:
         cfg = config.check(raw_cfg, manager_raft_cluster_config_iv)
-        if debug_enabled:
+        if log_level == LogSeverity.DEBUG:
             print("== Raft cluster configuration ==", file=sys.stderr)
             print(pformat(cfg), file=sys.stderr)
     except config.ConfigurationError as e:

src/ai/backend/manager/server.py

@@ -1086,7 +1086,7 @@ def main(
     Start the manager service as a foreground process.
     """
     cfg = load_config(config_path, LogSeverity.DEBUG if debug else log_level)
-    raft_cluster_cfg = load_raft_cluster_config(debug, raft_cluster_config_path)
+    raft_cluster_cfg = load_raft_cluster_config(raft_cluster_config_path, log_level)
 
     if ctx.invoked_subcommand is None:
         cfg["manager"]["pid-file"].write_text(str(os.getpid()))