Fixes and improvements

agent/agent.sh

@@ -8,9 +8,6 @@ STATE_ROOT="/etc/nodeflare"
 STATE_DIR="$STATE_ROOT/agent"
 SERVICE_FILE="/etc/systemd/system/$SERVICE_NAME.service"
 OPENRC_FILE="/etc/init.d/$SERVICE_NAME"
-# Holds NODEFLARE_AGENT_TOKEN for systemd; a 0600 file inside the 0750 state dir
-# is readable only by root/uninstall, unlike Environment= in the unit.
-TOKEN_ENV_FILE="$STATE_DIR/agent.env"
 temporary=""
 backup_agent=""
 backup_service=""
@@ -350,10 +347,6 @@ install_agent() {
   mv "$temporary" "$AGENT_FILE"
   log "正在启动 $init_system 服务"
   if [ "$init_system" = "systemd" ]; then
-    # The token goes into a 0600 file rather than Environment=, which any local
-    # user can read back with `systemctl show`.
-    printf 'NODEFLARE_AGENT_TOKEN=%s\n' "$token" > "$TOKEN_ENV_FILE"
-    chmod 600 "$TOKEN_ENV_FILE"
     printf '%s\n' \
     '[Unit]' \
     'Description=NodeFlare Agent' \
@@ -366,7 +359,7 @@ install_agent() {
     'Restart=always' \
     'RestartSec=10' \
     'NoNewPrivileges=true' \
-    "EnvironmentFile=$TOKEN_ENV_FILE" \
+    "Environment=NODEFLARE_AGENT_TOKEN=$token" \
     "Environment=NODEFLARE_STATE_DIR=$STATE_DIR" \
     '' \
     '[Install]' \
@@ -411,15 +404,7 @@ load_installed_agent_config() {
       service_exec=$(sed -n 's/^ExecStart=//p' "$SERVICE_FILE" | head -n 1)
       endpoint=$(printf '%s\n' "$service_exec" | sed -n 's/.* -e \([^[:space:]]*\) -i \([0-9][0-9]*\)$/\1/p')
       interval=$(printf '%s\n' "$service_exec" | sed -n 's/.* -e \([^[:space:]]*\) -i \([0-9][0-9]*\)$/\2/p')
-      token=""
-      if [ -f "$TOKEN_ENV_FILE" ]; then
-        token=$(sed -n 's/^NODEFLARE_AGENT_TOKEN=//p' "$TOKEN_ENV_FILE" | head -n 1)
-      fi
-      # Fall back to the unit file so installs created before the token moved
-      # into an EnvironmentFile still update cleanly.
-      if [ -z "$token" ]; then
-        token=$(sed -n 's/^Environment=NODEFLARE_AGENT_TOKEN=//p' "$SERVICE_FILE" | head -n 1)
-      fi
+      token=$(sed -n 's/^Environment=NODEFLARE_AGENT_TOKEN=//p' "$SERVICE_FILE" | head -n 1)
       ;;
     openrc)
       [ -f "$OPENRC_FILE" ] || fail "未找到 Agent 服务配置，请先完成安装"

agent/src/live.rs

@@ -574,6 +574,7 @@ pub(crate) fn live_sender_loop(endpoint: &str, token: &str, worker: LiveSenderWo
                                 stats.live_persistence_failed();
                                 drop_socket = true;
                             } else {
+                                backoff.reset();
                                 prune_live_queue(&pending, ack.persisted_through_ts);
                                 accepted_through = accepted_through.max(ack.persisted_through_ts);
                                 next_probe_at = Instant::now() + ack_persist_interval(&ack);
@@ -644,7 +645,6 @@ pub(crate) fn live_sender_loop(endpoint: &str, token: &str, worker: LiveSenderWo
             continue;
         }
         stats.live_batch_sent(batch_len);
-        backoff.reset();
         if let Some(timestamp) = batch_last {
             accepted_through = accepted_through.max(timestamp);
         }
@@ -663,6 +663,7 @@ pub(crate) fn live_sender_loop(endpoint: &str, token: &str, worker: LiveSenderWo
                         stats.live_persistence_failed();
                         drop_socket = true;
                     } else {
+                        backoff.reset();
                         prune_live_queue(&pending, ack.persisted_through_ts);
                         accepted_through = accepted_through.max(ack.persisted_through_ts);
                         next_probe_at = Instant::now() + ack_persist_interval(&ack);

agent/src/main.rs

@@ -120,12 +120,8 @@ struct CliOptions {
     #[arg(short = 'e', value_name = "URL")]
     endpoint: String,
     /// Agent token
-    #[arg(short = 't', value_name = "TOKEN", conflicts_with = "token_file")]
+    #[arg(short = 't', value_name = "TOKEN")]
     token: Option<String>,
-    /// Read the Agent token from a file instead of -t, which keeps it out of
-    /// the process list
-    #[arg(long = "token-file", value_name = "PATH")]
-    token_file: Option<PathBuf>,
     /// Initial report interval in seconds (15-3600)
     #[arg(short = 'i', value_name = "SECONDS", default_value_t = 60)]
     interval: u64,
@@ -455,26 +451,16 @@ fn advance_deadline(mut deadline: Instant, interval: Duration, now: Instant) ->
     deadline
 }
 
-/// Reads an Agent token from a file, trimming surrounding whitespace (editors
-/// and shell redirection usually append a trailing newline).
-fn read_token_file(path: &Path) -> Result<String> {
-    let contents = fs::read_to_string(path)
-        .map_err(|error| format!("failed to read token file {}: {error}", path.display()))?;
-    Ok(contents.trim().to_string())
-}
-
 fn runtime_config(options: &CliOptions) -> Result<RuntimeConfig> {
     let interval = options.interval;
     if !(15..=3600).contains(&interval) {
         return Err("interval must be between 15 and 3600 seconds".into());
     }
-    let token = match &options.token {
-        Some(token) => token.clone(),
-        None => match options.token_file.as_deref() {
-            Some(path) => read_token_file(path)?,
-            None => env::var("NODEFLARE_AGENT_TOKEN").unwrap_or_default(),
-        },
-    };
+    let token = options
+        .token
+        .clone()
+        .or_else(|| env::var("NODEFLARE_AGENT_TOKEN").ok())
+        .unwrap_or_default();
     let endpoint = options.endpoint.clone();
     if token.is_empty() || token.len() > 512 || token.chars().any(char::is_whitespace) {
         return Err("token is invalid".into());

agent/src/metrics.rs

@@ -57,6 +57,8 @@ pub(crate) fn output_with_timeout(
     mut process: Command,
     timeout: Duration,
 ) -> Option<std::process::Output> {
+    #[cfg(unix)]
+    process.process_group(0);
     let mut child = process
         .stdin(Stdio::null())
         .stdout(Stdio::piped())
@@ -75,13 +77,13 @@ pub(crate) fn output_with_timeout(
         match child.try_wait() {
             Ok(Some(status)) => break Some(status),
             Ok(None) if started.elapsed() >= timeout => {
-                let _ = child.kill();
+                terminate_command_tree(&mut child);
                 let _ = child.wait();
                 break None;
             }
             Ok(None) => thread::sleep(Duration::from_millis(20)),
             Err(_) => {
-                let _ = child.kill();
+                terminate_command_tree(&mut child);
                 let _ = child.wait();
                 break None;
             }
@@ -97,6 +99,29 @@ pub(crate) fn output_with_timeout(
     })
 }
 
+fn terminate_command_tree(child: &mut Child) {
+    #[cfg(unix)]
+    {
+        let group = format!("-{}", child.id());
+        let _ = Command::new("kill")
+            .args(["-KILL", "--", group.as_str()])
+            .stdin(Stdio::null())
+            .stdout(Stdio::null())
+            .stderr(Stdio::null())
+            .status();
+    }
+    #[cfg(target_os = "windows")]
+    {
+        let _ = Command::new("taskkill")
+            .args(["/PID", &child.id().to_string(), "/T", "/F"])
+            .stdin(Stdio::null())
+            .stdout(Stdio::null())
+            .stderr(Stdio::null())
+            .status();
+    }
+    let _ = child.kill();
+}
+
 pub(crate) fn command(name: &str, args: &[&str]) -> String {
     let mut process = Command::new(name);
     process.args(args);

agent/src/tests.rs

@@ -430,6 +430,16 @@ fn remote_commands_are_terminated_after_the_deadline() {
     assert!(result.result.contains("已终止"));
 }
 
+#[cfg(unix)]
+#[test]
+fn command_timeout_terminates_descendants_without_waiting_for_pipes() {
+    let mut command = std::process::Command::new("sh");
+    command.args(["-c", "sleep 2 & wait"]);
+    let started = Instant::now();
+    assert!(super::output_with_timeout(command, Duration::from_millis(50)).is_none());
+    assert!(started.elapsed() < Duration::from_secs(1));
+}
+
 #[test]
 fn combined_remote_output_is_bounded_at_utf8_boundaries() {
     let result = remote_result_text(
@@ -628,18 +638,6 @@ fn parses_cli_options() {
     assert!(parsed.once);
     assert!(CliOptions::try_parse_from(["agent", "-t"]).is_err());
     assert!(CliOptions::try_parse_from(["agent", "-t", "first", "-t", "second"]).is_err());
-    assert!(
-        CliOptions::try_parse_from([
-            "agent",
-            "-e",
-            "https://monitor.example.com",
-            "-t",
-            "agent-token",
-            "--token-file",
-            "/run/nodeflare/token",
-        ])
-        .is_err()
-    );
     assert!(CliOptions::try_parse_from(["agent", "--once", "--collect"]).is_err());
 }