Log --run-service and --send-data. Spool to separate files with .TS suffixes. Use /tmp until --lib is valid. Daemonize early. Replace __TS_MICRO__ in task options.

bin/pt-agent

@@ -4724,14 +4724,42 @@ sub main {
       exit $exit_status;
    }
 
+   # ######################################################################## 
+   # Daemonize first so all output goes to the --log.
+   # ######################################################################## 
+   my $daemon;
+   if ( !$o->get('send-data') ) {
+      if ( $o->get('daemonize') ) {
+         $daemon = new Daemon(o=>$o);
+         $daemon->daemonize();
+         PTDEBUG && _d('I am a daemon now');
+      }
+      elsif ( $o->get('pid') ) {
+         $daemon = new Daemon(o=>$o);
+         $daemon->make_PID_file();
+      }
+      # If we daemonized, the parent has already exited and we're the child.
+      # We shared a copy of every Cxn with the parent, and the parent's copies
+      # were destroyed but the dbhs were not disconnected because the parent
+      # attrib was true.  Now, as the child, set it false so the dbhs will be
+      # disconnected when our Cxn copies are destroyed.  If we didn't daemonize,
+      # then we're not really a parent (since we have no children), so set it
+      # false to auto-disconnect the dbhs when our Cxns are destroyed.
+      $cxn->{parent} = 0;
+   }
+
    # ########################################################################
    # Connect to the Percona web API.
    # ########################################################################
+   
+   # TODO: --send-data should not use this because it calls init_agent()
+   my $lib_dir = $o->get('lib');
+
    my ($client, $agent);
    eval {
       ($client, $agent) = connect_to_percona(
          api_key    => $api_key,
-         lib_dir    => $o->get('lib'),
+         lib_dir    => $lib_dir,
          Cxn        => $cxn,
          agent_uuid => $o->get('agent-uuid'),  # optional
       );
@@ -4765,26 +4793,6 @@ sub main {
    # retried forever.
    # ########################################################################
 
-   # Daemonize first so all output goes to the --log.
-   my $daemon;
-   if ( $o->get('daemonize') ) {
-      $daemon = new Daemon(o=>$o);
-      $daemon->daemonize();
-      PTDEBUG && _d('I am a daemon now');
-   }
-   elsif ( $o->get('pid') ) {
-      $daemon = new Daemon(o=>$o);
-      $daemon->make_PID_file();
-   }
-   # If we daemonized, the parent has already exited and we're the child.
-   # We shared a copy of every Cxn with the parent, and the parent's copies
-   # were destroyed but the dbhs were not disconnected because the parent
-   # attrib was true.  Now, as the child, set it false so the dbhs will be
-   # disconnected when our Cxn copies are destroyed.  If we didn't daemonize,
-   # then we're not really a parent (since we have no children), so set it
-   # false to auto-disconnect the dbhs when our Cxns are destroyed.
-   $cxn->{parent} = 0;
-
    # Check and init the config file.
    my $config_file = get_config_file();
    _info("Config file: $config_file");
@@ -4806,6 +4814,15 @@ sub main {
       }
    }
 
+   # Check --lib.  Until the agent is configured, the default lib dir
+   # may not work.  Save stuff in /tmp so if we stop and start again
+   # we don't try to create a new agent again.
+   if ( !-d $lib_dir || !-w $lib_dir ) {
+      _info("--lib $lib_dir does not exist or is not writeable,"
+         . " using /tmp until the agent is configured");
+      $lib_dir = '/tmp';
+   }
+
    # Wait time between checking for new config and services.
    # Use the tool's built-in default until a config is gotten,
    # then config->{check-interval} will be pass in.
@@ -4824,7 +4841,7 @@ sub main {
       agent    => $agent,
       client   => $client,
       interval => $check_wait,
-      lib_dir  => $o->get('lib'),
+      lib_dir  => $lib_dir,
    );
 
    _info("pt-agent exit $exit_status, oktorun $oktorun");
@@ -5056,7 +5073,12 @@ sub init_agent {
       link => $agent_uri,
    );
 
-   _info("Agent " . $agent->name . " initialized and ready");
+   save_agent(
+      agent   => $agent,
+      lib_dir => $lib_dir,
+   );
+
+   _info("Agent " . $agent->name . "(" . $agent->uuid . ") is ready");
    return $agent;
 }
 
@@ -5545,7 +5567,14 @@ sub run_service {
    my $lib_dir   = $args{lib_dir};
    my $cxn       = $args{Cxn};
 
-   # TODO: where should this output go?
+   my $log_file = "$lib_dir/$service.run.log";
+   close STDOUT;
+   open  STDOUT, '>>', $log_file
+      or die "Cannot open log file $log_file: $OS_ERROR";
+   close STDERR;
+   open  STDERR, ">&STDOUT"
+      or die "Cannot dupe STDERR to STDOUT: $OS_ERROR";
+
    _info("Running $service service");
 
    $service = load_service(
@@ -5574,7 +5603,7 @@ sub run_service {
 
    my @output_files;
    my $final_exit_status = 0;
-   my $spool_file        = "$spool_dir/" . $service->name;
+   my $spool_file        = "$spool_dir/" . $service->name . '.' . int(time);
    my $taskno            = 0;
    TASK:
    foreach my $task ( @$tasks ) {
@@ -5689,6 +5718,8 @@ sub replace_special_vars {
    my $cmd          = $args{cmd};
    my $output_files = $args{output_files};
 
+   my $ts_micro = sprintf('%.6f', time);
+
    my $new_cmd = join(' ',
       map {
          my $word = $_;
@@ -5700,6 +5731,7 @@ sub replace_special_vars {
                die "Run$runno has no output for $word to access.\n";
             }
          }
+         $word =~ s/__TS_MICRO__/$ts_micro/g;
          $word;
       }
       split(/\s+/, $cmd)
@@ -5737,6 +5769,14 @@ sub send_data {
    my $service_dir  = $spool_dir . '/' . $service;
    my $service_file = $lib_dir . '/services/' . $service;
 
+   my $log_file = "$lib_dir/$service.send.log";
+   close STDOUT;
+   open  STDOUT, '>>', $log_file
+      or die "Cannot open log file $log_file: $OS_ERROR";
+   close STDERR;
+   open  STDERR, ">&STDOUT"
+      or die "Cannot dupe STDERR to STDOUT: $OS_ERROR";
+
    # Re-create the Service resource object from the saved service file.   
    # TODO: test
    if ( !-f $service_file ) {