http://git-wip-us.apache.org/repos/asf/trafodion-site/blob/fb76ef66/docs/dcs_reference/xref/org/trafodion/dcs/master/ServerManager.html ---------------------------------------------------------------------- diff --git a/docs/dcs_reference/xref/org/trafodion/dcs/master/ServerManager.html b/docs/dcs_reference/xref/org/trafodion/dcs/master/ServerManager.html index e607d4e..65a572e 100644 --- a/docs/dcs_reference/xref/org/trafodion/dcs/master/ServerManager.html +++ b/docs/dcs_reference/xref/org/trafodion/dcs/master/ServerManager.html @@ -33,778 +33,822 @@ <a class="jxr_linenumber" name="23" href="#23">23</a> <strong class="jxr_keyword">package</strong> org.trafodion.dcs.master; <a class="jxr_linenumber" name="24" href="#24">24</a> <a class="jxr_linenumber" name="25" href="#25">25</a> <strong class="jxr_keyword">import</strong> java.net.InetAddress; -<a class="jxr_linenumber" name="26" href="#26">26</a> -<a class="jxr_linenumber" name="27" href="#27">27</a> <strong class="jxr_keyword">import</strong> java.io.IOException; -<a class="jxr_linenumber" name="28" href="#28">28</a> <strong class="jxr_keyword">import</strong> java.io.InputStream; -<a class="jxr_linenumber" name="29" href="#29">29</a> <strong class="jxr_keyword">import</strong> java.io.InputStreamReader; -<a class="jxr_linenumber" name="30" href="#30">30</a> <strong class="jxr_keyword">import</strong> java.io.BufferedReader; -<a class="jxr_linenumber" name="31" href="#31">31</a> <strong class="jxr_keyword">import</strong> java.io.FileReader; -<a class="jxr_linenumber" name="32" href="#32">32</a> <strong class="jxr_keyword">import</strong> java.io.FileNotFoundException; -<a class="jxr_linenumber" name="33" href="#33">33</a> -<a class="jxr_linenumber" name="34" href="#34">34</a> <strong class="jxr_keyword">import</strong> java.util.Scanner; -<a class="jxr_linenumber" name="35" href="#35">35</a> <strong class="jxr_keyword">import</strong> java.util.Collections; -<a class="jxr_linenumber" name="36" href="#36">36</a> <strong class="jxr_keyword">import</strong> java.util.Iterator; -<a class="jxr_linenumber" name="37" href="#37">37</a> <strong class="jxr_keyword">import</strong> java.util.LinkedList; -<a class="jxr_linenumber" name="38" href="#38">38</a> <strong class="jxr_keyword">import</strong> java.util.Queue; -<a class="jxr_linenumber" name="39" href="#39">39</a> <strong class="jxr_keyword">import</strong> java.util.List; -<a class="jxr_linenumber" name="40" href="#40">40</a> <strong class="jxr_keyword">import</strong> java.util.ArrayList; -<a class="jxr_linenumber" name="41" href="#41">41</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.Callable; -<a class="jxr_linenumber" name="42" href="#42">42</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.Executors; -<a class="jxr_linenumber" name="43" href="#43">43</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.ExecutorService; -<a class="jxr_linenumber" name="44" href="#44">44</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.Future; -<a class="jxr_linenumber" name="45" href="#45">45</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.ExecutionException; -<a class="jxr_linenumber" name="46" href="#46">46</a> <strong class="jxr_keyword">import</strong> java.util.Date; -<a class="jxr_linenumber" name="47" href="#47">47</a> <strong class="jxr_keyword">import</strong> java.util.Comparator; -<a class="jxr_linenumber" name="48" href="#48">48</a> <strong class="jxr_keyword">import</strong> java.util.Map; -<a class="jxr_linenumber" name="49" href="#49">49</a> <strong class="jxr_keyword">import</strong> java.util.HashMap; -<a class="jxr_linenumber" name="50" href="#50">50</a> -<a class="jxr_linenumber" name="51" href="#51">51</a> <strong class="jxr_keyword">import</strong> java.text.DateFormat; -<a class="jxr_linenumber" name="52" href="#52">52</a> -<a class="jxr_linenumber" name="53" href="#53">53</a> <strong class="jxr_keyword">import</strong> org.apache.zookeeper.*; -<a class="jxr_linenumber" name="54" href="#54">54</a> <strong class="jxr_keyword">import</strong> org.apache.zookeeper.data.Stat; -<a class="jxr_linenumber" name="55" href="#55">55</a> -<a class="jxr_linenumber" name="56" href="#56">56</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.conf.Configuration; -<a class="jxr_linenumber" name="57" href="#57">57</a> -<a class="jxr_linenumber" name="58" href="#58">58</a> <strong class="jxr_keyword">import</strong> org.apache.commons.logging.Log; -<a class="jxr_linenumber" name="59" href="#59">59</a> <strong class="jxr_keyword">import</strong> org.apache.commons.logging.LogFactory; -<a class="jxr_linenumber" name="60" href="#60">60</a> -<a class="jxr_linenumber" name="61" href="#61">61</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.master.RunningServer; -<a class="jxr_linenumber" name="62" href="#62">62</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.master.RegisteredServer; -<a class="jxr_linenumber" name="63" href="#63">63</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.master.Metrics; -<a class="jxr_linenumber" name="64" href="#64">64</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.script.ScriptManager; -<a class="jxr_linenumber" name="65" href="#65">65</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.script.ScriptContext; -<a class="jxr_linenumber" name="66" href="#66">66</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.Constants; -<a class="jxr_linenumber" name="67" href="#67">67</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.zookeeper.ZkClient; -<a class="jxr_linenumber" name="68" href="#68">68</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.util.*; -<a class="jxr_linenumber" name="69" href="#69">69</a> -<a class="jxr_linenumber" name="70" href="#70">70</a> <strong class="jxr_keyword">import</strong> org.codehaus.jettison.json.JSONArray; -<a class="jxr_linenumber" name="71" href="#71">71</a> <strong class="jxr_keyword">import</strong> org.codehaus.jettison.json.JSONException; -<a class="jxr_linenumber" name="72" href="#72">72</a> <strong class="jxr_keyword">import</strong> org.codehaus.jettison.json.JSONObject; -<a class="jxr_linenumber" name="73" href="#73">73</a> -<a class="jxr_linenumber" name="74" href="#74">74</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">class</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">ServerManager</a> <strong class="jxr_keyword">implements</strong> Callable { -<a class="jxr_linenumber" name="75" href="#75">75</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">static</strong> <strong class="jxr_keyword">final</strong> Log LOG = LogFactory.getLog(ServerManager.<strong class="jxr_keyword">class</strong>); -<a class="jxr_linenumber" name="76" href="#76">76</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/master/DcsMaster.html">DcsMaster</a> master; -<a class="jxr_linenumber" name="77" href="#77">77</a> <strong class="jxr_keyword">private</strong> Configuration conf; -<a class="jxr_linenumber" name="78" href="#78">78</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/util/DcsNetworkConfiguration.html">DcsNetworkConfiguration</a> netConf; -<a class="jxr_linenumber" name="79" href="#79">79</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/zookeeper/ZkClient.html">ZkClient</a> zkc = <strong class="jxr_keyword">null</strong>; -<a class="jxr_linenumber" name="80" href="#80">80</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">long</strong> startupTimestamp; -<a class="jxr_linenumber" name="81" href="#81">81</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">int</strong> maxRestartAttempts; -<a class="jxr_linenumber" name="82" href="#82">82</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">int</strong> retryIntervalMillis; -<a class="jxr_linenumber" name="83" href="#83">83</a> <strong class="jxr_keyword">private</strong> ExecutorService pool = <strong class="jxr_keyword">null</strong>; -<a class="jxr_linenumber" name="84" href="#84">84</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/master/Metrics.html">Metrics</a> metrics; -<a class="jxr_linenumber" name="85" href="#85">85</a> <strong class="jxr_keyword">private</strong> String parentZnode; -<a class="jxr_linenumber" name="86" href="#86">86</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/util/RetryCounterFactory.html">RetryCounterFactory</a> retryCounterFactory; -<a class="jxr_linenumber" name="87" href="#87">87</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ArrayList<String> configuredServers = <strong class="jxr_keyword">new</strong> ArrayList<String>(); -<a class="jxr_linenumber" name="88" href="#88">88</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> Map<String, ServerPortMap> serverPortMap = <strong class="jxr_keyword">new</strong> HashMap<String, ServerPortMap>(); -<a class="jxr_linenumber" name="89" href="#89">89</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ArrayList<String> runningServers = <strong class="jxr_keyword">new</strong> ArrayList<String>(); -<a class="jxr_linenumber" name="90" href="#90">90</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ArrayList<String> registeredServers = <strong class="jxr_keyword">new</strong> ArrayList<String>(); -<a class="jxr_linenumber" name="91" href="#91">91</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> Queue<RestartHandler> restartQueue = <strong class="jxr_keyword">new</strong> LinkedList<RestartHandler>(); -<a class="jxr_linenumber" name="92" href="#92">92</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ArrayList<ServerItem> serverItemList = <strong class="jxr_keyword">new</strong> ArrayList<ServerItem>(); -<a class="jxr_linenumber" name="93" href="#93">93</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">boolean</strong> trafodionQueryToolsEnabled; -<a class="jxr_linenumber" name="94" href="#94">94</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/util/JdbcT4Util.html">JdbcT4Util</a> jdbcT4Util = <strong class="jxr_keyword">null</strong>; -<a class="jxr_linenumber" name="95" href="#95">95</a> -<a class="jxr_linenumber" name="96" href="#96">96</a> <strong class="jxr_keyword">public</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">ServerManager</a>(<a href="../../../../org/trafodion/dcs/master/DcsMaster.html">DcsMaster</a> master, Configuration conf, <a href="../../../../org/trafodion/dcs/zookeeper/ZkClient.html">ZkClient</a> zkc, -<a class="jxr_linenumber" name="97" href="#97">97</a> <a href="../../../../org/trafodion/dcs/util/DcsNetworkConfiguration.html">DcsNetworkConfiguration</a> netConf, <strong class="jxr_keyword">long</strong> startupTimestamp, -<a class="jxr_linenumber" name="98" href="#98">98</a> <a href="../../../../org/trafodion/dcs/master/Metrics.html">Metrics</a> metrics) <strong class="jxr_keyword">throws</strong> Exception { -<a class="jxr_linenumber" name="99" href="#99">99</a> <strong class="jxr_keyword">try</strong> { -<a class="jxr_linenumber" name="100" href="#100">100</a> <strong class="jxr_keyword">this</strong>.master = master; -<a class="jxr_linenumber" name="101" href="#101">101</a> <strong class="jxr_keyword">this</strong>.conf = conf; -<a class="jxr_linenumber" name="102" href="#102">102</a> <strong class="jxr_keyword">this</strong>.zkc = zkc; -<a class="jxr_linenumber" name="103" href="#103">103</a> <strong class="jxr_keyword">this</strong>.netConf = netConf; -<a class="jxr_linenumber" name="104" href="#104">104</a> <strong class="jxr_keyword">this</strong>.startupTimestamp = startupTimestamp; -<a class="jxr_linenumber" name="105" href="#105">105</a> <strong class="jxr_keyword">this</strong>.metrics = metrics; -<a class="jxr_linenumber" name="106" href="#106">106</a> maxRestartAttempts = conf -<a class="jxr_linenumber" name="107" href="#107">107</a> .getInt(Constants.DCS_MASTER_SERVER_RESTART_HANDLER_ATTEMPTS, -<a class="jxr_linenumber" name="108" href="#108">108</a> Constants.DEFAULT_DCS_MASTER_SERVER_RESTART_HANDLER_ATTEMPTS); -<a class="jxr_linenumber" name="109" href="#109">109</a> retryIntervalMillis = conf -<a class="jxr_linenumber" name="110" href="#110">110</a> .getInt(Constants.DCS_MASTER_SERVER_RESTART_HANDLER_RETRY_INTERVAL_MILLIS, -<a class="jxr_linenumber" name="111" href="#111">111</a> Constants.DEFAULT_DCS_MASTER_SERVER_RESTART_HANDLER_RETRY_INTERVAL_MILLIS); -<a class="jxr_linenumber" name="112" href="#112">112</a> trafodionQueryToolsEnabled = conf.getBoolean( -<a class="jxr_linenumber" name="113" href="#113">113</a> Constants.DCS_MASTER_TRAFODION_QUERY_TOOLS, -<a class="jxr_linenumber" name="114" href="#114">114</a> Constants.DEFAULT_DCS_MASTER_TRAFODION_QUERY_TOOLS); -<a class="jxr_linenumber" name="115" href="#115">115</a> <strong class="jxr_keyword">if</strong> (trafodionQueryToolsEnabled) -<a class="jxr_linenumber" name="116" href="#116">116</a> jdbcT4Util = <strong class="jxr_keyword">new</strong> <a href="../../../../org/trafodion/dcs/util/JdbcT4Util.html">JdbcT4Util</a>(conf, netConf); -<a class="jxr_linenumber" name="117" href="#117">117</a> retryCounterFactory = <strong class="jxr_keyword">new</strong> <a href="../../../../org/trafodion/dcs/util/RetryCounterFactory.html">RetryCounterFactory</a>(maxRestartAttempts, -<a class="jxr_linenumber" name="118" href="#118">118</a> retryIntervalMillis); -<a class="jxr_linenumber" name="119" href="#119">119</a> parentZnode = conf.get(Constants.ZOOKEEPER_ZNODE_PARENT, -<a class="jxr_linenumber" name="120" href="#120">120</a> Constants.DEFAULT_ZOOKEEPER_ZNODE_PARENT); -<a class="jxr_linenumber" name="121" href="#121">121</a> pool = Executors.newSingleThreadExecutor(); -<a class="jxr_linenumber" name="122" href="#122">122</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { -<a class="jxr_linenumber" name="123" href="#123">123</a> e.printStackTrace(); -<a class="jxr_linenumber" name="124" href="#124">124</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) -<a class="jxr_linenumber" name="125" href="#125">125</a> LOG.error(e); -<a class="jxr_linenumber" name="126" href="#126">126</a> <strong class="jxr_keyword">throw</strong> e; -<a class="jxr_linenumber" name="127" href="#127">127</a> } -<a class="jxr_linenumber" name="128" href="#128">128</a> } -<a class="jxr_linenumber" name="129" href="#129">129</a> -<a class="jxr_linenumber" name="130" href="#130">130</a> <strong class="jxr_keyword">class</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RestartHandler</a> <strong class="jxr_keyword">implements</strong> Callable<ScriptContext> { -<a class="jxr_linenumber" name="131" href="#131">131</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/script/ScriptContext.html">ScriptContext</a> scriptContext = <strong class="jxr_keyword">new</strong> <a href="../../../../org/trafodion/dcs/script/ScriptContext.html">ScriptContext</a>(); -<a class="jxr_linenumber" name="132" href="#132">132</a> <strong class="jxr_keyword">private</strong> String znodePath; -<a class="jxr_linenumber" name="133" href="#133">133</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">int</strong> childCount; -<a class="jxr_linenumber" name="134" href="#134">134</a> -<a class="jxr_linenumber" name="135" href="#135">135</a> <strong class="jxr_keyword">public</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RestartHandler</a>(String znodePath, <strong class="jxr_keyword">int</strong> childCount) { -<a class="jxr_linenumber" name="136" href="#136">136</a> <strong class="jxr_keyword">this</strong>.znodePath = znodePath; -<a class="jxr_linenumber" name="137" href="#137">137</a> <strong class="jxr_keyword">this</strong>.childCount = childCount; -<a class="jxr_linenumber" name="138" href="#138">138</a> } -<a class="jxr_linenumber" name="139" href="#139">139</a> -<a class="jxr_linenumber" name="140" href="#140">140</a> @Override -<a class="jxr_linenumber" name="141" href="#141">141</a> <strong class="jxr_keyword">public</strong> <a href="../../../../org/trafodion/dcs/script/ScriptContext.html">ScriptContext</a> call() <strong class="jxr_keyword">throws</strong> Exception { -<a class="jxr_linenumber" name="142" href="#142">142</a> <strong class="jxr_keyword">try</strong> { -<a class="jxr_linenumber" name="143" href="#143">143</a> Scanner scn = <strong class="jxr_keyword">new</strong> Scanner(znodePath); -<a class="jxr_linenumber" name="144" href="#144">144</a> scn.useDelimiter(<span class="jxr_string">":"</span>); -<a class="jxr_linenumber" name="145" href="#145">145</a> String hostName = scn.next();<em class="jxr_comment">// host name</em> -<a class="jxr_linenumber" name="146" href="#146">146</a> String instance = scn.next();<em class="jxr_comment">// instance</em> -<a class="jxr_linenumber" name="147" href="#147">147</a> <strong class="jxr_keyword">int</strong> infoPort = Integer.parseInt(scn.next()); <em class="jxr_comment">// UI port</em> -<a class="jxr_linenumber" name="148" href="#148">148</a> <strong class="jxr_keyword">long</strong> serverStartTimestamp = Long.parseLong(scn.next()); -<a class="jxr_linenumber" name="149" href="#149">149</a> scn.close(); -<a class="jxr_linenumber" name="150" href="#150">150</a> -<a class="jxr_linenumber" name="151" href="#151">151</a> <em class="jxr_comment">// Get the --config property from classpath...it's always first</em> -<a class="jxr_linenumber" name="152" href="#152">152</a> <em class="jxr_comment">// in the classpath</em> -<a class="jxr_linenumber" name="153" href="#153">153</a> String cp = System.getProperty(<span class="jxr_string">"java.class.path"</span>); -<a class="jxr_linenumber" name="154" href="#154">154</a> scn = <strong class="jxr_keyword">new</strong> Scanner(cp); -<a class="jxr_linenumber" name="155" href="#155">155</a> scn.useDelimiter(<span class="jxr_string">":"</span>); -<a class="jxr_linenumber" name="156" href="#156">156</a> String confDir = scn.next(); -<a class="jxr_linenumber" name="157" href="#157">157</a> scn.close(); -<a class="jxr_linenumber" name="158" href="#158">158</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) -<a class="jxr_linenumber" name="159" href="#159">159</a> LOG.debug(<span class="jxr_string">"conf dir ["</span> + confDir + <span class="jxr_string">"]"</span>); -<a class="jxr_linenumber" name="160" href="#160">160</a> -<a class="jxr_linenumber" name="161" href="#161">161</a> <em class="jxr_comment">// Get -Ddcs.home.dir</em> -<a class="jxr_linenumber" name="162" href="#162">162</a> String dcsHome = System.getProperty(<span class="jxr_string">"dcs.home.dir"</span>); -<a class="jxr_linenumber" name="163" href="#163">163</a> -<a class="jxr_linenumber" name="164" href="#164">164</a> <em class="jxr_comment">// If stop-dcs.sh is executed and DCS_MANAGES_ZK then zookeeper</em> -<a class="jxr_linenumber" name="165" href="#165">165</a> <em class="jxr_comment">// is stopped abruptly.</em> -<a class="jxr_linenumber" name="166" href="#166">166</a> <em class="jxr_comment">// Second scenario is when ZooKeeper fails for some reason</em> -<a class="jxr_linenumber" name="167" href="#167">167</a> <em class="jxr_comment">// regardless of whether DCS</em> -<a class="jxr_linenumber" name="168" href="#168">168</a> <em class="jxr_comment">// manages it. When either happens the DcsServer running znodes</em> -<a class="jxr_linenumber" name="169" href="#169">169</a> <em class="jxr_comment">// still exist in ZooKeeper</em> -<a class="jxr_linenumber" name="170" href="#170">170</a> <em class="jxr_comment">// and we see them at next startup. When they eventually timeout</em> -<a class="jxr_linenumber" name="171" href="#171">171</a> <em class="jxr_comment">// we get node deleted events for a server that no longer</em> -<a class="jxr_linenumber" name="172" href="#172">172</a> <em class="jxr_comment">// exists. So, only recognize</em> -<a class="jxr_linenumber" name="173" href="#173">173</a> <em class="jxr_comment">// DcsServer running znodes that have timestamps after last</em> -<a class="jxr_linenumber" name="174" href="#174">174</a> <em class="jxr_comment">// DcsMaster startup.</em> -<a class="jxr_linenumber" name="175" href="#175">175</a> <em class="jxr_comment">//</em> -<a class="jxr_linenumber" name="176" href="#176">176</a> <em class="jxr_comment">// But, if we are DcsMaster follower that is taking over from</em> -<a class="jxr_linenumber" name="177" href="#177">177</a> <em class="jxr_comment">// failed one then ignore timestamp issues described above.</em> -<a class="jxr_linenumber" name="178" href="#178">178</a> <em class="jxr_comment">// See MasterLeaderElection.elect()</em> -<a class="jxr_linenumber" name="179" href="#179">179</a> <strong class="jxr_keyword">if</strong> ((master.isFollower() == false && serverStartTimestamp > startupTimestamp) -<a class="jxr_linenumber" name="180" href="#180">180</a> || (master.isFollower() && runningServers.size() < configuredServers.size())) { -<a class="jxr_linenumber" name="181" href="#181">181</a> scriptContext.setHostName(hostName); -<a class="jxr_linenumber" name="182" href="#182">182</a> scriptContext -<a class="jxr_linenumber" name="183" href="#183">183</a> .setScriptName(Constants.SYS_SHELL_SCRIPT_NAME); -<a class="jxr_linenumber" name="184" href="#184">184</a> -<a class="jxr_linenumber" name="185" href="#185">185</a> <strong class="jxr_keyword">if</strong> (hostName.equalsIgnoreCase(netConf.getHostName())) -<a class="jxr_linenumber" name="186" href="#186">186</a> scriptContext.setCommand(<span class="jxr_string">"bin/dcs-daemon.sh --config "</span> -<a class="jxr_linenumber" name="187" href="#187">187</a> + confDir + <span class="jxr_string">" start server "</span> + instance + <span class="jxr_string">" "</span> -<a class="jxr_linenumber" name="188" href="#188">188</a> + childCount); -<a class="jxr_linenumber" name="189" href="#189">189</a> <strong class="jxr_keyword">else</strong> -<a class="jxr_linenumber" name="190" href="#190">190</a> scriptContext.setCommand(<span class="jxr_string">"pdsh -w "</span> + hostName -<a class="jxr_linenumber" name="191" href="#191">191</a> + <span class="jxr_string">" \"cd "</span> + dcsHome -<a class="jxr_linenumber" name="192" href="#192">192</a> + <span class="jxr_string">";bin/dcs-daemon.sh --config "</span> + confDir -<a class="jxr_linenumber" name="193" href="#193">193</a> + <span class="jxr_string">" start server "</span> + instance + <span class="jxr_string">" "</span> -<a class="jxr_linenumber" name="194" href="#194">194</a> + childCount + <span class="jxr_string">"\""</span>); -<a class="jxr_linenumber" name="195" href="#195">195</a> -<a class="jxr_linenumber" name="196" href="#196">196</a> <a href="../../../../org/trafodion/dcs/util/RetryCounter.html">RetryCounter</a> retryCounter = retryCounterFactory.create(); -<a class="jxr_linenumber" name="197" href="#197">197</a> <strong class="jxr_keyword">while</strong> (<strong class="jxr_keyword">true</strong>) { -<a class="jxr_linenumber" name="198" href="#198">198</a> <strong class="jxr_keyword">if</strong> (scriptContext.getStdOut().length() > 0) -<a class="jxr_linenumber" name="199" href="#199">199</a> scriptContext.getStdOut().delete(0, -<a class="jxr_linenumber" name="200" href="#200">200</a> scriptContext.getStdOut().length()); -<a class="jxr_linenumber" name="201" href="#201">201</a> <strong class="jxr_keyword">if</strong> (scriptContext.getStdErr().length() > 0) -<a class="jxr_linenumber" name="202" href="#202">202</a> scriptContext.getStdErr().delete(0, -<a class="jxr_linenumber" name="203" href="#203">203</a> scriptContext.getStdErr().length()); -<a class="jxr_linenumber" name="204" href="#204">204</a> LOG.info(<span class="jxr_string">"Restarting DcsServer ["</span> + hostName + <span class="jxr_string">":"</span> -<a class="jxr_linenumber" name="205" href="#205">205</a> + instance + <span class="jxr_string">"], script [ "</span> -<a class="jxr_linenumber" name="206" href="#206">206</a> + scriptContext.toString() + <span class="jxr_string">" ]"</span>); -<a class="jxr_linenumber" name="207" href="#207">207</a> ScriptManager.getInstance().runScript(scriptContext); -<a class="jxr_linenumber" name="208" href="#208">208</a> -<a class="jxr_linenumber" name="209" href="#209">209</a> <strong class="jxr_keyword">if</strong> (scriptContext.getExitCode() == 0) { -<a class="jxr_linenumber" name="210" href="#210">210</a> LOG.info(<span class="jxr_string">"DcsServer ["</span> + hostName + <span class="jxr_string">":"</span> + instance -<a class="jxr_linenumber" name="211" href="#211">211</a> + <span class="jxr_string">"] restarted"</span>); -<a class="jxr_linenumber" name="212" href="#212">212</a> <strong class="jxr_keyword">break</strong>; -<a class="jxr_linenumber" name="213" href="#213">213</a> } <strong class="jxr_keyword">else</strong> { -<a class="jxr_linenumber" name="214" href="#214">214</a> StringBuilder sb = <strong class="jxr_keyword">new</strong> StringBuilder(); -<a class="jxr_linenumber" name="215" href="#215">215</a> sb.append(<span class="jxr_string">"exit code ["</span> -<a class="jxr_linenumber" name="216" href="#216">216</a> + scriptContext.getExitCode() + <span class="jxr_string">"]"</span>); -<a class="jxr_linenumber" name="217" href="#217">217</a> <strong class="jxr_keyword">if</strong> (!scriptContext.getStdOut().toString().isEmpty()) -<a class="jxr_linenumber" name="218" href="#218">218</a> sb.append(<span class="jxr_string">", stdout ["</span> -<a class="jxr_linenumber" name="219" href="#219">219</a> + scriptContext.getStdOut().toString() -<a class="jxr_linenumber" name="220" href="#220">220</a> + <span class="jxr_string">"]"</span>); -<a class="jxr_linenumber" name="221" href="#221">221</a> <strong class="jxr_keyword">if</strong> (!scriptContext.getStdErr().toString().isEmpty()) -<a class="jxr_linenumber" name="222" href="#222">222</a> sb.append(<span class="jxr_string">", stderr ["</span> -<a class="jxr_linenumber" name="223" href="#223">223</a> + scriptContext.getStdErr().toString() -<a class="jxr_linenumber" name="224" href="#224">224</a> + <span class="jxr_string">"]"</span>); -<a class="jxr_linenumber" name="225" href="#225">225</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) -<a class="jxr_linenumber" name="226" href="#226">226</a> LOG.error(sb.toString()); -<a class="jxr_linenumber" name="227" href="#227">227</a> -<a class="jxr_linenumber" name="228" href="#228">228</a> <strong class="jxr_keyword">if</strong> (!retryCounter.shouldRetry()) { -<a class="jxr_linenumber" name="229" href="#229">229</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) -<a class="jxr_linenumber" name="230" href="#230">230</a> LOG.error(<span class="jxr_string">"DcsServer ["</span> + hostName + <span class="jxr_string">":"</span> -<a class="jxr_linenumber" name="231" href="#231">231</a> + instance -<a class="jxr_linenumber" name="232" href="#232">232</a> + <span class="jxr_string">"] restart failed after "</span> -<a class="jxr_linenumber" name="233" href="#233">233</a> + retryCounter.getMaxRetries() -<a class="jxr_linenumber" name="234" href="#234">234</a> + <span class="jxr_string">" retries"</span>); -<a class="jxr_linenumber" name="235" href="#235">235</a> <strong class="jxr_keyword">break</strong>; -<a class="jxr_linenumber" name="236" href="#236">236</a> } <strong class="jxr_keyword">else</strong> { -<a class="jxr_linenumber" name="237" href="#237">237</a> retryCounter.sleepUntilNextRetry(); -<a class="jxr_linenumber" name="238" href="#238">238</a> retryCounter.useRetry(); -<a class="jxr_linenumber" name="239" href="#239">239</a> } -<a class="jxr_linenumber" name="240" href="#240">240</a> } -<a class="jxr_linenumber" name="241" href="#241">241</a> } -<a class="jxr_linenumber" name="242" href="#242">242</a> } <strong class="jxr_keyword">else</strong> { -<a class="jxr_linenumber" name="243" href="#243">243</a> StringBuffer sb = <strong class="jxr_keyword">new</strong> StringBuffer(); -<a class="jxr_linenumber" name="244" href="#244">244</a> sb.append(<span class="jxr_string">"No restart for "</span>).append(znodePath).append(System.getProperty(<span class="jxr_string">"line.separator"</span>)); -<a class="jxr_linenumber" name="245" href="#245">245</a> sb.append(<span class="jxr_string">"DCS Master isFollower ["</span>).append(master.isFollower()).append(<span class="jxr_string">"], "</span>); -<a class="jxr_linenumber" name="246" href="#246">246</a> sb.append(<span class="jxr_string">"DCS Master start time ["</span>) -<a class="jxr_linenumber" name="247" href="#247">247</a> .append(DateFormat.getDateTimeInstance().format(<strong class="jxr_keyword">new</strong> Date(startupTimestamp))).append(<span class="jxr_string">"], "</span>); -<a class="jxr_linenumber" name="248" href="#248">248</a> sb.append(<span class="jxr_string">"DCS Server start time ["</span>) -<a class="jxr_linenumber" name="249" href="#249">249</a> .append(DateFormat.getDateTimeInstance().format(<strong class="jxr_keyword">new</strong> Date(serverStartTimestamp))).append(<span class="jxr_string">"], "</span>); -<a class="jxr_linenumber" name="250" href="#250">250</a> sb.append(<span class="jxr_string">"running DCS Server num is ["</span>).append(runningServers.size()) -<a class="jxr_linenumber" name="251" href="#251">251</a> .append(<span class="jxr_string">"], registered DCS Server num is ["</span>).append(registeredServers.size()).append(<span class="jxr_string">"]."</span>); -<a class="jxr_linenumber" name="252" href="#252">252</a> -<a class="jxr_linenumber" name="253" href="#253">253</a> LOG.info(sb.toString()); -<a class="jxr_linenumber" name="254" href="#254">254</a> } -<a class="jxr_linenumber" name="255" href="#255">255</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { -<a class="jxr_linenumber" name="256" href="#256">256</a> e.printStackTrace(); -<a class="jxr_linenumber" name="257" href="#257">257</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) -<a class="jxr_linenumber" name="258" href="#258">258</a> LOG.error(e); -<a class="jxr_linenumber" name="259" href="#259">259</a> } -<a class="jxr_linenumber" name="260" href="#260">260</a> -<a class="jxr_linenumber" name="261" href="#261">261</a> <strong class="jxr_keyword">return</strong> scriptContext; -<a class="jxr_linenumber" name="262" href="#262">262</a> } -<a class="jxr_linenumber" name="263" href="#263">263</a> } -<a class="jxr_linenumber" name="264" href="#264">264</a> -<a class="jxr_linenumber" name="265" href="#265">265</a> <strong class="jxr_keyword">class</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RunningWatcher</a> <strong class="jxr_keyword">implements</strong> Watcher { -<a class="jxr_linenumber" name="266" href="#266">266</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> process(WatchedEvent event) { -<a class="jxr_linenumber" name="267" href="#267">267</a> <strong class="jxr_keyword">if</strong> (event.getType() == Event.EventType.NodeChildrenChanged) { -<a class="jxr_linenumber" name="268" href="#268">268</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) -<a class="jxr_linenumber" name="269" href="#269">269</a> LOG.debug(<span class="jxr_string">"Running children changed ["</span> + event.getPath() -<a class="jxr_linenumber" name="270" href="#270">270</a> + <span class="jxr_string">"]"</span>); -<a class="jxr_linenumber" name="271" href="#271">271</a> <strong class="jxr_keyword">try</strong> { -<a class="jxr_linenumber" name="272" href="#272">272</a> getZkRunning(); -<a class="jxr_linenumber" name="273" href="#273">273</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { -<a class="jxr_linenumber" name="274" href="#274">274</a> e.printStackTrace(); -<a class="jxr_linenumber" name="275" href="#275">275</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) -<a class="jxr_linenumber" name="276" href="#276">276</a> LOG.error(e); -<a class="jxr_linenumber" name="277" href="#277">277</a> } -<a class="jxr_linenumber" name="278" href="#278">278</a> } <strong class="jxr_keyword">else</strong> <strong class="jxr_keyword">if</strong> (event.getType() == Event.EventType.NodeDeleted) { -<a class="jxr_linenumber" name="279" href="#279">279</a> String znodePath = event.getPath(); -<a class="jxr_linenumber" name="280" href="#280">280</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) -<a class="jxr_linenumber" name="281" href="#281">281</a> LOG.debug(<span class="jxr_string">"Running znode deleted ["</span> + znodePath + <span class="jxr_string">"]"</span>); -<a class="jxr_linenumber" name="282" href="#282">282</a> <strong class="jxr_keyword">try</strong> { -<a class="jxr_linenumber" name="283" href="#283">283</a> restartServer(znodePath); -<a class="jxr_linenumber" name="284" href="#284">284</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { -<a class="jxr_linenumber" name="285" href="#285">285</a> e.printStackTrace(); -<a class="jxr_linenumber" name="286" href="#286">286</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) -<a class="jxr_linenumber" name="287" href="#287">287</a> LOG.error(e); -<a class="jxr_linenumber" name="288" href="#288">288</a> } -<a class="jxr_linenumber" name="289" href="#289">289</a> } -<a class="jxr_linenumber" name="290" href="#290">290</a> } -<a class="jxr_linenumber" name="291" href="#291">291</a> } -<a class="jxr_linenumber" name="292" href="#292">292</a> -<a class="jxr_linenumber" name="293" href="#293">293</a> <strong class="jxr_keyword">class</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RegisteredWatcher</a> <strong class="jxr_keyword">implements</strong> Watcher { -<a class="jxr_linenumber" name="294" href="#294">294</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> process(WatchedEvent event) { -<a class="jxr_linenumber" name="295" href="#295">295</a> <strong class="jxr_keyword">if</strong> (event.getType() == Event.EventType.NodeChildrenChanged) { -<a class="jxr_linenumber" name="296" href="#296">296</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) -<a class="jxr_linenumber" name="297" href="#297">297</a> LOG.debug(<span class="jxr_string">"Registered children changed ["</span> + event.getPath() -<a class="jxr_linenumber" name="298" href="#298">298</a> + <span class="jxr_string">"]"</span>); -<a class="jxr_linenumber" name="299" href="#299">299</a> <strong class="jxr_keyword">try</strong> { -<a class="jxr_linenumber" name="300" href="#300">300</a> getZkRegistered(); -<a class="jxr_linenumber" name="301" href="#301">301</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { -<a class="jxr_linenumber" name="302" href="#302">302</a> e.printStackTrace(); -<a class="jxr_linenumber" name="303" href="#303">303</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) -<a class="jxr_linenumber" name="304" href="#304">304</a> LOG.error(e); -<a class="jxr_linenumber" name="305" href="#305">305</a> } -<a class="jxr_linenumber" name="306" href="#306">306</a> } -<a class="jxr_linenumber" name="307" href="#307">307</a> } -<a class="jxr_linenumber" name="308" href="#308">308</a> } -<a class="jxr_linenumber" name="309" href="#309">309</a> -<a class="jxr_linenumber" name="310" href="#310">310</a> @Override -<a class="jxr_linenumber" name="311" href="#311">311</a> <strong class="jxr_keyword">public</strong> Boolean call() <strong class="jxr_keyword">throws</strong> Exception { -<a class="jxr_linenumber" name="312" href="#312">312</a> -<a class="jxr_linenumber" name="313" href="#313">313</a> <strong class="jxr_keyword">long</strong> timeoutMillis = 5000; +<a class="jxr_linenumber" name="26" href="#26">26</a> <strong class="jxr_keyword">import</strong> java.io.IOException; +<a class="jxr_linenumber" name="27" href="#27">27</a> <strong class="jxr_keyword">import</strong> java.io.InputStream; +<a class="jxr_linenumber" name="28" href="#28">28</a> <strong class="jxr_keyword">import</strong> java.io.InputStreamReader; +<a class="jxr_linenumber" name="29" href="#29">29</a> <strong class="jxr_keyword">import</strong> java.io.BufferedReader; +<a class="jxr_linenumber" name="30" href="#30">30</a> <strong class="jxr_keyword">import</strong> java.io.FileReader; +<a class="jxr_linenumber" name="31" href="#31">31</a> <strong class="jxr_keyword">import</strong> java.io.FileNotFoundException; +<a class="jxr_linenumber" name="32" href="#32">32</a> <strong class="jxr_keyword">import</strong> java.util.Scanner; +<a class="jxr_linenumber" name="33" href="#33">33</a> <strong class="jxr_keyword">import</strong> java.util.Collections; +<a class="jxr_linenumber" name="34" href="#34">34</a> <strong class="jxr_keyword">import</strong> java.util.Iterator; +<a class="jxr_linenumber" name="35" href="#35">35</a> <strong class="jxr_keyword">import</strong> java.util.LinkedList; +<a class="jxr_linenumber" name="36" href="#36">36</a> <strong class="jxr_keyword">import</strong> java.util.Queue; +<a class="jxr_linenumber" name="37" href="#37">37</a> <strong class="jxr_keyword">import</strong> java.util.List; +<a class="jxr_linenumber" name="38" href="#38">38</a> <strong class="jxr_keyword">import</strong> java.util.ArrayList; +<a class="jxr_linenumber" name="39" href="#39">39</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.Callable; +<a class="jxr_linenumber" name="40" href="#40">40</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.Executors; +<a class="jxr_linenumber" name="41" href="#41">41</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.ExecutorService; +<a class="jxr_linenumber" name="42" href="#42">42</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.Future; +<a class="jxr_linenumber" name="43" href="#43">43</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.ExecutionException; +<a class="jxr_linenumber" name="44" href="#44">44</a> <strong class="jxr_keyword">import</strong> java.util.Date; +<a class="jxr_linenumber" name="45" href="#45">45</a> <strong class="jxr_keyword">import</strong> java.util.Comparator; +<a class="jxr_linenumber" name="46" href="#46">46</a> <strong class="jxr_keyword">import</strong> java.util.Map; +<a class="jxr_linenumber" name="47" href="#47">47</a> <strong class="jxr_keyword">import</strong> java.util.HashMap; +<a class="jxr_linenumber" name="48" href="#48">48</a> <strong class="jxr_keyword">import</strong> java.text.DateFormat; +<a class="jxr_linenumber" name="49" href="#49">49</a> +<a class="jxr_linenumber" name="50" href="#50">50</a> <strong class="jxr_keyword">import</strong> org.apache.zookeeper.*; +<a class="jxr_linenumber" name="51" href="#51">51</a> <strong class="jxr_keyword">import</strong> org.apache.zookeeper.data.Stat; +<a class="jxr_linenumber" name="52" href="#52">52</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.conf.Configuration; +<a class="jxr_linenumber" name="53" href="#53">53</a> <strong class="jxr_keyword">import</strong> org.apache.commons.logging.Log; +<a class="jxr_linenumber" name="54" href="#54">54</a> <strong class="jxr_keyword">import</strong> org.apache.commons.logging.LogFactory; +<a class="jxr_linenumber" name="55" href="#55">55</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.master.RunningServer; +<a class="jxr_linenumber" name="56" href="#56">56</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.master.RegisteredServer; +<a class="jxr_linenumber" name="57" href="#57">57</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.master.Metrics; +<a class="jxr_linenumber" name="58" href="#58">58</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.script.ScriptManager; +<a class="jxr_linenumber" name="59" href="#59">59</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.script.ScriptContext; +<a class="jxr_linenumber" name="60" href="#60">60</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.Constants; +<a class="jxr_linenumber" name="61" href="#61">61</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.zookeeper.ZkClient; +<a class="jxr_linenumber" name="62" href="#62">62</a> <strong class="jxr_keyword">import</strong> org.trafodion.dcs.util.*; +<a class="jxr_linenumber" name="63" href="#63">63</a> <strong class="jxr_keyword">import</strong> org.codehaus.jettison.json.JSONArray; +<a class="jxr_linenumber" name="64" href="#64">64</a> <strong class="jxr_keyword">import</strong> org.codehaus.jettison.json.JSONException; +<a class="jxr_linenumber" name="65" href="#65">65</a> <strong class="jxr_keyword">import</strong> org.codehaus.jettison.json.JSONObject; +<a class="jxr_linenumber" name="66" href="#66">66</a> +<a class="jxr_linenumber" name="67" href="#67">67</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">class</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">ServerManager</a> <strong class="jxr_keyword">implements</strong> Callable { +<a class="jxr_linenumber" name="68" href="#68">68</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">static</strong> <strong class="jxr_keyword">final</strong> Log LOG = LogFactory.getLog(ServerManager.<strong class="jxr_keyword">class</strong>); +<a class="jxr_linenumber" name="69" href="#69">69</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/master/DcsMaster.html">DcsMaster</a> master; +<a class="jxr_linenumber" name="70" href="#70">70</a> <strong class="jxr_keyword">private</strong> Configuration conf; +<a class="jxr_linenumber" name="71" href="#71">71</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/util/DcsNetworkConfiguration.html">DcsNetworkConfiguration</a> netConf; +<a class="jxr_linenumber" name="72" href="#72">72</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/zookeeper/ZkClient.html">ZkClient</a> zkc = <strong class="jxr_keyword">null</strong>; +<a class="jxr_linenumber" name="73" href="#73">73</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">long</strong> startupTimestamp; +<a class="jxr_linenumber" name="74" href="#74">74</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">int</strong> maxRestartAttempts; +<a class="jxr_linenumber" name="75" href="#75">75</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">int</strong> retryIntervalMillis; +<a class="jxr_linenumber" name="76" href="#76">76</a> <strong class="jxr_keyword">private</strong> ExecutorService pool = <strong class="jxr_keyword">null</strong>; +<a class="jxr_linenumber" name="77" href="#77">77</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/master/Metrics.html">Metrics</a> metrics; +<a class="jxr_linenumber" name="78" href="#78">78</a> <strong class="jxr_keyword">private</strong> String parentZnode; +<a class="jxr_linenumber" name="79" href="#79">79</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/util/RetryCounterFactory.html">RetryCounterFactory</a> retryCounterFactory; +<a class="jxr_linenumber" name="80" href="#80">80</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ArrayList<String> configuredServers = <strong class="jxr_keyword">new</strong> ArrayList<String>(); +<a class="jxr_linenumber" name="81" href="#81">81</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> Map<String, ServerPortMap> serverPortMap = <strong class="jxr_keyword">new</strong> HashMap<String, ServerPortMap>(); +<a class="jxr_linenumber" name="82" href="#82">82</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ArrayList<String> runningServers = <strong class="jxr_keyword">new</strong> ArrayList<String>(); +<a class="jxr_linenumber" name="83" href="#83">83</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ArrayList<String> registeredServers = <strong class="jxr_keyword">new</strong> ArrayList<String>(); +<a class="jxr_linenumber" name="84" href="#84">84</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> Queue<RestartHandler> restartQueue = <strong class="jxr_keyword">new</strong> LinkedList<RestartHandler>(); +<a class="jxr_linenumber" name="85" href="#85">85</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ArrayList<ServerItem> serverItemList = <strong class="jxr_keyword">new</strong> ArrayList<ServerItem>(); +<a class="jxr_linenumber" name="86" href="#86">86</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">boolean</strong> trafodionQueryToolsEnabled; +<a class="jxr_linenumber" name="87" href="#87">87</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/util/JdbcT4Util.html">JdbcT4Util</a> jdbcT4Util = <strong class="jxr_keyword">null</strong>; +<a class="jxr_linenumber" name="88" href="#88">88</a> +<a class="jxr_linenumber" name="89" href="#89">89</a> <strong class="jxr_keyword">public</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">ServerManager</a>(<a href="../../../../org/trafodion/dcs/master/DcsMaster.html">DcsMaster</a> master, Configuration conf, <a href="../../../../org/trafodion/dcs/zookeeper/ZkClient.html">ZkClient</a> zkc, +<a class="jxr_linenumber" name="90" href="#90">90</a> <a href="../../../../org/trafodion/dcs/util/DcsNetworkConfiguration.html">DcsNetworkConfiguration</a> netConf, <strong class="jxr_keyword">long</strong> startupTimestamp, +<a class="jxr_linenumber" name="91" href="#91">91</a> <a href="../../../../org/trafodion/dcs/master/Metrics.html">Metrics</a> metrics) <strong class="jxr_keyword">throws</strong> Exception { +<a class="jxr_linenumber" name="92" href="#92">92</a> <strong class="jxr_keyword">try</strong> { +<a class="jxr_linenumber" name="93" href="#93">93</a> <strong class="jxr_keyword">this</strong>.master = master; +<a class="jxr_linenumber" name="94" href="#94">94</a> <strong class="jxr_keyword">this</strong>.conf = conf; +<a class="jxr_linenumber" name="95" href="#95">95</a> <strong class="jxr_keyword">this</strong>.zkc = zkc; +<a class="jxr_linenumber" name="96" href="#96">96</a> <strong class="jxr_keyword">this</strong>.netConf = netConf; +<a class="jxr_linenumber" name="97" href="#97">97</a> <strong class="jxr_keyword">this</strong>.startupTimestamp = startupTimestamp; +<a class="jxr_linenumber" name="98" href="#98">98</a> <strong class="jxr_keyword">this</strong>.metrics = metrics; +<a class="jxr_linenumber" name="99" href="#99">99</a> maxRestartAttempts = conf +<a class="jxr_linenumber" name="100" href="#100">100</a> .getInt(Constants.DCS_MASTER_SERVER_RESTART_HANDLER_ATTEMPTS, +<a class="jxr_linenumber" name="101" href="#101">101</a> Constants.DEFAULT_DCS_MASTER_SERVER_RESTART_HANDLER_ATTEMPTS); +<a class="jxr_linenumber" name="102" href="#102">102</a> retryIntervalMillis = conf +<a class="jxr_linenumber" name="103" href="#103">103</a> .getInt(Constants.DCS_MASTER_SERVER_RESTART_HANDLER_RETRY_INTERVAL_MILLIS, +<a class="jxr_linenumber" name="104" href="#104">104</a> Constants.DEFAULT_DCS_MASTER_SERVER_RESTART_HANDLER_RETRY_INTERVAL_MILLIS); +<a class="jxr_linenumber" name="105" href="#105">105</a> trafodionQueryToolsEnabled = conf.getBoolean( +<a class="jxr_linenumber" name="106" href="#106">106</a> Constants.DCS_MASTER_TRAFODION_QUERY_TOOLS, +<a class="jxr_linenumber" name="107" href="#107">107</a> Constants.DEFAULT_DCS_MASTER_TRAFODION_QUERY_TOOLS); +<a class="jxr_linenumber" name="108" href="#108">108</a> <strong class="jxr_keyword">if</strong> (trafodionQueryToolsEnabled) +<a class="jxr_linenumber" name="109" href="#109">109</a> jdbcT4Util = <strong class="jxr_keyword">new</strong> <a href="../../../../org/trafodion/dcs/util/JdbcT4Util.html">JdbcT4Util</a>(conf, netConf); +<a class="jxr_linenumber" name="110" href="#110">110</a> retryCounterFactory = <strong class="jxr_keyword">new</strong> <a href="../../../../org/trafodion/dcs/util/RetryCounterFactory.html">RetryCounterFactory</a>(maxRestartAttempts, +<a class="jxr_linenumber" name="111" href="#111">111</a> retryIntervalMillis); +<a class="jxr_linenumber" name="112" href="#112">112</a> parentZnode = conf.get(Constants.ZOOKEEPER_ZNODE_PARENT, +<a class="jxr_linenumber" name="113" href="#113">113</a> Constants.DEFAULT_ZOOKEEPER_ZNODE_PARENT); +<a class="jxr_linenumber" name="114" href="#114">114</a> pool = Executors.newSingleThreadExecutor(); +<a class="jxr_linenumber" name="115" href="#115">115</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { +<a class="jxr_linenumber" name="116" href="#116">116</a> e.printStackTrace(); +<a class="jxr_linenumber" name="117" href="#117">117</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) +<a class="jxr_linenumber" name="118" href="#118">118</a> LOG.error(e); +<a class="jxr_linenumber" name="119" href="#119">119</a> <strong class="jxr_keyword">throw</strong> e; +<a class="jxr_linenumber" name="120" href="#120">120</a> } +<a class="jxr_linenumber" name="121" href="#121">121</a> } +<a class="jxr_linenumber" name="122" href="#122">122</a> +<a class="jxr_linenumber" name="123" href="#123">123</a> <strong class="jxr_keyword">class</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RestartHandler</a> <strong class="jxr_keyword">implements</strong> Callable<ScriptContext> { +<a class="jxr_linenumber" name="124" href="#124">124</a> <strong class="jxr_keyword">private</strong> <a href="../../../../org/trafodion/dcs/script/ScriptContext.html">ScriptContext</a> scriptContext = <strong class="jxr_keyword">new</strong> <a href="../../../../org/trafodion/dcs/script/ScriptContext.html">ScriptContext</a>(); +<a class="jxr_linenumber" name="125" href="#125">125</a> <strong class="jxr_keyword">private</strong> String znodePath; +<a class="jxr_linenumber" name="126" href="#126">126</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">int</strong> childCount; +<a class="jxr_linenumber" name="127" href="#127">127</a> +<a class="jxr_linenumber" name="128" href="#128">128</a> <strong class="jxr_keyword">public</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RestartHandler</a>(String znodePath, <strong class="jxr_keyword">int</strong> childCount) { +<a class="jxr_linenumber" name="129" href="#129">129</a> <strong class="jxr_keyword">this</strong>.znodePath = znodePath; +<a class="jxr_linenumber" name="130" href="#130">130</a> <strong class="jxr_keyword">this</strong>.childCount = childCount; +<a class="jxr_linenumber" name="131" href="#131">131</a> } +<a class="jxr_linenumber" name="132" href="#132">132</a> +<a class="jxr_linenumber" name="133" href="#133">133</a> @Override +<a class="jxr_linenumber" name="134" href="#134">134</a> <strong class="jxr_keyword">public</strong> <a href="../../../../org/trafodion/dcs/script/ScriptContext.html">ScriptContext</a> call() <strong class="jxr_keyword">throws</strong> Exception { +<a class="jxr_linenumber" name="135" href="#135">135</a> <strong class="jxr_keyword">try</strong> { +<a class="jxr_linenumber" name="136" href="#136">136</a> Scanner scn = <strong class="jxr_keyword">new</strong> Scanner(znodePath); +<a class="jxr_linenumber" name="137" href="#137">137</a> scn.useDelimiter(<span class="jxr_string">":"</span>); +<a class="jxr_linenumber" name="138" href="#138">138</a> String hostName = scn.next();<em class="jxr_comment">// host name</em> +<a class="jxr_linenumber" name="139" href="#139">139</a> String instance = scn.next();<em class="jxr_comment">// instance</em> +<a class="jxr_linenumber" name="140" href="#140">140</a> <strong class="jxr_keyword">int</strong> infoPort = Integer.parseInt(scn.next()); <em class="jxr_comment">// UI port</em> +<a class="jxr_linenumber" name="141" href="#141">141</a> <strong class="jxr_keyword">long</strong> serverStartTimestamp = Long.parseLong(scn.next()); +<a class="jxr_linenumber" name="142" href="#142">142</a> scn.close(); +<a class="jxr_linenumber" name="143" href="#143">143</a> +<a class="jxr_linenumber" name="144" href="#144">144</a> <em class="jxr_comment">// Get the --config property from classpath...it's always first</em> +<a class="jxr_linenumber" name="145" href="#145">145</a> <em class="jxr_comment">// in the classpath</em> +<a class="jxr_linenumber" name="146" href="#146">146</a> String cp = System.getProperty(<span class="jxr_string">"java.class.path"</span>); +<a class="jxr_linenumber" name="147" href="#147">147</a> scn = <strong class="jxr_keyword">new</strong> Scanner(cp); +<a class="jxr_linenumber" name="148" href="#148">148</a> scn.useDelimiter(<span class="jxr_string">":"</span>); +<a class="jxr_linenumber" name="149" href="#149">149</a> String confDir = scn.next(); +<a class="jxr_linenumber" name="150" href="#150">150</a> scn.close(); +<a class="jxr_linenumber" name="151" href="#151">151</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) +<a class="jxr_linenumber" name="152" href="#152">152</a> LOG.debug(<span class="jxr_string">"conf dir ["</span> + confDir + <span class="jxr_string">"]"</span>); +<a class="jxr_linenumber" name="153" href="#153">153</a> +<a class="jxr_linenumber" name="154" href="#154">154</a> <em class="jxr_comment">// Get -Ddcs.home.dir</em> +<a class="jxr_linenumber" name="155" href="#155">155</a> String dcsHome = System.getProperty(<span class="jxr_string">"dcs.home.dir"</span>); +<a class="jxr_linenumber" name="156" href="#156">156</a> +<a class="jxr_linenumber" name="157" href="#157">157</a> <em class="jxr_comment">// If stop-dcs.sh is executed and DCS_MANAGES_ZK then zookeeper</em> +<a class="jxr_linenumber" name="158" href="#158">158</a> <em class="jxr_comment">// is stopped abruptly.</em> +<a class="jxr_linenumber" name="159" href="#159">159</a> <em class="jxr_comment">// Second scenario is when ZooKeeper fails for some reason</em> +<a class="jxr_linenumber" name="160" href="#160">160</a> <em class="jxr_comment">// regardless of whether DCS</em> +<a class="jxr_linenumber" name="161" href="#161">161</a> <em class="jxr_comment">// manages it. When either happens the DcsServer running znodes</em> +<a class="jxr_linenumber" name="162" href="#162">162</a> <em class="jxr_comment">// still exist in ZooKeeper</em> +<a class="jxr_linenumber" name="163" href="#163">163</a> <em class="jxr_comment">// and we see them at next startup. When they eventually timeout</em> +<a class="jxr_linenumber" name="164" href="#164">164</a> <em class="jxr_comment">// we get node deleted events for a server that no longer</em> +<a class="jxr_linenumber" name="165" href="#165">165</a> <em class="jxr_comment">// exists. So, only recognize</em> +<a class="jxr_linenumber" name="166" href="#166">166</a> <em class="jxr_comment">// DcsServer running znodes that have timestamps after last</em> +<a class="jxr_linenumber" name="167" href="#167">167</a> <em class="jxr_comment">// DcsMaster startup.</em> +<a class="jxr_linenumber" name="168" href="#168">168</a> <em class="jxr_comment">//</em> +<a class="jxr_linenumber" name="169" href="#169">169</a> <em class="jxr_comment">// But, if we are DcsMaster follower that is taking over from</em> +<a class="jxr_linenumber" name="170" href="#170">170</a> <em class="jxr_comment">// failed one then ignore timestamp issues described above.</em> +<a class="jxr_linenumber" name="171" href="#171">171</a> <em class="jxr_comment">// See MasterLeaderElection.elect()</em> +<a class="jxr_linenumber" name="172" href="#172">172</a> <strong class="jxr_keyword">if</strong> ((master.isFollower() == false && serverStartTimestamp > startupTimestamp) +<a class="jxr_linenumber" name="173" href="#173">173</a> || (master.isFollower() && runningServers.size() < configuredServers.size())) { +<a class="jxr_linenumber" name="174" href="#174">174</a> scriptContext.setHostName(hostName); +<a class="jxr_linenumber" name="175" href="#175">175</a> scriptContext +<a class="jxr_linenumber" name="176" href="#176">176</a> .setScriptName(Constants.SYS_SHELL_SCRIPT_NAME); +<a class="jxr_linenumber" name="177" href="#177">177</a> +<a class="jxr_linenumber" name="178" href="#178">178</a> <strong class="jxr_keyword">if</strong> (hostName.equalsIgnoreCase(netConf.getHostName())) +<a class="jxr_linenumber" name="179" href="#179">179</a> scriptContext.setCommand(<span class="jxr_string">"bin/dcs-daemon.sh --config "</span> +<a class="jxr_linenumber" name="180" href="#180">180</a> + confDir + <span class="jxr_string">" start server "</span> + instance + <span class="jxr_string">" "</span> +<a class="jxr_linenumber" name="181" href="#181">181</a> + childCount); +<a class="jxr_linenumber" name="182" href="#182">182</a> <strong class="jxr_keyword">else</strong> +<a class="jxr_linenumber" name="183" href="#183">183</a> scriptContext.setCommand(<span class="jxr_string">"pdsh -w "</span> + hostName +<a class="jxr_linenumber" name="184" href="#184">184</a> + <span class="jxr_string">" \"cd "</span> + dcsHome +<a class="jxr_linenumber" name="185" href="#185">185</a> + <span class="jxr_string">";bin/dcs-daemon.sh --config "</span> + confDir +<a class="jxr_linenumber" name="186" href="#186">186</a> + <span class="jxr_string">" start server "</span> + instance + <span class="jxr_string">" "</span> +<a class="jxr_linenumber" name="187" href="#187">187</a> + childCount + <span class="jxr_string">"\""</span>); +<a class="jxr_linenumber" name="188" href="#188">188</a> +<a class="jxr_linenumber" name="189" href="#189">189</a> <a href="../../../../org/trafodion/dcs/util/RetryCounter.html">RetryCounter</a> retryCounter = retryCounterFactory.create(); +<a class="jxr_linenumber" name="190" href="#190">190</a> <strong class="jxr_keyword">while</strong> (<strong class="jxr_keyword">true</strong>) { +<a class="jxr_linenumber" name="191" href="#191">191</a> <strong class="jxr_keyword">if</strong> (scriptContext.getStdOut().length() > 0) +<a class="jxr_linenumber" name="192" href="#192">192</a> scriptContext.getStdOut().delete(0, +<a class="jxr_linenumber" name="193" href="#193">193</a> scriptContext.getStdOut().length()); +<a class="jxr_linenumber" name="194" href="#194">194</a> <strong class="jxr_keyword">if</strong> (scriptContext.getStdErr().length() > 0) +<a class="jxr_linenumber" name="195" href="#195">195</a> scriptContext.getStdErr().delete(0, +<a class="jxr_linenumber" name="196" href="#196">196</a> scriptContext.getStdErr().length()); +<a class="jxr_linenumber" name="197" href="#197">197</a> LOG.info(<span class="jxr_string">"Restarting DcsServer ["</span> + hostName + <span class="jxr_string">":"</span> +<a class="jxr_linenumber" name="198" href="#198">198</a> + instance + <span class="jxr_string">"], script [ "</span> +<a class="jxr_linenumber" name="199" href="#199">199</a> + scriptContext.toString() + <span class="jxr_string">" ]"</span>); +<a class="jxr_linenumber" name="200" href="#200">200</a> ScriptManager.getInstance().runScript(scriptContext); +<a class="jxr_linenumber" name="201" href="#201">201</a> +<a class="jxr_linenumber" name="202" href="#202">202</a> <strong class="jxr_keyword">if</strong> (scriptContext.getExitCode() == 0) { +<a class="jxr_linenumber" name="203" href="#203">203</a> LOG.info(<span class="jxr_string">"DcsServer ["</span> + hostName + <span class="jxr_string">":"</span> + instance +<a class="jxr_linenumber" name="204" href="#204">204</a> + <span class="jxr_string">"] restarted"</span>); +<a class="jxr_linenumber" name="205" href="#205">205</a> <strong class="jxr_keyword">break</strong>; +<a class="jxr_linenumber" name="206" href="#206">206</a> } <strong class="jxr_keyword">else</strong> { +<a class="jxr_linenumber" name="207" href="#207">207</a> StringBuilder sb = <strong class="jxr_keyword">new</strong> StringBuilder(); +<a class="jxr_linenumber" name="208" href="#208">208</a> sb.append(<span class="jxr_string">"exit code ["</span> +<a class="jxr_linenumber" name="209" href="#209">209</a> + scriptContext.getExitCode() + <span class="jxr_string">"]"</span>); +<a class="jxr_linenumber" name="210" href="#210">210</a> <strong class="jxr_keyword">if</strong> (!scriptContext.getStdOut().toString().isEmpty()) +<a class="jxr_linenumber" name="211" href="#211">211</a> sb.append(<span class="jxr_string">", stdout ["</span> +<a class="jxr_linenumber" name="212" href="#212">212</a> + scriptContext.getStdOut().toString() +<a class="jxr_linenumber" name="213" href="#213">213</a> + <span class="jxr_string">"]"</span>); +<a class="jxr_linenumber" name="214" href="#214">214</a> <strong class="jxr_keyword">if</strong> (!scriptContext.getStdErr().toString().isEmpty()) +<a class="jxr_linenumber" name="215" href="#215">215</a> sb.append(<span class="jxr_string">", stderr ["</span> +<a class="jxr_linenumber" name="216" href="#216">216</a> + scriptContext.getStdErr().toString() +<a class="jxr_linenumber" name="217" href="#217">217</a> + <span class="jxr_string">"]"</span>); +<a class="jxr_linenumber" name="218" href="#218">218</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) +<a class="jxr_linenumber" name="219" href="#219">219</a> LOG.error(sb.toString()); +<a class="jxr_linenumber" name="220" href="#220">220</a> +<a class="jxr_linenumber" name="221" href="#221">221</a> <strong class="jxr_keyword">if</strong> (!retryCounter.shouldRetry()) { +<a class="jxr_linenumber" name="222" href="#222">222</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) +<a class="jxr_linenumber" name="223" href="#223">223</a> LOG.error(<span class="jxr_string">"DcsServer ["</span> + hostName + <span class="jxr_string">":"</span> +<a class="jxr_linenumber" name="224" href="#224">224</a> + instance +<a class="jxr_linenumber" name="225" href="#225">225</a> + <span class="jxr_string">"] restart failed after "</span> +<a class="jxr_linenumber" name="226" href="#226">226</a> + retryCounter.getMaxRetries() +<a class="jxr_linenumber" name="227" href="#227">227</a> + <span class="jxr_string">" retries"</span>); +<a class="jxr_linenumber" name="228" href="#228">228</a> <strong class="jxr_keyword">break</strong>; +<a class="jxr_linenumber" name="229" href="#229">229</a> } <strong class="jxr_keyword">else</strong> { +<a class="jxr_linenumber" name="230" href="#230">230</a> retryCounter.sleepUntilNextRetry(); +<a class="jxr_linenumber" name="231" href="#231">231</a> retryCounter.useRetry(); +<a class="jxr_linenumber" name="232" href="#232">232</a> } +<a class="jxr_linenumber" name="233" href="#233">233</a> } +<a class="jxr_linenumber" name="234" href="#234">234</a> } +<a class="jxr_linenumber" name="235" href="#235">235</a> } <strong class="jxr_keyword">else</strong> { +<a class="jxr_linenumber" name="236" href="#236">236</a> StringBuffer sb = <strong class="jxr_keyword">new</strong> StringBuffer(); +<a class="jxr_linenumber" name="237" href="#237">237</a> sb.append(<span class="jxr_string">"No restart for "</span>).append(znodePath).append(System.getProperty(<span class="jxr_string">"line.separator"</span>)); +<a class="jxr_linenumber" name="238" href="#238">238</a> sb.append(<span class="jxr_string">"DCS Master isFollower ["</span>).append(master.isFollower()).append(<span class="jxr_string">"], "</span>); +<a class="jxr_linenumber" name="239" href="#239">239</a> sb.append(<span class="jxr_string">"DCS Master start time ["</span>) +<a class="jxr_linenumber" name="240" href="#240">240</a> .append(DateFormat.getDateTimeInstance().format(<strong class="jxr_keyword">new</strong> Date(startupTimestamp))).append(<span class="jxr_string">"], "</span>); +<a class="jxr_linenumber" name="241" href="#241">241</a> sb.append(<span class="jxr_string">"DCS Server start time ["</span>) +<a class="jxr_linenumber" name="242" href="#242">242</a> .append(DateFormat.getDateTimeInstance().format(<strong class="jxr_keyword">new</strong> Date(serverStartTimestamp))).append(<span class="jxr_string">"], "</span>); +<a class="jxr_linenumber" name="243" href="#243">243</a> sb.append(<span class="jxr_string">"running DCS Server num is ["</span>).append(runningServers.size()) +<a class="jxr_linenumber" name="244" href="#244">244</a> .append(<span class="jxr_string">"], registered DCS Server num is ["</span>).append(registeredServers.size()).append(<span class="jxr_string">"]."</span>); +<a class="jxr_linenumber" name="245" href="#245">245</a> +<a class="jxr_linenumber" name="246" href="#246">246</a> LOG.info(sb.toString()); +<a class="jxr_linenumber" name="247" href="#247">247</a> } +<a class="jxr_linenumber" name="248" href="#248">248</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { +<a class="jxr_linenumber" name="249" href="#249">249</a> e.printStackTrace(); +<a class="jxr_linenumber" name="250" href="#250">250</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) +<a class="jxr_linenumber" name="251" href="#251">251</a> LOG.error(e); +<a class="jxr_linenumber" name="252" href="#252">252</a> } +<a class="jxr_linenumber" name="253" href="#253">253</a> +<a class="jxr_linenumber" name="254" href="#254">254</a> <strong class="jxr_keyword">return</strong> scriptContext; +<a class="jxr_linenumber" name="255" href="#255">255</a> } +<a class="jxr_linenumber" name="256" href="#256">256</a> } +<a class="jxr_linenumber" name="257" href="#257">257</a> +<a class="jxr_linenumber" name="258" href="#258">258</a> <strong class="jxr_keyword">class</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RunningWatcher</a> <strong class="jxr_keyword">implements</strong> Watcher { +<a class="jxr_linenumber" name="259" href="#259">259</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> process(WatchedEvent event) { +<a class="jxr_linenumber" name="260" href="#260">260</a> <strong class="jxr_keyword">if</strong> (event.getType() == Event.EventType.NodeChildrenChanged) { +<a class="jxr_linenumber" name="261" href="#261">261</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) +<a class="jxr_linenumber" name="262" href="#262">262</a> LOG.debug(<span class="jxr_string">"Running children changed ["</span> + event.getPath() +<a class="jxr_linenumber" name="263" href="#263">263</a> + <span class="jxr_string">"]"</span>); +<a class="jxr_linenumber" name="264" href="#264">264</a> <strong class="jxr_keyword">try</strong> { +<a class="jxr_linenumber" name="265" href="#265">265</a> getZkRunning(); +<a class="jxr_linenumber" name="266" href="#266">266</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { +<a class="jxr_linenumber" name="267" href="#267">267</a> e.printStackTrace(); +<a class="jxr_linenumber" name="268" href="#268">268</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) +<a class="jxr_linenumber" name="269" href="#269">269</a> LOG.error(e); +<a class="jxr_linenumber" name="270" href="#270">270</a> } +<a class="jxr_linenumber" name="271" href="#271">271</a> } <strong class="jxr_keyword">else</strong> <strong class="jxr_keyword">if</strong> (event.getType() == Event.EventType.NodeDeleted) { +<a class="jxr_linenumber" name="272" href="#272">272</a> String znodePath = event.getPath(); +<a class="jxr_linenumber" name="273" href="#273">273</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) +<a class="jxr_linenumber" name="274" href="#274">274</a> LOG.debug(<span class="jxr_string">"Running znode deleted ["</span> + znodePath + <span class="jxr_string">"]"</span>); +<a class="jxr_linenumber" name="275" href="#275">275</a> <strong class="jxr_keyword">try</strong> { +<a class="jxr_linenumber" name="276" href="#276">276</a> restartServer(znodePath); +<a class="jxr_linenumber" name="277" href="#277">277</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { +<a class="jxr_linenumber" name="278" href="#278">278</a> e.printStackTrace(); +<a class="jxr_linenumber" name="279" href="#279">279</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) +<a class="jxr_linenumber" name="280" href="#280">280</a> LOG.error(e); +<a class="jxr_linenumber" name="281" href="#281">281</a> } +<a class="jxr_linenumber" name="282" href="#282">282</a> } +<a class="jxr_linenumber" name="283" href="#283">283</a> } +<a class="jxr_linenumber" name="284" href="#284">284</a> } +<a class="jxr_linenumber" name="285" href="#285">285</a> +<a class="jxr_linenumber" name="286" href="#286">286</a> <strong class="jxr_keyword">class</strong> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RegisteredWatcher</a> <strong class="jxr_keyword">implements</strong> Watcher { +<a class="jxr_linenumber" name="287" href="#287">287</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> process(WatchedEvent event) { +<a class="jxr_linenumber" name="288" href="#288">288</a> <strong class="jxr_keyword">if</strong> (event.getType() == Event.EventType.NodeChildrenChanged) { +<a class="jxr_linenumber" name="289" href="#289">289</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) +<a class="jxr_linenumber" name="290" href="#290">290</a> LOG.debug(<span class="jxr_string">"Registered children changed ["</span> + event.getPath() +<a class="jxr_linenumber" name="291" href="#291">291</a> + <span class="jxr_string">"]"</span>); +<a class="jxr_linenumber" name="292" href="#292">292</a> <strong class="jxr_keyword">try</strong> { +<a class="jxr_linenumber" name="293" href="#293">293</a> getZkRegistered(); +<a class="jxr_linenumber" name="294" href="#294">294</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { +<a class="jxr_linenumber" name="295" href="#295">295</a> e.printStackTrace(); +<a class="jxr_linenumber" name="296" href="#296">296</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) +<a class="jxr_linenumber" name="297" href="#297">297</a> LOG.error(e); +<a class="jxr_linenumber" name="298" href="#298">298</a> } +<a class="jxr_linenumber" name="299" href="#299">299</a> } +<a class="jxr_linenumber" name="300" href="#300">300</a> } +<a class="jxr_linenumber" name="301" href="#301">301</a> } +<a class="jxr_linenumber" name="302" href="#302">302</a> +<a class="jxr_linenumber" name="303" href="#303">303</a> @Override +<a class="jxr_linenumber" name="304" href="#304">304</a> <strong class="jxr_keyword">public</strong> Boolean call() <strong class="jxr_keyword">throws</strong> Exception { +<a class="jxr_linenumber" name="305" href="#305">305</a> +<a class="jxr_linenumber" name="306" href="#306">306</a> <strong class="jxr_keyword">long</strong> timeoutMillis = 5000; +<a class="jxr_linenumber" name="307" href="#307">307</a> +<a class="jxr_linenumber" name="308" href="#308">308</a> <strong class="jxr_keyword">try</strong> { +<a class="jxr_linenumber" name="309" href="#309">309</a> getServersFile(); +<a class="jxr_linenumber" name="310" href="#310">310</a> createServersPortMap(); +<a class="jxr_linenumber" name="311" href="#311">311</a> getZkRunning(); +<a class="jxr_linenumber" name="312" href="#312">312</a> getUnwathedServers(); +<a class="jxr_linenumber" name="313" href="#313">313</a> getZkRegistered(); <a class="jxr_linenumber" name="314" href="#314">314</a> -<a class="jxr_linenumber" name="315" href="#315">315</a> <strong class="jxr_keyword">try</strong> { -<a class="jxr_linenumber" name="316" href="#316">316</a> getServersFile(); -<a class="jxr_linenumber" name="317" href="#317">317</a> createServersPortMap(); -<a class="jxr_linenumber" name="318" href="#318">318</a> getZkRunning(); -<a class="jxr_linenumber" name="319" href="#319">319</a> getZkRegistered(); -<a class="jxr_linenumber" name="320" href="#320">320</a> -<a class="jxr_linenumber" name="321" href="#321">321</a> <strong class="jxr_keyword">while</strong> (<strong class="jxr_keyword">true</strong>) { -<a class="jxr_linenumber" name="322" href="#322">322</a> <strong class="jxr_keyword">while</strong> (!restartQueue.isEmpty()) { -<a class="jxr_linenumber" name="323" href="#323">323</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) -<a class="jxr_linenumber" name="324" href="#324">324</a> LOG.debug(<span class="jxr_string">"Restart queue size ["</span> + restartQueue.size() -<a class="jxr_linenumber" name="325" href="#325">325</a> + <span class="jxr_string">"]"</span>); -<a class="jxr_linenumber" name="326" href="#326">326</a> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RestartHandler</a> handler = restartQueue.poll(); -<a class="jxr_linenumber" name="327" href="#327">327</a> Future<ScriptContext> runner = pool.submit(handler); -<a class="jxr_linenumber" name="328" href="#328">328</a> <a href="../../../../org/trafodion/dcs/script/ScriptContext.html">ScriptContext</a> scriptContext = runner.get();<em class="jxr_comment">// blocking call</em> -<a class="jxr_linenumber" name="329" href="#329">329</a> <strong class="jxr_keyword">if</strong> (scriptContext.getExitCode() != 0) -<a class="jxr_linenumber" name="330" href="#330">330</a> restartQueue.add(handler); -<a class="jxr_linenumber" name="331" href="#331">331</a> } +<a class="jxr_linenumber" name="315" href="#315">315</a> <strong class="jxr_keyword">while</strong> (<strong class="jxr_keyword">true</strong>) { +<a class="jxr_linenumber" name="316" href="#316">316</a> <strong class="jxr_keyword">while</strong> (!restartQueue.isEmpty()) { +<a class="jxr_linenumber" name="317" href="#317">317</a> <strong class="jxr_keyword">if</strong> (LOG.isDebugEnabled()) +<a class="jxr_linenumber" name="318" href="#318">318</a> LOG.debug(<span class="jxr_string">"Restart queue size ["</span> + restartQueue.size() +<a class="jxr_linenumber" name="319" href="#319">319</a> + <span class="jxr_string">"]"</span>); +<a class="jxr_linenumber" name="320" href="#320">320</a> <a href="../../../../org/trafodion/dcs/master/ServerManager.html">RestartHandler</a> handler = restartQueue.poll(); +<a class="jxr_linenumber" name="321" href="#321">321</a> Future<ScriptContext> runner = pool.submit(handler); +<a class="jxr_linenumber" name="322" href="#322">322</a> <a href="../../../../org/trafodion/dcs/script/ScriptContext.html">ScriptContext</a> scriptContext = runner.get();<em class="jxr_comment">// blocking call</em> +<a class="jxr_linenumber" name="323" href="#323">323</a> <strong class="jxr_keyword">if</strong> (scriptContext.getExitCode() != 0) +<a class="jxr_linenumber" name="324" href="#324">324</a> restartQueue.add(handler); +<a class="jxr_linenumber" name="325" href="#325">325</a> } +<a class="jxr_linenumber" name="326" href="#326">326</a> +<a class="jxr_linenumber" name="327" href="#327">327</a> <strong class="jxr_keyword">try</strong> { +<a class="jxr_linenumber" name="328" href="#328">328</a> Thread.sleep(timeoutMillis); +<a class="jxr_linenumber" name="329" href="#329">329</a> } <strong class="jxr_keyword">catch</strong> (InterruptedException e) { +<a class="jxr_linenumber" name="330" href="#330">330</a> } +<a class="jxr_linenumber" name="331" href="#331">331</a> } <a class="jxr_linenumber" name="332" href="#332">332</a> -<a class="jxr_linenumber" name="333" href="#333">333</a> <strong class="jxr_keyword">try</strong> { -<a class="jxr_linenumber" name="334" href="#334">334</a> Thread.sleep(timeoutMillis); -<a class="jxr_linenumber" name="335" href="#335">335</a> } <strong class="jxr_keyword">catch</strong> (InterruptedException e) { -<a class="jxr_linenumber" name="336" href="#336">336</a> } -<a class="jxr_linenumber" name="337" href="#337">337</a> } -<a class="jxr_linenumber" name="338" href="#338">338</a> -<a class="jxr_linenumber" name="339" href="#339">339</a> } <strong class="jxr_keyword">catch</strong> (Exception e) { -<a class="jxr_linenumber" name="340" href="#340">340</a> e.printStackTrace(); -<a class="jxr_linenumber" name="341" href="#341">341</a> <strong class="jxr_keyword">if</strong> (LOG.isErrorEnabled()) -<a class="jxr_linenumber" name="342" href="#342">342</a> LOG.error(e); -<a class="jxr_linenumber" name="343" href="#343">343</a> pool.shutdown(); -<a class="jxr_linenumber" name="344" href="#344">344</a> <strong class="jxr_keyword">throw</strong> e; -<a class="jxr_linenumber" name="345" href="#345">345</a> } -<a class="jxr_linenumber" name="346" href="#346">346</a> } -<a class="jxr_linenumber" name="347" href="#347">347</a> -<a class="jxr_linenumber" name="348" href="#348">348</a> <strong class="jxr_keyword">private</strong> List<String> getChildren(String znode, Watcher watcher) -<a class="jxr_linenumber" name="349" href="#349">349</a> <strong class="jxr_keyword">throws</strong> Exception { -<a class="jxr_linenumber" name="350" href="#350">350</a> List<String> children = <strong class="jxr_keyword">null</strong>; -<a class="jxr_linenumber" name="351" href="#351">351</a> children = zkc.getChildren(znode, watcher); -<a class="jxr_linenumber" name="352" href="#352">352</a> <strong class="jxr_keyword">if</strong> (!children.isEmpty()) -<a class="jxr_linenumber" name="353" href="#353">353</a> Collections.sort(children); -<a class="jxr_linenumber" name="354" href="#354">354</a> <strong class="jxr_keyword">return</strong> children; -<a class="jxr_linenumber" name="355" href="#355">355</a> } -<a class="jxr_linenumber" name="356" href="#356">356</a> -<a class="jxr_linenumber" name="357" href="#357">357</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">void</strong> getServersFile() <strong class="jxr_keyword">throws</strong> Exception { -<a class="jxr_linenumber" name="358" href="#358">358</a> InputStream is = <strong class="jxr_keyword">this</strong>.getClass().getResourceAsStream(<span class="jxr_string">"/servers"</span>); -<a class="jxr_linenumber" name="359" href="#359">359</a> <strong class="jxr_keyword">if</strong> (is == <strong class="jxr_keyword">null</strong>) -<a class="jxr_linenumber" name="360" href="#360">360</a> <strong class="jxr_keyword">throw</strong> <strong class="jxr_keyword">new</strong> IOException(<span class="jxr_string">"Cannot find servers file"</span>); -<a class="jxr_linenumber" name="361" href="#361">361</a> -<a class="jxr_linenumber" name="362" href="#362">362</a> BufferedReader br = <strong class="jxr_keyword">new</strong> BufferedReader(<strong class="jxr_keyword">new</strong> InputStreamReader(is)); -<a class="jxr_linenumber" name="363" href="#363">363</a> configuredServers.clear(); -<a class="jxr_linenumber" name="364" href="#364">364</a> String line; -<a class="jxr_linenumber" name="365" href="#365">365</a> <strong class="jxr_keyword">int</strong> lineNum = 1; -<a class="jxr_linenumber" name="366" href="#366">366</a> <strong class="jxr_keyword">while</strong> ((line = br.readLine()) != <strong class="jxr_keyword">null</strong>) { -<a class="jxr_linenumber" name="367" href="#367">367</a> Scanner scn = <strong class="jxr_keyword">new</strong> Scanner(line); -<a class="jxr_linenumber" name="368" href="#368">368</a> scn.useDelimiter(<span class="jxr_string">" "</span>); -<a class="jxr_linenumber" name="369" href="#369">369</a> String hostName = <strong class="jxr_keyword">null</strong>; -<a class="jxr_linenumber" name="370" href="#370">370</a>
<TRUNCATED>
