http://git-wip-us.apache.org/repos/asf/hbase-site/blob/58f51408/xref/org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html ---------------------------------------------------------------------- diff --git a/xref/org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html b/xref/org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html index 4ef24c8..9c50f37 100644 --- a/xref/org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html +++ b/xref/org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html @@ -27,392 +27,345 @@ <a class="jxr_linenumber" name="17" href="#17">17</a> <em class="jxr_javadoccomment"> */</em> <a class="jxr_linenumber" name="18" href="#18">18</a> <strong class="jxr_keyword">package</strong> org.apache.hadoop.hbase.client; <a class="jxr_linenumber" name="19" href="#19">19</a> -<a class="jxr_linenumber" name="20" href="#20">20</a> <strong class="jxr_keyword">import</strong> java.io.EOFException; -<a class="jxr_linenumber" name="21" href="#21">21</a> <strong class="jxr_keyword">import</strong> java.io.IOException; -<a class="jxr_linenumber" name="22" href="#22">22</a> <strong class="jxr_keyword">import</strong> java.io.SyncFailedException; -<a class="jxr_linenumber" name="23" href="#23">23</a> <strong class="jxr_keyword">import</strong> java.lang.reflect.UndeclaredThrowableException; -<a class="jxr_linenumber" name="24" href="#24">24</a> <strong class="jxr_keyword">import</strong> java.net.ConnectException; -<a class="jxr_linenumber" name="25" href="#25">25</a> <strong class="jxr_keyword">import</strong> java.net.SocketTimeoutException; -<a class="jxr_linenumber" name="26" href="#26">26</a> <strong class="jxr_keyword">import</strong> java.nio.channels.ClosedChannelException; -<a class="jxr_linenumber" name="27" href="#27">27</a> <strong class="jxr_keyword">import</strong> java.util.Map.Entry; -<a class="jxr_linenumber" name="28" href="#28">28</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.ConcurrentHashMap; -<a class="jxr_linenumber" name="29" href="#29">29</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.ConcurrentMap; -<a class="jxr_linenumber" name="30" href="#30">30</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.TimeoutException; -<a class="jxr_linenumber" name="31" href="#31">31</a> -<a class="jxr_linenumber" name="32" href="#32">32</a> <strong class="jxr_keyword">import</strong> org.apache.commons.lang.mutable.MutableBoolean; -<a class="jxr_linenumber" name="33" href="#33">33</a> <strong class="jxr_keyword">import</strong> org.apache.commons.logging.Log; -<a class="jxr_linenumber" name="34" href="#34">34</a> <strong class="jxr_keyword">import</strong> org.apache.commons.logging.LogFactory; -<a class="jxr_linenumber" name="35" href="#35">35</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.conf.Configuration; -<a class="jxr_linenumber" name="36" href="#36">36</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.DoNotRetryIOException; -<a class="jxr_linenumber" name="37" href="#37">37</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.HConstants; -<a class="jxr_linenumber" name="38" href="#38">38</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.ServerName; -<a class="jxr_linenumber" name="39" href="#39">39</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.classification.InterfaceAudience; -<a class="jxr_linenumber" name="40" href="#40">40</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.exceptions.ConnectionClosingException; -<a class="jxr_linenumber" name="41" href="#41">41</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.exceptions.PreemptiveFastFailException; -<a class="jxr_linenumber" name="42" href="#42">42</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.ipc.FailedServerException; -<a class="jxr_linenumber" name="43" href="#43">43</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.util.EnvironmentEdgeManager; -<a class="jxr_linenumber" name="44" href="#44">44</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.ipc.RemoteException; -<a class="jxr_linenumber" name="45" href="#45">45</a> -<a class="jxr_linenumber" name="46" href="#46">46</a> <em class="jxr_javadoccomment">/**</em> -<a class="jxr_linenumber" name="47" href="#47">47</a> <em class="jxr_javadoccomment"> * </em> -<a class="jxr_linenumber" name="48" href="#48">48</a> <em class="jxr_javadoccomment"> * The concrete {@link RetryingCallerInterceptor} class that implements the preemptive fast fail</em> -<a class="jxr_linenumber" name="49" href="#49">49</a> <em class="jxr_javadoccomment"> * feature.</em> -<a class="jxr_linenumber" name="50" href="#50">50</a> <em class="jxr_javadoccomment"> * </em> -<a class="jxr_linenumber" name="51" href="#51">51</a> <em class="jxr_javadoccomment"> * The motivation is as follows : </em> -<a class="jxr_linenumber" name="52" href="#52">52</a> <em class="jxr_javadoccomment"> * In case where a large number of clients try and talk to a particular region server in hbase, if</em> -<a class="jxr_linenumber" name="53" href="#53">53</a> <em class="jxr_javadoccomment"> * the region server goes down due to network problems, we might end up in a scenario where</em> -<a class="jxr_linenumber" name="54" href="#54">54</a> <em class="jxr_javadoccomment"> * the clients would go into a state where they all start to retry.</em> -<a class="jxr_linenumber" name="55" href="#55">55</a> <em class="jxr_javadoccomment"> * This behavior will set off many of the threads in pretty much the same path and they all would be</em> -<a class="jxr_linenumber" name="56" href="#56">56</a> <em class="jxr_javadoccomment"> * sleeping giving rise to a state where the client either needs to create more threads to send new</em> -<a class="jxr_linenumber" name="57" href="#57">57</a> <em class="jxr_javadoccomment"> * requests to other hbase machines or block because the client cannot create anymore threads.</em> -<a class="jxr_linenumber" name="58" href="#58">58</a> <em class="jxr_javadoccomment"> * </em> -<a class="jxr_linenumber" name="59" href="#59">59</a> <em class="jxr_javadoccomment"> * In most cases the clients might prefer to have a bound on the number of threads that are created</em> -<a class="jxr_linenumber" name="60" href="#60">60</a> <em class="jxr_javadoccomment"> * in order to send requests to hbase. This would mostly result in the client thread starvation.</em> -<a class="jxr_linenumber" name="61" href="#61">61</a> <em class="jxr_javadoccomment"> * </em> -<a class="jxr_linenumber" name="62" href="#62">62</a> <em class="jxr_javadoccomment"> * To circumvent this problem, the approach that is being taken here under is to let 1 of the many</em> -<a class="jxr_linenumber" name="63" href="#63">63</a> <em class="jxr_javadoccomment"> * threads who are trying to contact the regionserver with connection problems and let the other</em> -<a class="jxr_linenumber" name="64" href="#64">64</a> <em class="jxr_javadoccomment"> * threads get a {@link PreemptiveFastFailException} so that they can move on and take other</em> -<a class="jxr_linenumber" name="65" href="#65">65</a> <em class="jxr_javadoccomment"> * requests.</em> -<a class="jxr_linenumber" name="66" href="#66">66</a> <em class="jxr_javadoccomment"> * </em> -<a class="jxr_linenumber" name="67" href="#67">67</a> <em class="jxr_javadoccomment"> * This would give the client more flexibility on the kind of action he would want to take in cases</em> -<a class="jxr_linenumber" name="68" href="#68">68</a> <em class="jxr_javadoccomment"> * where the regionserver is down. He can either discard the requests and send a nack upstream</em> -<a class="jxr_linenumber" name="69" href="#69">69</a> <em class="jxr_javadoccomment"> * faster or have an application level retry or buffer the requests up so as to send them down to</em> -<a class="jxr_linenumber" name="70" href="#70">70</a> <em class="jxr_javadoccomment"> * hbase later.</em> -<a class="jxr_linenumber" name="71" href="#71">71</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="72" href="#72">72</a> <em class="jxr_javadoccomment"> */</em> -<a class="jxr_linenumber" name="73" href="#73">73</a> @InterfaceAudience.Private -<a class="jxr_linenumber" name="74" href="#74">74</a> <strong class="jxr_keyword">class</strong> <a href="../../../../../org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html">PreemptiveFastFailInterceptor</a> <strong class="jxr_keyword">extends</strong> <a href="../../../../../org/apache/hadoop/hbase/client/RetryingCallerInterceptor.html">RetryingCallerInterceptor</a> { -<a class="jxr_linenumber" name="75" href="#75">75</a> -<a class="jxr_linenumber" name="76" href="#76">76</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">static</strong> <strong class="jxr_keyword">final</strong> Log LOG = LogFactory -<a class="jxr_linenumber" name="77" href="#77">77</a> .getLog(PreemptiveFastFailInterceptor.<strong class="jxr_keyword">class</strong>); -<a class="jxr_linenumber" name="78" href="#78">78</a> -<a class="jxr_linenumber" name="79" href="#79">79</a> <em class="jxr_comment">// amount of time to wait before we consider a server to be in fast fail</em> -<a class="jxr_linenumber" name="80" href="#80">80</a> <em class="jxr_comment">// mode</em> -<a class="jxr_linenumber" name="81" href="#81">81</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">final</strong> <strong class="jxr_keyword">long</strong> fastFailThresholdMilliSec; -<a class="jxr_linenumber" name="82" href="#82">82</a> -<a class="jxr_linenumber" name="83" href="#83">83</a> <em class="jxr_comment">// Keeps track of failures when we cannot talk to a server. Helps in</em> -<a class="jxr_linenumber" name="84" href="#84">84</a> <em class="jxr_comment">// fast failing clients if the server is down for a long time.</em> -<a class="jxr_linenumber" name="85" href="#85">85</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">final</strong> ConcurrentMap<ServerName, FailureInfo> repeatedFailuresMap = -<a class="jxr_linenumber" name="86" href="#86">86</a> <strong class="jxr_keyword">new</strong> ConcurrentHashMap<ServerName, FailureInfo>(); -<a class="jxr_linenumber" name="87" href="#87">87</a> -<a class="jxr_linenumber" name="88" href="#88">88</a> <em class="jxr_comment">// We populate repeatedFailuresMap every time there is a failure. So, to</em> -<a class="jxr_linenumber" name="89" href="#89">89</a> <em class="jxr_comment">// keep it from growing unbounded, we garbage collect the failure information</em> -<a class="jxr_linenumber" name="90" href="#90">90</a> <em class="jxr_comment">// every cleanupInterval.</em> -<a class="jxr_linenumber" name="91" href="#91">91</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">final</strong> <strong class="jxr_keyword">long</strong> failureMapCleanupIntervalMilliSec; -<a class="jxr_linenumber" name="92" href="#92">92</a> -<a class="jxr_linenumber" name="93" href="#93">93</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">volatile</strong> <strong class="jxr_keyword">long</strong> lastFailureMapCleanupTimeMilliSec; -<a class="jxr_linenumber" name="94" href="#94">94</a> -<a class="jxr_linenumber" name="95" href="#95">95</a> <em class="jxr_comment">// clear failure Info. Used to clean out all entries.</em> -<a class="jxr_linenumber" name="96" href="#96">96</a> <em class="jxr_comment">// A safety valve, in case the client does not exit the</em> -<a class="jxr_linenumber" name="97" href="#97">97</a> <em class="jxr_comment">// fast fail mode for any reason.</em> -<a class="jxr_linenumber" name="98" href="#98">98</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">long</strong> fastFailClearingTimeMilliSec; -<a class="jxr_linenumber" name="99" href="#99">99</a> -<a class="jxr_linenumber" name="100" href="#100">100</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ThreadLocal<MutableBoolean> threadRetryingInFastFailMode = -<a class="jxr_linenumber" name="101" href="#101">101</a> <strong class="jxr_keyword">new</strong> ThreadLocal<MutableBoolean>(); -<a class="jxr_linenumber" name="102" href="#102">102</a> -<a class="jxr_linenumber" name="103" href="#103">103</a> <strong class="jxr_keyword">public</strong> <a href="../../../../../org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html">PreemptiveFastFailInterceptor</a>(Configuration conf) { -<a class="jxr_linenumber" name="104" href="#104">104</a> <strong class="jxr_keyword">this</strong>.fastFailThresholdMilliSec = conf.getLong( -<a class="jxr_linenumber" name="105" href="#105">105</a> HConstants.HBASE_CLIENT_FAST_FAIL_THREASHOLD_MS, -<a class="jxr_linenumber" name="106" href="#106">106</a> HConstants.HBASE_CLIENT_FAST_FAIL_THREASHOLD_MS_DEFAULT); -<a class="jxr_linenumber" name="107" href="#107">107</a> <strong class="jxr_keyword">this</strong>.failureMapCleanupIntervalMilliSec = conf.getLong( -<a class="jxr_linenumber" name="108" href="#108">108</a> HConstants.HBASE_CLIENT_FAST_FAIL_CLEANUP_MS_DURATION_MS, -<a class="jxr_linenumber" name="109" href="#109">109</a> HConstants.HBASE_CLIENT_FAST_FAIL_CLEANUP_DURATION_MS_DEFAULT); -<a class="jxr_linenumber" name="110" href="#110">110</a> lastFailureMapCleanupTimeMilliSec = EnvironmentEdgeManager.currentTime(); -<a class="jxr_linenumber" name="111" href="#111">111</a> } -<a class="jxr_linenumber" name="112" href="#112">112</a> -<a class="jxr_linenumber" name="113" href="#113">113</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> intercept(<a href="../../../../../org/apache/hadoop/hbase/client/FastFailInterceptorContext.html">FastFailInterceptorContext</a> context) -<a class="jxr_linenumber" name="114" href="#114">114</a> <strong class="jxr_keyword">throws</strong> <a href="../../../../../org/apache/hadoop/hbase/exceptions/PreemptiveFastFailException.html">PreemptiveFastFailException</a> { -<a class="jxr_linenumber" name="115" href="#115">115</a> context.setFailureInfo(repeatedFailuresMap.get(context.getServer())); -<a class="jxr_linenumber" name="116" href="#116">116</a> <strong class="jxr_keyword">if</strong> (inFastFailMode(context.getServer()) && !currentThreadInFastFailMode()) { -<a class="jxr_linenumber" name="117" href="#117">117</a> <em class="jxr_comment">// In Fast-fail mode, all but one thread will fast fail. Check</em> -<a class="jxr_linenumber" name="118" href="#118">118</a> <em class="jxr_comment">// if we are that one chosen thread.</em> -<a class="jxr_linenumber" name="119" href="#119">119</a> context.setRetryDespiteFastFailMode(shouldRetryInspiteOfFastFail(context -<a class="jxr_linenumber" name="120" href="#120">120</a> .getFailureInfo())); -<a class="jxr_linenumber" name="121" href="#121">121</a> <strong class="jxr_keyword">if</strong> (!context.isRetryDespiteFastFailMode()) { <em class="jxr_comment">// we don't have to retry</em> -<a class="jxr_linenumber" name="122" href="#122">122</a> LOG.debug(<span class="jxr_string">"Throwing PFFE : "</span> + context.getFailureInfo() + <span class="jxr_string">" tries : "</span> -<a class="jxr_linenumber" name="123" href="#123">123</a> + context.getTries()); -<a class="jxr_linenumber" name="124" href="#124">124</a> <strong class="jxr_keyword">throw</strong> <strong class="jxr_keyword">new</strong> <a href="../../../../../org/apache/hadoop/hbase/exceptions/PreemptiveFastFailException.html">PreemptiveFastFailException</a>( -<a class="jxr_linenumber" name="125" href="#125">125</a> context.getFailureInfo().numConsecutiveFailures.get(), -<a class="jxr_linenumber" name="126" href="#126">126</a> context.getFailureInfo().timeOfFirstFailureMilliSec, -<a class="jxr_linenumber" name="127" href="#127">127</a> context.getFailureInfo().timeOfLatestAttemptMilliSec, context.getServer()); -<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> context.setDidTry(<strong class="jxr_keyword">true</strong>); -<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> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> handleFailure(<a href="../../../../../org/apache/hadoop/hbase/client/FastFailInterceptorContext.html">FastFailInterceptorContext</a> context, -<a class="jxr_linenumber" name="134" href="#134">134</a> Throwable t) <strong class="jxr_keyword">throws</strong> IOException { -<a class="jxr_linenumber" name="135" href="#135">135</a> handleThrowable(t, context.getServer(), -<a class="jxr_linenumber" name="136" href="#136">136</a> context.getCouldNotCommunicateWithServer()); -<a class="jxr_linenumber" name="137" href="#137">137</a> } -<a class="jxr_linenumber" name="138" href="#138">138</a> -<a class="jxr_linenumber" name="139" href="#139">139</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> updateFailureInfo(<a href="../../../../../org/apache/hadoop/hbase/client/FastFailInterceptorContext.html">FastFailInterceptorContext</a> context) { -<a class="jxr_linenumber" name="140" href="#140">140</a> updateFailureInfoForServer(context.getServer(), context.getFailureInfo(), -<a class="jxr_linenumber" name="141" href="#141">141</a> context.didTry(), context.getCouldNotCommunicateWithServer() -<a class="jxr_linenumber" name="142" href="#142">142</a> .booleanValue(), context.isRetryDespiteFastFailMode()); -<a class="jxr_linenumber" name="143" href="#143">143</a> } -<a class="jxr_linenumber" name="144" href="#144">144</a> -<a class="jxr_linenumber" name="145" href="#145">145</a> <em class="jxr_javadoccomment">/**</em> -<a class="jxr_linenumber" name="146" href="#146">146</a> <em class="jxr_javadoccomment"> * Handles failures encountered when communicating with a server.</em> -<a class="jxr_linenumber" name="147" href="#147">147</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="148" href="#148">148</a> <em class="jxr_javadoccomment"> * Updates the FailureInfo in repeatedFailuresMap to reflect the failure.</em> -<a class="jxr_linenumber" name="149" href="#149">149</a> <em class="jxr_javadoccomment"> * Throws RepeatedConnectException if the client is in Fast fail mode.</em> -<a class="jxr_linenumber" name="150" href="#150">150</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="151" href="#151">151</a> <em class="jxr_javadoccomment"> * @param serverName</em> -<a class="jxr_linenumber" name="152" href="#152">152</a> <em class="jxr_javadoccomment"> * @param t</em> -<a class="jxr_linenumber" name="153" href="#153">153</a> <em class="jxr_javadoccomment"> * - the throwable to be handled.</em> -<a class="jxr_linenumber" name="154" href="#154">154</a> <em class="jxr_javadoccomment"> * @throws PreemptiveFastFailException</em> -<a class="jxr_linenumber" name="155" href="#155">155</a> <em class="jxr_javadoccomment"> */</em> -<a class="jxr_linenumber" name="156" href="#156">156</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">void</strong> handleFailureToServer(<a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> serverName, Throwable t) { -<a class="jxr_linenumber" name="157" href="#157">157</a> <strong class="jxr_keyword">if</strong> (serverName == <strong class="jxr_keyword">null</strong> || t == <strong class="jxr_keyword">null</strong>) { -<a class="jxr_linenumber" name="158" href="#158">158</a> <strong class="jxr_keyword">return</strong>; -<a class="jxr_linenumber" name="159" href="#159">159</a> } -<a class="jxr_linenumber" name="160" href="#160">160</a> <strong class="jxr_keyword">long</strong> currentTime = EnvironmentEdgeManager.currentTime(); -<a class="jxr_linenumber" name="161" href="#161">161</a> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> fInfo = repeatedFailuresMap.get(serverName); -<a class="jxr_linenumber" name="162" href="#162">162</a> <strong class="jxr_keyword">if</strong> (fInfo == <strong class="jxr_keyword">null</strong>) { -<a class="jxr_linenumber" name="163" href="#163">163</a> fInfo = <strong class="jxr_keyword">new</strong> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a>(currentTime); -<a class="jxr_linenumber" name="164" href="#164">164</a> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> oldfInfo = repeatedFailuresMap.putIfAbsent(serverName, fInfo); -<a class="jxr_linenumber" name="165" href="#165">165</a> -<a class="jxr_linenumber" name="166" href="#166">166</a> <strong class="jxr_keyword">if</strong> (oldfInfo != <strong class="jxr_keyword">null</strong>) { -<a class="jxr_linenumber" name="167" href="#167">167</a> fInfo = oldfInfo; -<a class="jxr_linenumber" name="168" href="#168">168</a> } -<a class="jxr_linenumber" name="169" href="#169">169</a> } -<a class="jxr_linenumber" name="170" href="#170">170</a> fInfo.timeOfLatestAttemptMilliSec = currentTime; -<a class="jxr_linenumber" name="171" href="#171">171</a> fInfo.numConsecutiveFailures.incrementAndGet(); -<a class="jxr_linenumber" name="172" href="#172">172</a> } -<a class="jxr_linenumber" name="173" href="#173">173</a> -<a class="jxr_linenumber" name="174" href="#174">174</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> handleThrowable(Throwable t1, <a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> serverName, -<a class="jxr_linenumber" name="175" href="#175">175</a> MutableBoolean couldNotCommunicateWithServer) <strong class="jxr_keyword">throws</strong> IOException { -<a class="jxr_linenumber" name="176" href="#176">176</a> Throwable t2 = translateException(t1); -<a class="jxr_linenumber" name="177" href="#177">177</a> <strong class="jxr_keyword">boolean</strong> isLocalException = !(t2 instanceof RemoteException); -<a class="jxr_linenumber" name="178" href="#178">178</a> <strong class="jxr_keyword">if</strong> (isLocalException && isConnectionException(t2)) { +<a class="jxr_linenumber" name="20" href="#20">20</a> <strong class="jxr_keyword">import</strong> java.io.IOException; +<a class="jxr_linenumber" name="21" href="#21">21</a> <strong class="jxr_keyword">import</strong> java.util.Map.Entry; +<a class="jxr_linenumber" name="22" href="#22">22</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.ConcurrentHashMap; +<a class="jxr_linenumber" name="23" href="#23">23</a> <strong class="jxr_keyword">import</strong> java.util.concurrent.ConcurrentMap; +<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> org.apache.commons.lang.mutable.MutableBoolean; +<a class="jxr_linenumber" name="26" href="#26">26</a> <strong class="jxr_keyword">import</strong> org.apache.commons.logging.Log; +<a class="jxr_linenumber" name="27" href="#27">27</a> <strong class="jxr_keyword">import</strong> org.apache.commons.logging.LogFactory; +<a class="jxr_linenumber" name="28" href="#28">28</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.conf.Configuration; +<a class="jxr_linenumber" name="29" href="#29">29</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.HConstants; +<a class="jxr_linenumber" name="30" href="#30">30</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.ServerName; +<a class="jxr_linenumber" name="31" href="#31">31</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.classification.InterfaceAudience; +<a class="jxr_linenumber" name="32" href="#32">32</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.exceptions.ClientExceptionsUtil; +<a class="jxr_linenumber" name="33" href="#33">33</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.exceptions.PreemptiveFastFailException; +<a class="jxr_linenumber" name="34" href="#34">34</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.ipc.CallTimeoutException; +<a class="jxr_linenumber" name="35" href="#35">35</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.hbase.util.EnvironmentEdgeManager; +<a class="jxr_linenumber" name="36" href="#36">36</a> <strong class="jxr_keyword">import</strong> org.apache.hadoop.ipc.RemoteException; +<a class="jxr_linenumber" name="37" href="#37">37</a> +<a class="jxr_linenumber" name="38" href="#38">38</a> <strong class="jxr_keyword">import</strong> com.google.common.annotations.VisibleForTesting; +<a class="jxr_linenumber" name="39" href="#39">39</a> +<a class="jxr_linenumber" name="40" href="#40">40</a> <em class="jxr_javadoccomment">/**</em> +<a class="jxr_linenumber" name="41" href="#41">41</a> <em class="jxr_javadoccomment"> * </em> +<a class="jxr_linenumber" name="42" href="#42">42</a> <em class="jxr_javadoccomment"> * The concrete {@link RetryingCallerInterceptor} class that implements the preemptive fast fail</em> +<a class="jxr_linenumber" name="43" href="#43">43</a> <em class="jxr_javadoccomment"> * feature.</em> +<a class="jxr_linenumber" name="44" href="#44">44</a> <em class="jxr_javadoccomment"> * </em> +<a class="jxr_linenumber" name="45" href="#45">45</a> <em class="jxr_javadoccomment"> * The motivation is as follows : </em> +<a class="jxr_linenumber" name="46" href="#46">46</a> <em class="jxr_javadoccomment"> * In case where a large number of clients try and talk to a particular region server in hbase, if</em> +<a class="jxr_linenumber" name="47" href="#47">47</a> <em class="jxr_javadoccomment"> * the region server goes down due to network problems, we might end up in a scenario where</em> +<a class="jxr_linenumber" name="48" href="#48">48</a> <em class="jxr_javadoccomment"> * the clients would go into a state where they all start to retry.</em> +<a class="jxr_linenumber" name="49" href="#49">49</a> <em class="jxr_javadoccomment"> * This behavior will set off many of the threads in pretty much the same path and they all would be</em> +<a class="jxr_linenumber" name="50" href="#50">50</a> <em class="jxr_javadoccomment"> * sleeping giving rise to a state where the client either needs to create more threads to send new</em> +<a class="jxr_linenumber" name="51" href="#51">51</a> <em class="jxr_javadoccomment"> * requests to other hbase machines or block because the client cannot create anymore threads.</em> +<a class="jxr_linenumber" name="52" href="#52">52</a> <em class="jxr_javadoccomment"> * </em> +<a class="jxr_linenumber" name="53" href="#53">53</a> <em class="jxr_javadoccomment"> * In most cases the clients might prefer to have a bound on the number of threads that are created</em> +<a class="jxr_linenumber" name="54" href="#54">54</a> <em class="jxr_javadoccomment"> * in order to send requests to hbase. This would mostly result in the client thread starvation.</em> +<a class="jxr_linenumber" name="55" href="#55">55</a> <em class="jxr_javadoccomment"> * </em> +<a class="jxr_linenumber" name="56" href="#56">56</a> <em class="jxr_javadoccomment"> * To circumvent this problem, the approach that is being taken here under is to let 1 of the many</em> +<a class="jxr_linenumber" name="57" href="#57">57</a> <em class="jxr_javadoccomment"> * threads who are trying to contact the regionserver with connection problems and let the other</em> +<a class="jxr_linenumber" name="58" href="#58">58</a> <em class="jxr_javadoccomment"> * threads get a {@link PreemptiveFastFailException} so that they can move on and take other</em> +<a class="jxr_linenumber" name="59" href="#59">59</a> <em class="jxr_javadoccomment"> * requests.</em> +<a class="jxr_linenumber" name="60" href="#60">60</a> <em class="jxr_javadoccomment"> * </em> +<a class="jxr_linenumber" name="61" href="#61">61</a> <em class="jxr_javadoccomment"> * This would give the client more flexibility on the kind of action he would want to take in cases</em> +<a class="jxr_linenumber" name="62" href="#62">62</a> <em class="jxr_javadoccomment"> * where the regionserver is down. He can either discard the requests and send a nack upstream</em> +<a class="jxr_linenumber" name="63" href="#63">63</a> <em class="jxr_javadoccomment"> * faster or have an application level retry or buffer the requests up so as to send them down to</em> +<a class="jxr_linenumber" name="64" href="#64">64</a> <em class="jxr_javadoccomment"> * hbase later.</em> +<a class="jxr_linenumber" name="65" href="#65">65</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="66" href="#66">66</a> <em class="jxr_javadoccomment"> */</em> +<a class="jxr_linenumber" name="67" href="#67">67</a> @InterfaceAudience.Private +<a class="jxr_linenumber" name="68" href="#68">68</a> <strong class="jxr_keyword">class</strong> <a href="../../../../../org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html">PreemptiveFastFailInterceptor</a> <strong class="jxr_keyword">extends</strong> <a href="../../../../../org/apache/hadoop/hbase/client/RetryingCallerInterceptor.html">RetryingCallerInterceptor</a> { +<a class="jxr_linenumber" name="69" href="#69">69</a> +<a class="jxr_linenumber" name="70" href="#70">70</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">static</strong> <strong class="jxr_keyword">final</strong> Log LOG = LogFactory +<a class="jxr_linenumber" name="71" href="#71">71</a> .getLog(PreemptiveFastFailInterceptor.<strong class="jxr_keyword">class</strong>); +<a class="jxr_linenumber" name="72" href="#72">72</a> +<a class="jxr_linenumber" name="73" href="#73">73</a> <em class="jxr_comment">// amount of time to wait before we consider a server to be in fast fail</em> +<a class="jxr_linenumber" name="74" href="#74">74</a> <em class="jxr_comment">// mode</em> +<a class="jxr_linenumber" name="75" href="#75">75</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">final</strong> <strong class="jxr_keyword">long</strong> fastFailThresholdMilliSec; +<a class="jxr_linenumber" name="76" href="#76">76</a> +<a class="jxr_linenumber" name="77" href="#77">77</a> <em class="jxr_comment">// Keeps track of failures when we cannot talk to a server. Helps in</em> +<a class="jxr_linenumber" name="78" href="#78">78</a> <em class="jxr_comment">// fast failing clients if the server is down for a long time.</em> +<a class="jxr_linenumber" name="79" href="#79">79</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">final</strong> ConcurrentMap<ServerName, FailureInfo> repeatedFailuresMap = +<a class="jxr_linenumber" name="80" href="#80">80</a> <strong class="jxr_keyword">new</strong> ConcurrentHashMap<ServerName, FailureInfo>(); +<a class="jxr_linenumber" name="81" href="#81">81</a> +<a class="jxr_linenumber" name="82" href="#82">82</a> <em class="jxr_comment">// We populate repeatedFailuresMap every time there is a failure. So, to</em> +<a class="jxr_linenumber" name="83" href="#83">83</a> <em class="jxr_comment">// keep it from growing unbounded, we garbage collect the failure information</em> +<a class="jxr_linenumber" name="84" href="#84">84</a> <em class="jxr_comment">// every cleanupInterval.</em> +<a class="jxr_linenumber" name="85" href="#85">85</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">final</strong> <strong class="jxr_keyword">long</strong> failureMapCleanupIntervalMilliSec; +<a class="jxr_linenumber" name="86" href="#86">86</a> +<a class="jxr_linenumber" name="87" href="#87">87</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">volatile</strong> <strong class="jxr_keyword">long</strong> lastFailureMapCleanupTimeMilliSec; +<a class="jxr_linenumber" name="88" href="#88">88</a> +<a class="jxr_linenumber" name="89" href="#89">89</a> <em class="jxr_comment">// clear failure Info. Used to clean out all entries.</em> +<a class="jxr_linenumber" name="90" href="#90">90</a> <em class="jxr_comment">// A safety valve, in case the client does not exit the</em> +<a class="jxr_linenumber" name="91" href="#91">91</a> <em class="jxr_comment">// fast fail mode for any reason.</em> +<a class="jxr_linenumber" name="92" href="#92">92</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">long</strong> fastFailClearingTimeMilliSec; +<a class="jxr_linenumber" name="93" href="#93">93</a> +<a class="jxr_linenumber" name="94" href="#94">94</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">final</strong> ThreadLocal<MutableBoolean> threadRetryingInFastFailMode = +<a class="jxr_linenumber" name="95" href="#95">95</a> <strong class="jxr_keyword">new</strong> ThreadLocal<MutableBoolean>(); +<a class="jxr_linenumber" name="96" href="#96">96</a> +<a class="jxr_linenumber" name="97" href="#97">97</a> <strong class="jxr_keyword">public</strong> <a href="../../../../../org/apache/hadoop/hbase/client/PreemptiveFastFailInterceptor.html">PreemptiveFastFailInterceptor</a>(Configuration conf) { +<a class="jxr_linenumber" name="98" href="#98">98</a> <strong class="jxr_keyword">this</strong>.fastFailThresholdMilliSec = conf.getLong( +<a class="jxr_linenumber" name="99" href="#99">99</a> HConstants.HBASE_CLIENT_FAST_FAIL_THREASHOLD_MS, +<a class="jxr_linenumber" name="100" href="#100">100</a> HConstants.HBASE_CLIENT_FAST_FAIL_THREASHOLD_MS_DEFAULT); +<a class="jxr_linenumber" name="101" href="#101">101</a> <strong class="jxr_keyword">this</strong>.failureMapCleanupIntervalMilliSec = conf.getLong( +<a class="jxr_linenumber" name="102" href="#102">102</a> HConstants.HBASE_CLIENT_FAST_FAIL_CLEANUP_MS_DURATION_MS, +<a class="jxr_linenumber" name="103" href="#103">103</a> HConstants.HBASE_CLIENT_FAST_FAIL_CLEANUP_DURATION_MS_DEFAULT); +<a class="jxr_linenumber" name="104" href="#104">104</a> lastFailureMapCleanupTimeMilliSec = EnvironmentEdgeManager.currentTime(); +<a class="jxr_linenumber" name="105" href="#105">105</a> } +<a class="jxr_linenumber" name="106" href="#106">106</a> +<a class="jxr_linenumber" name="107" href="#107">107</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> intercept(<a href="../../../../../org/apache/hadoop/hbase/client/FastFailInterceptorContext.html">FastFailInterceptorContext</a> context) +<a class="jxr_linenumber" name="108" href="#108">108</a> <strong class="jxr_keyword">throws</strong> <a href="../../../../../org/apache/hadoop/hbase/exceptions/PreemptiveFastFailException.html">PreemptiveFastFailException</a> { +<a class="jxr_linenumber" name="109" href="#109">109</a> context.setFailureInfo(repeatedFailuresMap.get(context.getServer())); +<a class="jxr_linenumber" name="110" href="#110">110</a> <strong class="jxr_keyword">if</strong> (inFastFailMode(context.getServer()) && !currentThreadInFastFailMode()) { +<a class="jxr_linenumber" name="111" href="#111">111</a> <em class="jxr_comment">// In Fast-fail mode, all but one thread will fast fail. Check</em> +<a class="jxr_linenumber" name="112" href="#112">112</a> <em class="jxr_comment">// if we are that one chosen thread.</em> +<a class="jxr_linenumber" name="113" href="#113">113</a> context.setRetryDespiteFastFailMode(shouldRetryInspiteOfFastFail(context +<a class="jxr_linenumber" name="114" href="#114">114</a> .getFailureInfo())); +<a class="jxr_linenumber" name="115" href="#115">115</a> <strong class="jxr_keyword">if</strong> (!context.isRetryDespiteFastFailMode()) { <em class="jxr_comment">// we don't have to retry</em> +<a class="jxr_linenumber" name="116" href="#116">116</a> LOG.debug(<span class="jxr_string">"Throwing PFFE : "</span> + context.getFailureInfo() + <span class="jxr_string">" tries : "</span> +<a class="jxr_linenumber" name="117" href="#117">117</a> + context.getTries()); +<a class="jxr_linenumber" name="118" href="#118">118</a> <strong class="jxr_keyword">throw</strong> <strong class="jxr_keyword">new</strong> <a href="../../../../../org/apache/hadoop/hbase/exceptions/PreemptiveFastFailException.html">PreemptiveFastFailException</a>( +<a class="jxr_linenumber" name="119" href="#119">119</a> context.getFailureInfo().numConsecutiveFailures.get(), +<a class="jxr_linenumber" name="120" href="#120">120</a> context.getFailureInfo().timeOfFirstFailureMilliSec, +<a class="jxr_linenumber" name="121" href="#121">121</a> context.getFailureInfo().timeOfLatestAttemptMilliSec, context.getServer(), +<a class="jxr_linenumber" name="122" href="#122">122</a> context.getGuaranteedClientSideOnly().isTrue()); +<a class="jxr_linenumber" name="123" href="#123">123</a> } +<a class="jxr_linenumber" name="124" href="#124">124</a> } +<a class="jxr_linenumber" name="125" href="#125">125</a> context.setDidTry(<strong class="jxr_keyword">true</strong>); +<a class="jxr_linenumber" name="126" href="#126">126</a> } +<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> <strong class="jxr_keyword">void</strong> handleFailure(<a href="../../../../../org/apache/hadoop/hbase/client/FastFailInterceptorContext.html">FastFailInterceptorContext</a> context, +<a class="jxr_linenumber" name="129" href="#129">129</a> Throwable t) <strong class="jxr_keyword">throws</strong> IOException { +<a class="jxr_linenumber" name="130" href="#130">130</a> handleThrowable(t, context.getServer(), +<a class="jxr_linenumber" name="131" href="#131">131</a> context.getCouldNotCommunicateWithServer(), +<a class="jxr_linenumber" name="132" href="#132">132</a> context.getGuaranteedClientSideOnly()); +<a class="jxr_linenumber" name="133" href="#133">133</a> } +<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> <strong class="jxr_keyword">void</strong> updateFailureInfo(<a href="../../../../../org/apache/hadoop/hbase/client/FastFailInterceptorContext.html">FastFailInterceptorContext</a> context) { +<a class="jxr_linenumber" name="136" href="#136">136</a> updateFailureInfoForServer(context.getServer(), context.getFailureInfo(), +<a class="jxr_linenumber" name="137" href="#137">137</a> context.didTry(), context.getCouldNotCommunicateWithServer() +<a class="jxr_linenumber" name="138" href="#138">138</a> .booleanValue(), context.isRetryDespiteFastFailMode()); +<a class="jxr_linenumber" name="139" href="#139">139</a> } +<a class="jxr_linenumber" name="140" href="#140">140</a> +<a class="jxr_linenumber" name="141" href="#141">141</a> <em class="jxr_javadoccomment">/**</em> +<a class="jxr_linenumber" name="142" href="#142">142</a> <em class="jxr_javadoccomment"> * Handles failures encountered when communicating with a server.</em> +<a class="jxr_linenumber" name="143" href="#143">143</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="144" href="#144">144</a> <em class="jxr_javadoccomment"> * Updates the FailureInfo in repeatedFailuresMap to reflect the failure.</em> +<a class="jxr_linenumber" name="145" href="#145">145</a> <em class="jxr_javadoccomment"> * Throws RepeatedConnectException if the client is in Fast fail mode.</em> +<a class="jxr_linenumber" name="146" href="#146">146</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="147" href="#147">147</a> <em class="jxr_javadoccomment"> * @param serverName</em> +<a class="jxr_linenumber" name="148" href="#148">148</a> <em class="jxr_javadoccomment"> * @param t</em> +<a class="jxr_linenumber" name="149" href="#149">149</a> <em class="jxr_javadoccomment"> * - the throwable to be handled.</em> +<a class="jxr_linenumber" name="150" href="#150">150</a> <em class="jxr_javadoccomment"> * @throws PreemptiveFastFailException</em> +<a class="jxr_linenumber" name="151" href="#151">151</a> <em class="jxr_javadoccomment"> */</em> +<a class="jxr_linenumber" name="152" href="#152">152</a> @VisibleForTesting +<a class="jxr_linenumber" name="153" href="#153">153</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">void</strong> handleFailureToServer(<a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> serverName, Throwable t) { +<a class="jxr_linenumber" name="154" href="#154">154</a> <strong class="jxr_keyword">if</strong> (serverName == <strong class="jxr_keyword">null</strong> || t == <strong class="jxr_keyword">null</strong>) { +<a class="jxr_linenumber" name="155" href="#155">155</a> <strong class="jxr_keyword">return</strong>; +<a class="jxr_linenumber" name="156" href="#156">156</a> } +<a class="jxr_linenumber" name="157" href="#157">157</a> <strong class="jxr_keyword">long</strong> currentTime = EnvironmentEdgeManager.currentTime(); +<a class="jxr_linenumber" name="158" href="#158">158</a> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> fInfo = repeatedFailuresMap.get(serverName); +<a class="jxr_linenumber" name="159" href="#159">159</a> <strong class="jxr_keyword">if</strong> (fInfo == <strong class="jxr_keyword">null</strong>) { +<a class="jxr_linenumber" name="160" href="#160">160</a> fInfo = <strong class="jxr_keyword">new</strong> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a>(currentTime); +<a class="jxr_linenumber" name="161" href="#161">161</a> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> oldfInfo = repeatedFailuresMap.putIfAbsent(serverName, fInfo); +<a class="jxr_linenumber" name="162" href="#162">162</a> +<a class="jxr_linenumber" name="163" href="#163">163</a> <strong class="jxr_keyword">if</strong> (oldfInfo != <strong class="jxr_keyword">null</strong>) { +<a class="jxr_linenumber" name="164" href="#164">164</a> fInfo = oldfInfo; +<a class="jxr_linenumber" name="165" href="#165">165</a> } +<a class="jxr_linenumber" name="166" href="#166">166</a> } +<a class="jxr_linenumber" name="167" href="#167">167</a> fInfo.timeOfLatestAttemptMilliSec = currentTime; +<a class="jxr_linenumber" name="168" href="#168">168</a> fInfo.numConsecutiveFailures.incrementAndGet(); +<a class="jxr_linenumber" name="169" href="#169">169</a> } +<a class="jxr_linenumber" name="170" href="#170">170</a> +<a class="jxr_linenumber" name="171" href="#171">171</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> handleThrowable(Throwable t1, <a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> serverName, +<a class="jxr_linenumber" name="172" href="#172">172</a> MutableBoolean couldNotCommunicateWithServer, +<a class="jxr_linenumber" name="173" href="#173">173</a> MutableBoolean guaranteedClientSideOnly) <strong class="jxr_keyword">throws</strong> IOException { +<a class="jxr_linenumber" name="174" href="#174">174</a> Throwable t2 = ClientExceptionsUtil.translatePFFE(t1); +<a class="jxr_linenumber" name="175" href="#175">175</a> <strong class="jxr_keyword">boolean</strong> isLocalException = !(t2 instanceof RemoteException); +<a class="jxr_linenumber" name="176" href="#176">176</a> +<a class="jxr_linenumber" name="177" href="#177">177</a> <strong class="jxr_keyword">if</strong> ((isLocalException && ClientExceptionsUtil.isConnectionException(t2)) || +<a class="jxr_linenumber" name="178" href="#178">178</a> ClientExceptionsUtil.isCallQueueTooBigException(t2)) { <a class="jxr_linenumber" name="179" href="#179">179</a> couldNotCommunicateWithServer.setValue(<strong class="jxr_keyword">true</strong>); -<a class="jxr_linenumber" name="180" href="#180">180</a> handleFailureToServer(serverName, t2); -<a class="jxr_linenumber" name="181" href="#181">181</a> } -<a class="jxr_linenumber" name="182" href="#182">182</a> } -<a class="jxr_linenumber" name="183" href="#183">183</a> -<a class="jxr_linenumber" name="184" href="#184">184</a> <strong class="jxr_keyword">private</strong> Throwable translateException(Throwable t) <strong class="jxr_keyword">throws</strong> IOException { -<a class="jxr_linenumber" name="185" href="#185">185</a> <strong class="jxr_keyword">if</strong> (t instanceof NoSuchMethodError) { -<a class="jxr_linenumber" name="186" href="#186">186</a> <em class="jxr_comment">// We probably can't recover from this exception by retrying.</em> -<a class="jxr_linenumber" name="187" href="#187">187</a> LOG.error(t); -<a class="jxr_linenumber" name="188" href="#188">188</a> <strong class="jxr_keyword">throw</strong> (NoSuchMethodError) t; -<a class="jxr_linenumber" name="189" href="#189">189</a> } -<a class="jxr_linenumber" name="190" href="#190">190</a> -<a class="jxr_linenumber" name="191" href="#191">191</a> <strong class="jxr_keyword">if</strong> (t instanceof NullPointerException) { -<a class="jxr_linenumber" name="192" href="#192">192</a> <em class="jxr_comment">// The same here. This is probably a bug.</em> -<a class="jxr_linenumber" name="193" href="#193">193</a> LOG.error(t.getMessage(), t); -<a class="jxr_linenumber" name="194" href="#194">194</a> <strong class="jxr_keyword">throw</strong> (NullPointerException) t; -<a class="jxr_linenumber" name="195" href="#195">195</a> } -<a class="jxr_linenumber" name="196" href="#196">196</a> -<a class="jxr_linenumber" name="197" href="#197">197</a> <strong class="jxr_keyword">if</strong> (t instanceof UndeclaredThrowableException) { -<a class="jxr_linenumber" name="198" href="#198">198</a> t = t.getCause(); -<a class="jxr_linenumber" name="199" href="#199">199</a> } -<a class="jxr_linenumber" name="200" href="#200">200</a> <strong class="jxr_keyword">if</strong> (t instanceof RemoteException) { -<a class="jxr_linenumber" name="201" href="#201">201</a> t = ((RemoteException) t).unwrapRemoteException(); -<a class="jxr_linenumber" name="202" href="#202">202</a> } -<a class="jxr_linenumber" name="203" href="#203">203</a> <strong class="jxr_keyword">if</strong> (t instanceof DoNotRetryIOException) { -<a class="jxr_linenumber" name="204" href="#204">204</a> <strong class="jxr_keyword">throw</strong> (DoNotRetryIOException) t; -<a class="jxr_linenumber" name="205" href="#205">205</a> } -<a class="jxr_linenumber" name="206" href="#206">206</a> <strong class="jxr_keyword">if</strong> (t instanceof Error) { -<a class="jxr_linenumber" name="207" href="#207">207</a> <strong class="jxr_keyword">throw</strong> (Error) t; -<a class="jxr_linenumber" name="208" href="#208">208</a> } -<a class="jxr_linenumber" name="209" href="#209">209</a> <strong class="jxr_keyword">return</strong> t; -<a class="jxr_linenumber" name="210" href="#210">210</a> } -<a class="jxr_linenumber" name="211" href="#211">211</a> -<a class="jxr_linenumber" name="212" href="#212">212</a> <em class="jxr_javadoccomment">/**</em> -<a class="jxr_linenumber" name="213" href="#213">213</a> <em class="jxr_javadoccomment"> * Check if the exception is something that indicates that we cannot</em> -<a class="jxr_linenumber" name="214" href="#214">214</a> <em class="jxr_javadoccomment"> * contact/communicate with the server.</em> -<a class="jxr_linenumber" name="215" href="#215">215</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="216" href="#216">216</a> <em class="jxr_javadoccomment"> * @param e</em> -<a class="jxr_linenumber" name="217" href="#217">217</a> <em class="jxr_javadoccomment"> * @return true when exception indicates that the client wasn't able to make contact with server</em> -<a class="jxr_linenumber" name="218" href="#218">218</a> <em class="jxr_javadoccomment"> */</em> -<a class="jxr_linenumber" name="219" href="#219">219</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">boolean</strong> isConnectionException(Throwable e) { -<a class="jxr_linenumber" name="220" href="#220">220</a> <strong class="jxr_keyword">if</strong> (e == <strong class="jxr_keyword">null</strong>) -<a class="jxr_linenumber" name="221" href="#221">221</a> <strong class="jxr_keyword">return</strong> false; -<a class="jxr_linenumber" name="222" href="#222">222</a> <em class="jxr_comment">// This list covers most connectivity exceptions but not all.</em> -<a class="jxr_linenumber" name="223" href="#223">223</a> <em class="jxr_comment">// For example, in SocketOutputStream a plain IOException is thrown</em> -<a class="jxr_linenumber" name="224" href="#224">224</a> <em class="jxr_comment">// at times when the channel is closed.</em> -<a class="jxr_linenumber" name="225" href="#225">225</a> <strong class="jxr_keyword">return</strong> (e instanceof SocketTimeoutException -<a class="jxr_linenumber" name="226" href="#226">226</a> || e instanceof ConnectException || e instanceof ClosedChannelException -<a class="jxr_linenumber" name="227" href="#227">227</a> || e instanceof SyncFailedException || e instanceof EOFException -<a class="jxr_linenumber" name="228" href="#228">228</a> || e instanceof TimeoutException -<a class="jxr_linenumber" name="229" href="#229">229</a> || e instanceof <a href="../../../../../org/apache/hadoop/hbase/exceptions/ConnectionClosingException.html">ConnectionClosingException</a> || e instanceof <a href="../../../../../org/apache/hadoop/hbase/ipc/FailedServerException.html">FailedServerException</a>); -<a class="jxr_linenumber" name="230" href="#230">230</a> } -<a class="jxr_linenumber" name="231" href="#231">231</a> -<a class="jxr_linenumber" name="232" href="#232">232</a> <em class="jxr_javadoccomment">/**</em> -<a class="jxr_linenumber" name="233" href="#233">233</a> <em class="jxr_javadoccomment"> * Occasionally cleans up unused information in repeatedFailuresMap.</em> -<a class="jxr_linenumber" name="234" href="#234">234</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="235" href="#235">235</a> <em class="jxr_javadoccomment"> * repeatedFailuresMap stores the failure information for all remote hosts</em> -<a class="jxr_linenumber" name="236" href="#236">236</a> <em class="jxr_javadoccomment"> * that had failures. In order to avoid these from growing indefinitely,</em> -<a class="jxr_linenumber" name="237" href="#237">237</a> <em class="jxr_javadoccomment"> * occassionallyCleanupFailureInformation() will clear these up once every</em> -<a class="jxr_linenumber" name="238" href="#238">238</a> <em class="jxr_javadoccomment"> * cleanupInterval ms.</em> -<a class="jxr_linenumber" name="239" href="#239">239</a> <em class="jxr_javadoccomment"> */</em> -<a class="jxr_linenumber" name="240" href="#240">240</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">void</strong> occasionallyCleanupFailureInformation() { -<a class="jxr_linenumber" name="241" href="#241">241</a> <strong class="jxr_keyword">long</strong> now = System.currentTimeMillis(); -<a class="jxr_linenumber" name="242" href="#242">242</a> <strong class="jxr_keyword">if</strong> (!(now > lastFailureMapCleanupTimeMilliSec -<a class="jxr_linenumber" name="243" href="#243">243</a> + failureMapCleanupIntervalMilliSec)) -<a class="jxr_linenumber" name="244" href="#244">244</a> <strong class="jxr_keyword">return</strong>; -<a class="jxr_linenumber" name="245" href="#245">245</a> -<a class="jxr_linenumber" name="246" href="#246">246</a> <em class="jxr_comment">// remove entries that haven't been attempted in a while</em> -<a class="jxr_linenumber" name="247" href="#247">247</a> <em class="jxr_comment">// No synchronization needed. It is okay if multiple threads try to</em> -<a class="jxr_linenumber" name="248" href="#248">248</a> <em class="jxr_comment">// remove the entry again and again from a concurrent hash map.</em> -<a class="jxr_linenumber" name="249" href="#249">249</a> StringBuilder sb = <strong class="jxr_keyword">new</strong> StringBuilder(); -<a class="jxr_linenumber" name="250" href="#250">250</a> <strong class="jxr_keyword">for</strong> (Entry<ServerName, FailureInfo> entry : repeatedFailuresMap.entrySet()) { -<a class="jxr_linenumber" name="251" href="#251">251</a> <strong class="jxr_keyword">if</strong> (now > entry.getValue().timeOfLatestAttemptMilliSec -<a class="jxr_linenumber" name="252" href="#252">252</a> + failureMapCleanupIntervalMilliSec) { <em class="jxr_comment">// no recent failures</em> -<a class="jxr_linenumber" name="253" href="#253">253</a> repeatedFailuresMap.remove(entry.getKey()); -<a class="jxr_linenumber" name="254" href="#254">254</a> } <strong class="jxr_keyword">else</strong> <strong class="jxr_keyword">if</strong> (now > entry.getValue().timeOfFirstFailureMilliSec -<a class="jxr_linenumber" name="255" href="#255">255</a> + <strong class="jxr_keyword">this</strong>.fastFailClearingTimeMilliSec) { <em class="jxr_comment">// been failing for a long</em> -<a class="jxr_linenumber" name="256" href="#256">256</a> <em class="jxr_comment">// time</em> -<a class="jxr_linenumber" name="257" href="#257">257</a> LOG.error(entry.getKey() -<a class="jxr_linenumber" name="258" href="#258">258</a> + <span class="jxr_string">" been failing for a long time. clearing out."</span> -<a class="jxr_linenumber" name="259" href="#259">259</a> + entry.getValue().toString()); -<a class="jxr_linenumber" name="260" href="#260">260</a> repeatedFailuresMap.remove(entry.getKey()); -<a class="jxr_linenumber" name="261" href="#261">261</a> } <strong class="jxr_keyword">else</strong> { -<a class="jxr_linenumber" name="262" href="#262">262</a> sb.append(entry.getKey().toString()).append(<span class="jxr_string">" failing "</span>) -<a class="jxr_linenumber" name="263" href="#263">263</a> .append(entry.getValue().toString()).append(<span class="jxr_string">"\n"</span>); -<a class="jxr_linenumber" name="264" href="#264">264</a> } -<a class="jxr_linenumber" name="265" href="#265">265</a> } -<a class="jxr_linenumber" name="266" href="#266">266</a> <strong class="jxr_keyword">if</strong> (sb.length() > 0) { -<a class="jxr_linenumber" name="267" href="#267">267</a> LOG.warn(<span class="jxr_string">"Preemptive failure enabled for : "</span> + sb.toString()); -<a class="jxr_linenumber" name="268" href="#268">268</a> } -<a class="jxr_linenumber" name="269" href="#269">269</a> lastFailureMapCleanupTimeMilliSec = now; -<a class="jxr_linenumber" name="270" href="#270">270</a> } -<a class="jxr_linenumber" name="271" href="#271">271</a> -<a class="jxr_linenumber" name="272" href="#272">272</a> <em class="jxr_javadoccomment">/**</em> -<a class="jxr_linenumber" name="273" href="#273">273</a> <em class="jxr_javadoccomment"> * Checks to see if we are in the Fast fail mode for requests to the server.</em> -<a class="jxr_linenumber" name="274" href="#274">274</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="275" href="#275">275</a> <em class="jxr_javadoccomment"> * If a client is unable to contact a server for more than</em> -<a class="jxr_linenumber" name="276" href="#276">276</a> <em class="jxr_javadoccomment"> * fastFailThresholdMilliSec the client will get into fast fail mode.</em> -<a class="jxr_linenumber" name="277" href="#277">277</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="278" href="#278">278</a> <em class="jxr_javadoccomment"> * @param server</em> -<a class="jxr_linenumber" name="279" href="#279">279</a> <em class="jxr_javadoccomment"> * @return true if the client is in fast fail mode for the server.</em> -<a class="jxr_linenumber" name="280" href="#280">280</a> <em class="jxr_javadoccomment"> */</em> -<a class="jxr_linenumber" name="281" href="#281">281</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">boolean</strong> inFastFailMode(<a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> server) { -<a class="jxr_linenumber" name="282" href="#282">282</a> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> fInfo = repeatedFailuresMap.get(server); -<a class="jxr_linenumber" name="283" href="#283">283</a> <em class="jxr_comment">// if fInfo is null --> The server is considered good.</em> -<a class="jxr_linenumber" name="284" href="#284">284</a> <em class="jxr_comment">// If the server is bad, wait long enough to believe that the server is</em> -<a class="jxr_linenumber" name="285" href="#285">285</a> <em class="jxr_comment">// down.</em> -<a class="jxr_linenumber" name="286" href="#286">286</a> <strong class="jxr_keyword">return</strong> (fInfo != <strong class="jxr_keyword">null</strong> && -<a class="jxr_linenumber" name="287" href="#287">287</a> EnvironmentEdgeManager.currentTime() > -<a class="jxr_linenumber" name="288" href="#288">288</a> (fInfo.timeOfFirstFailureMilliSec + <strong class="jxr_keyword">this</strong>.fastFailThresholdMilliSec)); -<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> <em class="jxr_javadoccomment">/**</em> -<a class="jxr_linenumber" name="292" href="#292">292</a> <em class="jxr_javadoccomment"> * Checks to see if the current thread is already in FastFail mode for *some*</em> -<a class="jxr_linenumber" name="293" href="#293">293</a> <em class="jxr_javadoccomment"> * server.</em> -<a class="jxr_linenumber" name="294" href="#294">294</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="295" href="#295">295</a> <em class="jxr_javadoccomment"> * @return true, if the thread is already in FF mode.</em> -<a class="jxr_linenumber" name="296" href="#296">296</a> <em class="jxr_javadoccomment"> */</em> -<a class="jxr_linenumber" name="297" href="#297">297</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">boolean</strong> currentThreadInFastFailMode() { -<a class="jxr_linenumber" name="298" href="#298">298</a> <strong class="jxr_keyword">return</strong> (<strong class="jxr_keyword">this</strong>.threadRetryingInFastFailMode.get() != <strong class="jxr_keyword">null</strong> && (<strong class="jxr_keyword">this</strong>.threadRetryingInFastFailMode -<a class="jxr_linenumber" name="299" href="#299">299</a> .get().booleanValue() == <strong class="jxr_keyword">true</strong>)); -<a class="jxr_linenumber" name="300" href="#300">300</a> } +<a class="jxr_linenumber" name="180" href="#180">180</a> guaranteedClientSideOnly.setValue(!(t2 instanceof <a href="../../../../../org/apache/hadoop/hbase/ipc/CallTimeoutException.html">CallTimeoutException</a>)); +<a class="jxr_linenumber" name="181" href="#181">181</a> handleFailureToServer(serverName, t2); +<a class="jxr_linenumber" name="182" href="#182">182</a> } +<a class="jxr_linenumber" name="183" href="#183">183</a> } +<a class="jxr_linenumber" name="184" href="#184">184</a> +<a class="jxr_linenumber" name="185" href="#185">185</a> <em class="jxr_javadoccomment">/**</em> +<a class="jxr_linenumber" name="186" href="#186">186</a> <em class="jxr_javadoccomment"> * Occasionally cleans up unused information in repeatedFailuresMap.</em> +<a class="jxr_linenumber" name="187" href="#187">187</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="188" href="#188">188</a> <em class="jxr_javadoccomment"> * repeatedFailuresMap stores the failure information for all remote hosts</em> +<a class="jxr_linenumber" name="189" href="#189">189</a> <em class="jxr_javadoccomment"> * that had failures. In order to avoid these from growing indefinitely,</em> +<a class="jxr_linenumber" name="190" href="#190">190</a> <em class="jxr_javadoccomment"> * occassionallyCleanupFailureInformation() will clear these up once every</em> +<a class="jxr_linenumber" name="191" href="#191">191</a> <em class="jxr_javadoccomment"> * cleanupInterval ms.</em> +<a class="jxr_linenumber" name="192" href="#192">192</a> <em class="jxr_javadoccomment"> */</em> +<a class="jxr_linenumber" name="193" href="#193">193</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">void</strong> occasionallyCleanupFailureInformation() { +<a class="jxr_linenumber" name="194" href="#194">194</a> <strong class="jxr_keyword">long</strong> now = System.currentTimeMillis(); +<a class="jxr_linenumber" name="195" href="#195">195</a> <strong class="jxr_keyword">if</strong> (!(now > lastFailureMapCleanupTimeMilliSec +<a class="jxr_linenumber" name="196" href="#196">196</a> + failureMapCleanupIntervalMilliSec)) +<a class="jxr_linenumber" name="197" href="#197">197</a> <strong class="jxr_keyword">return</strong>; +<a class="jxr_linenumber" name="198" href="#198">198</a> +<a class="jxr_linenumber" name="199" href="#199">199</a> <em class="jxr_comment">// remove entries that haven't been attempted in a while</em> +<a class="jxr_linenumber" name="200" href="#200">200</a> <em class="jxr_comment">// No synchronization needed. It is okay if multiple threads try to</em> +<a class="jxr_linenumber" name="201" href="#201">201</a> <em class="jxr_comment">// remove the entry again and again from a concurrent hash map.</em> +<a class="jxr_linenumber" name="202" href="#202">202</a> StringBuilder sb = <strong class="jxr_keyword">new</strong> StringBuilder(); +<a class="jxr_linenumber" name="203" href="#203">203</a> <strong class="jxr_keyword">for</strong> (Entry<ServerName, FailureInfo> entry : repeatedFailuresMap.entrySet()) { +<a class="jxr_linenumber" name="204" href="#204">204</a> <strong class="jxr_keyword">if</strong> (now > entry.getValue().timeOfLatestAttemptMilliSec +<a class="jxr_linenumber" name="205" href="#205">205</a> + failureMapCleanupIntervalMilliSec) { <em class="jxr_comment">// no recent failures</em> +<a class="jxr_linenumber" name="206" href="#206">206</a> repeatedFailuresMap.remove(entry.getKey()); +<a class="jxr_linenumber" name="207" href="#207">207</a> } <strong class="jxr_keyword">else</strong> <strong class="jxr_keyword">if</strong> (now > entry.getValue().timeOfFirstFailureMilliSec +<a class="jxr_linenumber" name="208" href="#208">208</a> + <strong class="jxr_keyword">this</strong>.fastFailClearingTimeMilliSec) { <em class="jxr_comment">// been failing for a long</em> +<a class="jxr_linenumber" name="209" href="#209">209</a> <em class="jxr_comment">// time</em> +<a class="jxr_linenumber" name="210" href="#210">210</a> LOG.error(entry.getKey() +<a class="jxr_linenumber" name="211" href="#211">211</a> + <span class="jxr_string">" been failing for a long time. clearing out."</span> +<a class="jxr_linenumber" name="212" href="#212">212</a> + entry.getValue().toString()); +<a class="jxr_linenumber" name="213" href="#213">213</a> repeatedFailuresMap.remove(entry.getKey()); +<a class="jxr_linenumber" name="214" href="#214">214</a> } <strong class="jxr_keyword">else</strong> { +<a class="jxr_linenumber" name="215" href="#215">215</a> sb.append(entry.getKey().toString()).append(<span class="jxr_string">" failing "</span>) +<a class="jxr_linenumber" name="216" href="#216">216</a> .append(entry.getValue().toString()).append(<span class="jxr_string">"\n"</span>); +<a class="jxr_linenumber" name="217" href="#217">217</a> } +<a class="jxr_linenumber" name="218" href="#218">218</a> } +<a class="jxr_linenumber" name="219" href="#219">219</a> <strong class="jxr_keyword">if</strong> (sb.length() > 0) { +<a class="jxr_linenumber" name="220" href="#220">220</a> LOG.warn(<span class="jxr_string">"Preemptive failure enabled for : "</span> + sb.toString()); +<a class="jxr_linenumber" name="221" href="#221">221</a> } +<a class="jxr_linenumber" name="222" href="#222">222</a> lastFailureMapCleanupTimeMilliSec = now; +<a class="jxr_linenumber" name="223" href="#223">223</a> } +<a class="jxr_linenumber" name="224" href="#224">224</a> +<a class="jxr_linenumber" name="225" href="#225">225</a> <em class="jxr_javadoccomment">/**</em> +<a class="jxr_linenumber" name="226" href="#226">226</a> <em class="jxr_javadoccomment"> * Checks to see if we are in the Fast fail mode for requests to the server.</em> +<a class="jxr_linenumber" name="227" href="#227">227</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="228" href="#228">228</a> <em class="jxr_javadoccomment"> * If a client is unable to contact a server for more than</em> +<a class="jxr_linenumber" name="229" href="#229">229</a> <em class="jxr_javadoccomment"> * fastFailThresholdMilliSec the client will get into fast fail mode.</em> +<a class="jxr_linenumber" name="230" href="#230">230</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="231" href="#231">231</a> <em class="jxr_javadoccomment"> * @param server</em> +<a class="jxr_linenumber" name="232" href="#232">232</a> <em class="jxr_javadoccomment"> * @return true if the client is in fast fail mode for the server.</em> +<a class="jxr_linenumber" name="233" href="#233">233</a> <em class="jxr_javadoccomment"> */</em> +<a class="jxr_linenumber" name="234" href="#234">234</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">boolean</strong> inFastFailMode(<a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> server) { +<a class="jxr_linenumber" name="235" href="#235">235</a> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> fInfo = repeatedFailuresMap.get(server); +<a class="jxr_linenumber" name="236" href="#236">236</a> <em class="jxr_comment">// if fInfo is null --> The server is considered good.</em> +<a class="jxr_linenumber" name="237" href="#237">237</a> <em class="jxr_comment">// If the server is bad, wait long enough to believe that the server is</em> +<a class="jxr_linenumber" name="238" href="#238">238</a> <em class="jxr_comment">// down.</em> +<a class="jxr_linenumber" name="239" href="#239">239</a> <strong class="jxr_keyword">return</strong> (fInfo != <strong class="jxr_keyword">null</strong> && +<a class="jxr_linenumber" name="240" href="#240">240</a> EnvironmentEdgeManager.currentTime() > +<a class="jxr_linenumber" name="241" href="#241">241</a> (fInfo.timeOfFirstFailureMilliSec + <strong class="jxr_keyword">this</strong>.fastFailThresholdMilliSec)); +<a class="jxr_linenumber" name="242" href="#242">242</a> } +<a class="jxr_linenumber" name="243" href="#243">243</a> +<a class="jxr_linenumber" name="244" href="#244">244</a> <em class="jxr_javadoccomment">/**</em> +<a class="jxr_linenumber" name="245" href="#245">245</a> <em class="jxr_javadoccomment"> * Checks to see if the current thread is already in FastFail mode for *some*</em> +<a class="jxr_linenumber" name="246" href="#246">246</a> <em class="jxr_javadoccomment"> * server.</em> +<a class="jxr_linenumber" name="247" href="#247">247</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="248" href="#248">248</a> <em class="jxr_javadoccomment"> * @return true, if the thread is already in FF mode.</em> +<a class="jxr_linenumber" name="249" href="#249">249</a> <em class="jxr_javadoccomment"> */</em> +<a class="jxr_linenumber" name="250" href="#250">250</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">boolean</strong> currentThreadInFastFailMode() { +<a class="jxr_linenumber" name="251" href="#251">251</a> <strong class="jxr_keyword">return</strong> (<strong class="jxr_keyword">this</strong>.threadRetryingInFastFailMode.get() != <strong class="jxr_keyword">null</strong> && (<strong class="jxr_keyword">this</strong>.threadRetryingInFastFailMode +<a class="jxr_linenumber" name="252" href="#252">252</a> .get().booleanValue() == <strong class="jxr_keyword">true</strong>)); +<a class="jxr_linenumber" name="253" href="#253">253</a> } +<a class="jxr_linenumber" name="254" href="#254">254</a> +<a class="jxr_linenumber" name="255" href="#255">255</a> <em class="jxr_javadoccomment">/**</em> +<a class="jxr_linenumber" name="256" href="#256">256</a> <em class="jxr_javadoccomment"> * Check to see if the client should try to connnect to the server, inspite of</em> +<a class="jxr_linenumber" name="257" href="#257">257</a> <em class="jxr_javadoccomment"> * knowing that it is in the fast fail mode.</em> +<a class="jxr_linenumber" name="258" href="#258">258</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="259" href="#259">259</a> <em class="jxr_javadoccomment"> * The idea here is that we want just one client thread to be actively trying</em> +<a class="jxr_linenumber" name="260" href="#260">260</a> <em class="jxr_javadoccomment"> * to reconnect, while all the other threads trying to reach the server will</em> +<a class="jxr_linenumber" name="261" href="#261">261</a> <em class="jxr_javadoccomment"> * short circuit.</em> +<a class="jxr_linenumber" name="262" href="#262">262</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="263" href="#263">263</a> <em class="jxr_javadoccomment"> * @param fInfo</em> +<a class="jxr_linenumber" name="264" href="#264">264</a> <em class="jxr_javadoccomment"> * @return true if the client should try to connect to the server.</em> +<a class="jxr_linenumber" name="265" href="#265">265</a> <em class="jxr_javadoccomment"> */</em> +<a class="jxr_linenumber" name="266" href="#266">266</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">boolean</strong> shouldRetryInspiteOfFastFail(<a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> fInfo) { +<a class="jxr_linenumber" name="267" href="#267">267</a> <em class="jxr_comment">// We believe that the server is down, But, we want to have just one</em> +<a class="jxr_linenumber" name="268" href="#268">268</a> <em class="jxr_comment">// client</em> +<a class="jxr_linenumber" name="269" href="#269">269</a> <em class="jxr_comment">// actively trying to connect. If we are the chosen one, we will retry</em> +<a class="jxr_linenumber" name="270" href="#270">270</a> <em class="jxr_comment">// and not throw an exception.</em> +<a class="jxr_linenumber" name="271" href="#271">271</a> <strong class="jxr_keyword">if</strong> (fInfo != <strong class="jxr_keyword">null</strong> +<a class="jxr_linenumber" name="272" href="#272">272</a> && fInfo.exclusivelyRetringInspiteOfFastFail.compareAndSet(false, <strong class="jxr_keyword">true</strong>)) { +<a class="jxr_linenumber" name="273" href="#273">273</a> MutableBoolean threadAlreadyInFF = <strong class="jxr_keyword">this</strong>.threadRetryingInFastFailMode +<a class="jxr_linenumber" name="274" href="#274">274</a> .get(); +<a class="jxr_linenumber" name="275" href="#275">275</a> <strong class="jxr_keyword">if</strong> (threadAlreadyInFF == <strong class="jxr_keyword">null</strong>) { +<a class="jxr_linenumber" name="276" href="#276">276</a> threadAlreadyInFF = <strong class="jxr_keyword">new</strong> MutableBoolean(); +<a class="jxr_linenumber" name="277" href="#277">277</a> <strong class="jxr_keyword">this</strong>.threadRetryingInFastFailMode.set(threadAlreadyInFF); +<a class="jxr_linenumber" name="278" href="#278">278</a> } +<a class="jxr_linenumber" name="279" href="#279">279</a> threadAlreadyInFF.setValue(<strong class="jxr_keyword">true</strong>); +<a class="jxr_linenumber" name="280" href="#280">280</a> <strong class="jxr_keyword">return</strong> <strong class="jxr_keyword">true</strong>; +<a class="jxr_linenumber" name="281" href="#281">281</a> } <strong class="jxr_keyword">else</strong> { +<a class="jxr_linenumber" name="282" href="#282">282</a> <strong class="jxr_keyword">return</strong> false; +<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> <em class="jxr_javadoccomment">/**</em> +<a class="jxr_linenumber" name="287" href="#287">287</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="288" href="#288">288</a> <em class="jxr_javadoccomment"> * This function updates the Failure info for a particular server after the</em> +<a class="jxr_linenumber" name="289" href="#289">289</a> <em class="jxr_javadoccomment"> * attempt to </em> +<a class="jxr_linenumber" name="290" href="#290">290</a> <em class="jxr_javadoccomment"> *</em> +<a class="jxr_linenumber" name="291" href="#291">291</a> <em class="jxr_javadoccomment"> * @param server</em> +<a class="jxr_linenumber" name="292" href="#292">292</a> <em class="jxr_javadoccomment"> * @param fInfo</em> +<a class="jxr_linenumber" name="293" href="#293">293</a> <em class="jxr_javadoccomment"> * @param couldNotCommunicate</em> +<a class="jxr_linenumber" name="294" href="#294">294</a> <em class="jxr_javadoccomment"> * @param retryDespiteFastFailMode</em> +<a class="jxr_linenumber" name="295" href="#295">295</a> <em class="jxr_javadoccomment"> */</em> +<a class="jxr_linenumber" name="296" href="#296">296</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">void</strong> updateFailureInfoForServer(<a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> server, +<a class="jxr_linenumber" name="297" href="#297">297</a> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> fInfo, <strong class="jxr_keyword">boolean</strong> didTry, <strong class="jxr_keyword">boolean</strong> couldNotCommunicate, +<a class="jxr_linenumber" name="298" href="#298">298</a> <strong class="jxr_keyword">boolean</strong> retryDespiteFastFailMode) { +<a class="jxr_linenumber" name="299" href="#299">299</a> <strong class="jxr_keyword">if</strong> (server == <strong class="jxr_keyword">null</strong> || fInfo == <strong class="jxr_keyword">null</strong> || didTry == false) +<a class="jxr_linenumber" name="300" href="#300">300</a> <strong class="jxr_keyword">return</strong>; <a class="jxr_linenumber" name="301" href="#301">301</a> -<a class="jxr_linenumber" name="302" href="#302">302</a> <em class="jxr_javadoccomment">/**</em> -<a class="jxr_linenumber" name="303" href="#303">303</a> <em class="jxr_javadoccomment"> * Check to see if the client should try to connnect to the server, inspite of</em> -<a class="jxr_linenumber" name="304" href="#304">304</a> <em class="jxr_javadoccomment"> * knowing that it is in the fast fail mode.</em> -<a class="jxr_linenumber" name="305" href="#305">305</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="306" href="#306">306</a> <em class="jxr_javadoccomment"> * The idea here is that we want just one client thread to be actively trying</em> -<a class="jxr_linenumber" name="307" href="#307">307</a> <em class="jxr_javadoccomment"> * to reconnect, while all the other threads trying to reach the server will</em> -<a class="jxr_linenumber" name="308" href="#308">308</a> <em class="jxr_javadoccomment"> * short circuit.</em> -<a class="jxr_linenumber" name="309" href="#309">309</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="310" href="#310">310</a> <em class="jxr_javadoccomment"> * @param fInfo</em> -<a class="jxr_linenumber" name="311" href="#311">311</a> <em class="jxr_javadoccomment"> * @return true if the client should try to connect to the server.</em> -<a class="jxr_linenumber" name="312" href="#312">312</a> <em class="jxr_javadoccomment"> */</em> -<a class="jxr_linenumber" name="313" href="#313">313</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">boolean</strong> shouldRetryInspiteOfFastFail(<a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> fInfo) { -<a class="jxr_linenumber" name="314" href="#314">314</a> <em class="jxr_comment">// We believe that the server is down, But, we want to have just one</em> -<a class="jxr_linenumber" name="315" href="#315">315</a> <em class="jxr_comment">// client</em> -<a class="jxr_linenumber" name="316" href="#316">316</a> <em class="jxr_comment">// actively trying to connect. If we are the chosen one, we will retry</em> -<a class="jxr_linenumber" name="317" href="#317">317</a> <em class="jxr_comment">// and not throw an exception.</em> -<a class="jxr_linenumber" name="318" href="#318">318</a> <strong class="jxr_keyword">if</strong> (fInfo != <strong class="jxr_keyword">null</strong> -<a class="jxr_linenumber" name="319" href="#319">319</a> && fInfo.exclusivelyRetringInspiteOfFastFail.compareAndSet(false, <strong class="jxr_keyword">true</strong>)) { -<a class="jxr_linenumber" name="320" href="#320">320</a> MutableBoolean threadAlreadyInFF = <strong class="jxr_keyword">this</strong>.threadRetryingInFastFailMode -<a class="jxr_linenumber" name="321" href="#321">321</a> .get(); -<a class="jxr_linenumber" name="322" href="#322">322</a> <strong class="jxr_keyword">if</strong> (threadAlreadyInFF == <strong class="jxr_keyword">null</strong>) { -<a class="jxr_linenumber" name="323" href="#323">323</a> threadAlreadyInFF = <strong class="jxr_keyword">new</strong> MutableBoolean(); -<a class="jxr_linenumber" name="324" href="#324">324</a> <strong class="jxr_keyword">this</strong>.threadRetryingInFastFailMode.set(threadAlreadyInFF); -<a class="jxr_linenumber" name="325" href="#325">325</a> } -<a class="jxr_linenumber" name="326" href="#326">326</a> threadAlreadyInFF.setValue(<strong class="jxr_keyword">true</strong>); -<a class="jxr_linenumber" name="327" href="#327">327</a> <strong class="jxr_keyword">return</strong> <strong class="jxr_keyword">true</strong>; -<a class="jxr_linenumber" name="328" href="#328">328</a> } <strong class="jxr_keyword">else</strong> { -<a class="jxr_linenumber" name="329" href="#329">329</a> <strong class="jxr_keyword">return</strong> false; -<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> <em class="jxr_javadoccomment">/**</em> -<a class="jxr_linenumber" name="334" href="#334">334</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="335" href="#335">335</a> <em class="jxr_javadoccomment"> * This function updates the Failure info for a particular server after the</em> -<a class="jxr_linenumber" name="336" href="#336">336</a> <em class="jxr_javadoccomment"> * attempt to </em> -<a class="jxr_linenumber" name="337" href="#337">337</a> <em class="jxr_javadoccomment"> *</em> -<a class="jxr_linenumber" name="338" href="#338">338</a> <em class="jxr_javadoccomment"> * @param server</em> -<a class="jxr_linenumber" name="339" href="#339">339</a> <em class="jxr_javadoccomment"> * @param fInfo</em> -<a class="jxr_linenumber" name="340" href="#340">340</a> <em class="jxr_javadoccomment"> * @param couldNotCommunicate</em> -<a class="jxr_linenumber" name="341" href="#341">341</a> <em class="jxr_javadoccomment"> * @param retryDespiteFastFailMode</em> -<a class="jxr_linenumber" name="342" href="#342">342</a> <em class="jxr_javadoccomment"> */</em> -<a class="jxr_linenumber" name="343" href="#343">343</a> <strong class="jxr_keyword">private</strong> <strong class="jxr_keyword">void</strong> updateFailureInfoForServer(<a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> server, -<a class="jxr_linenumber" name="344" href="#344">344</a> <a href="../../../../../org/apache/hadoop/hbase/client/FailureInfo.html">FailureInfo</a> fInfo, <strong class="jxr_keyword">boolean</strong> didTry, <strong class="jxr_keyword">boolean</strong> couldNotCommunicate, -<a class="jxr_linenumber" name="345" href="#345">345</a> <strong class="jxr_keyword">boolean</strong> retryDespiteFastFailMode) { -<a class="jxr_linenumber" name="346" href="#346">346</a> <strong class="jxr_keyword">if</strong> (server == <strong class="jxr_keyword">null</strong> || fInfo == <strong class="jxr_keyword">null</strong> || didTry == false) -<a class="jxr_linenumber" name="347" href="#347">347</a> <strong class="jxr_keyword">return</strong>; -<a class="jxr_linenumber" name="348" href="#348">348</a> -<a class="jxr_linenumber" name="349" href="#349">349</a> <em class="jxr_comment">// If we were able to connect to the server, reset the failure</em> -<a class="jxr_linenumber" name="350" href="#350">350</a> <em class="jxr_comment">// information.</em> -<a class="jxr_linenumber" name="351" href="#351">351</a> <strong class="jxr_keyword">if</strong> (couldNotCommunicate == false) { -<a class="jxr_linenumber" name="352" href="#352">352</a> LOG.info(<span class="jxr_string">"Clearing out PFFE for server "</span> + server.getServerName()); -<a class="jxr_linenumber" name="353" href="#353">353</a> repeatedFailuresMap.remove(server); -<a class="jxr_linenumber" name="354" href="#354">354</a> } <strong class="jxr_keyword">else</strong> { -<a class="jxr_linenumber" name="355" href="#355">355</a> <em class="jxr_comment">// update time of last attempt</em> -<a class="jxr_linenumber" name="356" href="#356">356</a> <strong class="jxr_keyword">long</strong> currentTime = System.currentTimeMillis(); -<a class="jxr_linenumber" name="357" href="#357">357</a> fInfo.timeOfLatestAttemptMilliSec = currentTime; -<a class="jxr_linenumber" name="358" href="#358">358</a> -<a class="jxr_linenumber" name="359" href="#359">359</a> <em class="jxr_comment">// Release the lock if we were retrying inspite of FastFail</em> -<a class="jxr_linenumber" name="360" href="#360">360</a> <strong class="jxr_keyword">if</strong> (retryDespiteFastFailMode) { -<a class="jxr_linenumber" name="361" href="#361">361</a> fInfo.exclusivelyRetringInspiteOfFastFail.set(false); -<a class="jxr_linenumber" name="362" href="#362">362</a> threadRetryingInFastFailMode.get().setValue(false); -<a class="jxr_linenumber" name="363" href="#363">363</a> } -<a class="jxr_linenumber" name="364" href="#364">364</a> } -<a class="jxr_linenumber" name="365" href="#365">365</a> -<a class="jxr_linenumber" name="366" href="#366">366</a> occasionallyCleanupFailureInformation(); -<a class="jxr_linenumber" name="367" href="#367">367</a> } -<a class="jxr_linenumber" name="368" href="#368">368</a> -<a class="jxr_linenumber" name="369" href="#369">369</a> @Override -<a class="jxr_linenumber" name="370" href="#370">370</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> intercept(<a href="../../../../../org/apache/hadoop/hbase/client/RetryingCallerInterceptorContext.html">RetryingCallerInterceptorContext</a> context) -<a class="jxr_linenumber" name="371" href="#371">371</a> <strong class="jxr_keyword">throws</strong> <a href="../../../../../org/apache/hadoop/hbase/exceptions/PreemptiveFastFailException.html">PreemptiveFastFailException</a> { -<a class="jxr_linenumber" name="372" href="#372">372</a> <strong class="jxr_keyword">if</strong> (context instanceof FastFailInterceptorContext) { -<a class="jxr_linenumber" name="373" href="#373">373</a> intercept((FastFailInterceptorContext) context); -<a class="jxr_linenumber" name="374" href="#374">374</a> } -<a class="jxr_linenumber" name="375" href="#375">375</a> } -<a class="jxr_linenumber" name="376" href="#376">376</a> -<a class="jxr_linenumber" name="377" href="#377">377</a> @Override -<a class="jxr_linenumber" name="378" href="#378">378</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> handleFailure(<a href="../../../../../org/apache/hadoop/hbase/client/RetryingCallerInterceptorContext.html">RetryingCallerInterceptorContext</a> context, -<a class="jxr_linenumber" name="379" href="#379">379</a> Throwable t) <strong class="jxr_keyword">throws</strong> IOException { -<a class="jxr_linenumber" name="380" href="#380">380</a> <strong class="jxr_keyword">if</strong> (context instanceof FastFailInterceptorContext) { -<a class="jxr_linenumber" name="381" href="#381">381</a> handleFailure((FastFailInterceptorContext) context, t); -<a class="jxr_linenumber" name="382" href="#382">382</a> } -<a class="jxr_linenumber" name="383" href="#383">383</a> } -<a class="jxr_linenumber" name="384" href="#384">384</a> -<a class="jxr_linenumber" name="385" href="#385">385</a> @Override -<a class="jxr_linenumber" name="386" href="#386">386</a> <strong class="jxr_keyword">public</strong> <strong class="jxr_keyword">void</strong> updateFailureInfo(<a href="../../../../../org/apache/hadoop/hbase/client/RetryingCallerInterceptorContext.html">RetryingCallerInterceptorContext</a> context) { -<a class="jxr_linenumber" name="387" href="#387">387</a> <strong class="jxr_keyword">if</strong> (context instanceof FastFailInterceptorContext) { -<a class="jxr_linenumber" name="388" href="#388">388</a> updateFailureInfo((FastFailInterceptorContext) context); -<a class="jxr_linenumber" name="389" href="#389">389</a> } -<a class="jxr_linenumber" name="390" href="#390">390</a> } -<a class="jxr_linenumber" name="391" href="#391">391</a> -<a class="jxr_linenumber" name="392" href="#392">392</a> @Override -<a class="jxr_linenumber" name="393" href="#393">393</a> <strong class="jxr_keyword">public</strong> <a href="../../../../../org/apache/hadoop/hbase/client/RetryingCallerInterceptorContext.html">RetryingCallerInterceptorContext</a> createEmptyContext() { -<a class="jxr_linenumber" name="394" href="#394">394</a> <strong class="jxr_keyword">return</strong> <strong class="jxr_keyword">new</strong> <a href="../../../../../org/apache/hadoop/hbase/client/FastFailInterceptorContext.html">FastFailInterceptorContext</a>(); -<a class="jxr_linenumber" name="395" href="#395">395</a> } -<a class="jxr_linenumber" name="396" href="#396">396</a> -<a class="jxr_linenumber" name="397" href="#397">397</a> <strong class="jxr_keyword">protected</strong> <strong class="jxr_keyword">boolean</strong> isServerInFailureMap(<a href="../../../../../org/apache/hadoop/hbase/ServerName.html">ServerName</a> serverName) { -<a class="jxr_linenumber" name="398" href="#398">398</a> <strong class="jxr_keyword">return</strong> <strong class="jxr_keyword">this</strong>.repeatedFailuresMap.containsKey(serverName); -<a class="jxr_linenumber" name="399" href="#399">399</a> } -<a class="jxr_linenumber" name="400" href="#400">400</a> -<a class="jxr_linenumber" name="401" href="#401">401</a> @Override -<a class="jxr_linenumber" name="402" href="#402">402</a> <strong class="jxr_keyword">public</strong> String toString() { -<a class="jxr_linenumber" name="403" href="#403">403</a> <strong class="jxr_keyword">return</strong> <span class="jxr_string">"PreemptiveFastFailInterceptor"</span>; -<a class="jxr_linenumber" name="404" href="#404">404</a> } -<a class="jxr_linenumber" name="405" href="#405">405</a> } +<a class="jxr_linenumber" name="302" href="#302">302</a> <em clas
<TRUNCATED>
