<h1 id="logging">Logging<a aria-hidden="true" class="anchor-heading icon-link" href="#logging"></a></h1>
<ul>
<li>what:recording events as they occur.</li>
<li>why: need to use log to restore</li>
<li>how: combine with checkpoint, periodically checkpoint to save complete state. increment log to maintain updates from that state.
<ul>
<li>synchronous logging: spend so many time to recovery</li>
<li>asynchronous: log buffered in memory, reduce overhead, but leave log entries vulnerable to loss.
<ul>
<li>GDV, global dependency vectors</li>
<li>interval is how many messages I got since beginning</li>
<li>if someone vector is bigger than the original one, it means it is inconsistence</li>
</ul>
</li>
</ul>
</li>
<li>problem: we can use prune log to reduce playback time
<ul>
<li>it may resend messages that having negative effects on system. Use incarnation numbers to enable recipients to do the same.</li>
<li>expensive in times</li>
</ul>
</li>
</ul>
<h2 id="checkpoint">checkpoint<a aria-hidden="true" class="anchor-heading icon-link" href="#checkpoint"></a></h2>
<ul>
<li>what: used to restore state. </li>
<li>why:for global consistent copy.</li>
<li>how: set point, freezing the system,no write to maintain consistency. inbound messages will not frozen but write into buffer.
<ul>
<li>uncoordinated checkpoint: 
<ul>
<li>periodically recorded its state</li>
<li>need recovery line to recover</li>
</ul>
</li>
<li>coordinated checkpoints:
<ul>
<li>record message sequences: know who sent us message since last checkpoint</li>
<li>synchronized clocks, add timestamp in all messages and act as a sequence number;:</li>
</ul>
</li>
</ul>
</li>
<li>problem: freeze</li>
<li>incarnation number: restarted will +1, use this to distinguish whether I should accept this message.</li>
</ul>
<hr>
Children
<ol>
<li><a href="/notes/c54dc5a8-42db-4a75-a182-c6c55f2c0c1d">LSM</a></li>
</ol>