<html><head></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><br><div><div>On 13 Jun 2012, at 09:05, Dan Berindei wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div>Sanne and I resumed the meeting later yesterday afternoon, but we<br>basically just rehashed the stuff that we've been discussing before<br>lunch. Logs here:<br><br>(07:10:10 PM) jbott: Meeting ended Tue Jun 12 16:09:55 2012 UTC.<br>Information about MeetBot at <a href="http://wiki.debian.org/MeetBot">http://wiki.debian.org/MeetBot</a> . (v<br>0.1.4)<br>(07:10:10 PM) jbott: Minutes:<br><a href="http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-15.26.html">http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-15.26.html</a><br>(07:10:10 PM) jbott: Minutes (text):<br>http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-15.26.txt<br>(07:10:10 PM) jbott: Log:<br>http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-15.26.log.html<br><br><br>The main conclusion was that the number of total virtual nodes/hash<br>segments will be fixed per cluster, not per node. Kind of like the old<br>AbstractWheelConsistentHash.HASH_SPACE, only configurable. A physical<br>node will have a variable number of vnodes/segments over its lifetime.<br><br>We also decided to add a pull component to our state transfer. The<br>current NBST design requires all the nodes to push state to a joiner<br>more or less at the same time, which results in lots of congestion at<br>the network layer and sometimes even in the joiner being excluded from<br>the cluster. We have decided that a node will not start pushing data<br>as soon as it receives the PREPARE_VIEW command from the coordinator,<br>but instead it will wait for a START_PUSH command from the receiver.<br>The receiver will only ask one previous owner at a time, thus<br>eliminating the congestion.<br><br><br>We've had a lot of back-and-forth discussions about whether the CH<br>should be "non-deterministic". We agreed in the end that (I think)<br>that it's fine if the creation of the CH is not based solely on the<br>current members list, and it depends on the previous CH as well. This<br>is quite important, I think it would be hard to find an algorithm<br>based only on member list that doesn't change ownership for a lot of<br>nodes in case of a leave (even if we use the previous members list as<br>well): see https://issues.jboss.org/browse/ISPN-1275.<br></div></blockquote><div><br></div><div>Would this still be encapsulated by the existing ConsistentHash interface? We should be careful about impl details leaking into the rest of the codebase.</div><br><blockquote type="cite"><div><br>I had an idea (that I'm pretty sure I didn't explained properly in the<br>chat) that we could avoid state transfer blocking everything while<br>receiving the transaction table from a previous owner by splitting the<br>state transfer in two:<br>* In the first phase, we'd pick the new backup owners for each<br>segment, and we'd transfer all the state to them (entries, transaction<br>table, etc.)<br>* In the second phase, we'd pick a new primary owner for each segment,<br>but the primary owner can only be one of the existing backup owners.<br>Since the data has already been transferred, we can now also remove<br>the extra owners.<br></div></blockquote><div><br></div><div>Who is "we"? The joiner? The coordinator? Everyone (deterministic, triggered on an event/message)?</div><br><blockquote type="cite"><div>During the first phase, a segment could have more than numOwners<br>owners, and commands would reach both the new owners and the old<br>owners. We will need to handle commit commands for transactions that<br>the new owner doesn't have yet in its transaction table, but we would<br>not need to block prepare commands (like the current NBST design<br>does). During the second phase, the new primary owner already has the<br>transaction table, so we don't need a blocking phase either.<br></div></blockquote><div><br></div><div>So the new primary owner would queue up these commit commands? Or just ignore them/respond with a +1?</div><br><blockquote type="cite"><div><br>I didn't explain this properly in the chat because I was certain it<br>would only make sense if the coordinator initiated state transfer one<br>node at a time, making it non-deterministic. But I think if we allow<br>the CH creation algorithm to use the previous CH, we can<br>deterministically decide if the backup owners are properly balanced<br>(if not, we need to start phase 1) and if the primary owners are<br>properly balanced (if not, we need to start phase 2).<br></div></blockquote><div><br></div><div>+1 to a deterministic CH.</div><br><blockquote type="cite"><div><br><br>There is something else that I've been thinking about since yesterday<br>that might improve performance and even simplify the state transfer at<br>the cost of determinism. When state transfer fails (usually because a<br>node has died, but not necessarily), the coordinator could ask each<br>node how far it got with the state transfer in progress (how many<br>segments they got, from which owners, etc). The coordinator would then<br>create a new "base CH" based on the actually transferred data instead<br>of the actual start CH or the "pending CH", or even the whole<br>chain/tree of CHs, none of which reflect how data is effectively<br>stored in the clustered at that moment. Because this base CH would<br>reflect the actual owners of each segment, there would be less data<br>moving around in the new state transfer and we wouldn't need to keep a<br>chain/tree of previous owner lists either.<br></div></blockquote><div><br></div><div>When you say at the cost of determinism, what are the consequences of this? Some nodes may get "wrong answers" from their CH instances? And if so, then what? When these nodes contact "wrong nodes" for a specific key, would this "wrong node" then proxy to the coordinator (who has the definitive CH)?</div><br><blockquote type="cite"><div><br><br>I'm going to take a stab at implementing a new CH with a fixed number<br>of vnodes, that can take an existing CH as input and change owners as<br>little as possible. Then I'm going to try and implement the balanced<br>backup owners/balanced primary owners check as well, just to see if<br>it's really possible. I'm not going to modify the design document just<br>yet, I need to see first if it does work and what you guys think about<br>it…<br></div></blockquote><div><br></div><div>+1. It will also give you an idea of the effort involved in implementing this, and how to break up the work into subtasks, how to test, etc.</div><div><br></div><div>Cheers</div><div>Manik</div><div><br></div><br><blockquote type="cite"><div><br><br>Cheers<br>Dan<br><br><br>On Tue, Jun 12, 2012 at 4:02 PM, Manik Surtani <<a href="mailto:manik@jboss.org">manik@jboss.org</a>> wrote:<br><blockquote type="cite">Meeting minutes from part 1. Had to break for lunch. :)<br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite">Meeting ended Tue Jun 12 13:00:43 2012 UTC. Information about MeetBot <a href="athttp://wiki.debian.org/MeetBot">athttp://wiki.debian.org/MeetBot</a> . (v 0.1.4)<br></blockquote><blockquote type="cite">14:01<br></blockquote><blockquote type="cite">Minutes: <a href="http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-09.58.html">http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-09.58.html</a><br></blockquote><blockquote type="cite">14:01<br></blockquote><blockquote type="cite">Minutes (text): <a href="http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-09.58.txt">http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-09.58.txt</a><br></blockquote><blockquote type="cite">14:01<br></blockquote><blockquote type="cite">Log: <a href="http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-09.58.log.html">http://transcripts.jboss.org/meeting/irc.freenode.org/infinispan/2012/infinispan.2012-06-12-09.58.log.html</a><br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite">--<br></blockquote><blockquote type="cite">Manik Surtani<br></blockquote><blockquote type="cite"><a href="mailto:manik@jboss.org">manik@jboss.org</a><br></blockquote><blockquote type="cite"><a href="http://twitter.com/maniksurtani">twitter.com/maniksurtani</a><br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite">Project Lead, Infinispan<br></blockquote><blockquote type="cite"><a href="http://www.infinispan.org">http://www.infinispan.org</a><br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite">Platform Architect, JBoss Data Grid<br></blockquote><blockquote type="cite"><a href="http://www.redhat.com/promo/dg6beta">http://www.redhat.com/promo/dg6beta</a><br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite">_______________________________________________<br></blockquote><blockquote type="cite">infinispan-dev mailing list<br></blockquote><blockquote type="cite"><a href="mailto:infinispan-dev@lists.jboss.org">infinispan-dev@lists.jboss.org</a><br></blockquote><blockquote type="cite"><a href="https://lists.jboss.org/mailman/listinfo/infinispan-dev">https://lists.jboss.org/mailman/listinfo/infinispan-dev</a><br></blockquote><br>_______________________________________________<br>infinispan-dev mailing list<br><a href="mailto:infinispan-dev@lists.jboss.org">infinispan-dev@lists.jboss.org</a><br>https://lists.jboss.org/mailman/listinfo/infinispan-dev<br></div></blockquote></div><br><div apple-content-edited="true">
<span class="Apple-style-span" style="border-collapse: separate; color: rgb(0, 0, 0); font-family: Helvetica; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; orphans: 2; text-align: -webkit-auto; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-border-horizontal-spacing: 0px; -webkit-border-vertical-spacing: 0px; -webkit-text-decorations-in-effect: none; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px; font-size: medium; "><span class="Apple-style-span" style="border-collapse: separate; color: rgb(0, 0, 0); font-family: Helvetica; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; orphans: 2; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-border-horizontal-spacing: 0px; -webkit-border-vertical-spacing: 0px; -webkit-text-decorations-in-effect: none; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px; font-size: medium; "><div style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><div><div><div>--</div><div>Manik Surtani</div><div><a href="mailto:manik@jboss.org">manik@jboss.org</a></div><div><a href="http://twitter.com/maniksurtani">twitter.com/maniksurtani</a></div><div><br></div><div>Project Lead, Infinispan</div><div><a href="http://www.infinispan.org">http://www.infinispan.org</a></div><div><br></div><div>Platform Architect, JBoss Data Grid</div><div><a href="http://www.redhat.com/promo/dg6beta">http://www.redhat.com/promo/dg6beta</a></div></div></div></div></span></span>
</div>
<br></body></html>