[infinispan-issues] [JBoss JIRA] Created: (ISPN-366) Document CAS-like operations in Hot Rod clearly

Document CAS-like operations in Hot Rod clearly ----------------------------------------------- Key: ISPN-366 URL: https://jira.jboss.org/jira/browse/ISPN-366 Project: Infinispan Issue Type: Task Components: Cache Server Reporter: Galder Zamarreno Assignee: Galder Zamarreno Fix For: 4.1.0.Final <mmarkus_> galderz: ping <mmarkus_> what is the meaning of "replaceIfUnmodified"? * amontenegro_ (~amonteneg(a)200.111.187.74) has joined #infinispan * amontenegro_ is now known as aamonten * clebert (~clebert@redhat/jboss/clebert) has joined #infinispan * baileyje (~baileyje@redhat/jboss/baileyje) has joined #infinispan * clebert (~clebert@redhat/jboss/clebert) has left #infinispan <galderz> manik__, hi mmarkus_ <galderz> replaceIfUnmodified ? <galderz> one sec <galderz> mmarkus_, manik_, r u planning to produce diff jars for server and client? <galderz> if so, it'd make sense for the client to be a separate module <galderz> wouldn't it? <manik_> galderz: yes, I think diff jars would make sense <mmarkus_> galderz: i think yes, because I'd like the client to be light <galderz> same here <manik_> no one would want server code (+ server deps) on their client <galderz> exactly <galderz> so, the client should be in server/hotrod-client or similar module <mmarkus_> not sure we need another module for that, only different packaging <galderz> mmarkus_, u're lucky i don't have java classes <galderz> and u can use src/java <galderz> while I use main/scala <galderz> i mean, u use main/java <galderz> but once u give a maven module scala nature <galderz> i think it ignores main/java <galderz> at least that's what I found out doing some testing <mmarkus_> even so, can't we produce two jars for the same module? <galderz> mmarkus_, u prob can but u have to start filtering classes./..etc <manik_> mmarkus_: thats a real PITA <galderz> iow, it's a probably a PITA <manik_> try wrestling with maven assemblies to do that ! ;) <sannegrinovero> hi all; nice to see activity here :) hi galderz recovered from all the walking? <galderz> sannegrinovero, yup, fully recovered now :) <sannegrinovero> nice :) <manik_> mmarkus_: much easier to just use a separate maven module <mmarkus_> manik_: I don't want to do that for sure. Not a big fan o mvn (not to say more) <galderz> mmarkus_, replaceIfUnmodified is all about the cas <sannegrinovero> can't build the snapshot currently as I'm missing a jgroups:jar:2.10.0.Alpha2 <galderz> it's a way to do a replace remotely <galderz> u want to replace, only if no one changed the value <galderz> so what do u do? <galderz> each modification has a unique value associated to it <galderz> so, when u want to mayeb replace, u get the cas first with: getWithCas <galderz> and then use the cas to make the modification <sannegrinovero> galderz is this "replaceIfUnmodified" already in trunk? sounds very useful <galderz> if the cas values do not match, someone has changed the value and replace is not execited <galderz> sannegrinovero, not yet <sannegrinovero> ok <galderz> mmarkus_, it's the same technique used by memcached: see http://github.com/trondn/memcached/blob/master/doc/protocol.txt <galderz> see cas command in that document <galderz> cas is our replaceIfUnmodified <galderz> mmarkus_, is it clear? <mmarkus_> what is cas? <galderz> Compare And Swap <galderz> used all over java.util.concurrent <mmarkus_> I know that one, just I don't see how it maps in hotrod <mmarkus_> e.g. getWithCas? <galderz> mmarkus_, each modification has a unique value associated: <galderz> for example: <galderz> if u key=k value=v1, let's say the unique value is 12345 <galderz> get with cas returns that 12345 <galderz> 12345 and v1 of course <galderz> now, u call replaceIfUnmodified on key=k with value=v2 and cas 12345 <galderz> key=k will only be replaced by v2 if, and only if, the cas in that node is still 12345 <galderz> if someone else had made a modification in the mean time, the unique value would have changed <galderz> hence, replaceIfUnmodified call would not have succeded <galderz> mmarkus_, does it make sense now? <galderz> it's a way to do a remote CAS <galderz> as memcached says: " "cas" is a check and set operation which means "store this data but <galderz> only if no one else has updated since I last fetched it."" <mmarkus_> so cas is generated on the server, each modification to the cache (cluster) resulting in a new cas value? <galderz> that cas operation does exactly the same as replaceIfUnmodified <galderz> indeed mmarkus_ <galderz> see the memcached module for an example <galderz> we discussed this in the list as well <mmarkus_> is the cache set per key or per whole cache? <galderz> cas u mean? <galderz> it's associated with each value, so is per key <mmarkus_> cas, sorry <galderz> mmarkus_, see http://anonsvn.jboss.org/repos/infinispan/trunk/server/memcached/src/main... <galderz> and <galderz> AddCommand <galderz> or SetCommand <galderz> in memcached, we set it to current time millis <galderz> and we increment it for each modification <mmarkus_> yep, makes sense. And it's not returned by default, the only way to retrieve it is through getWithCas, right? <galderz> mmarkus_, indeed <mmarkus_> galderz: I'm just thinking, is there a way to do a put AND obtain the cas value atomically? * mmarkus_ i.e. the way I understand it. if one does a 1.put (k1,v1) <galderz> not at the moment <mmarkus_> then he will need to to a getWithCas(k1) to get this value back <galderz> i wonder how often this would be needed <galderz> indeed <galderz> by making getWithCas a separate command and not using cas anywhere, i'm assuming this is rare occurrence <galderz> but maybe it's not that rar <galderz> rare <mmarkus_> now, the cas he gets on getWithCas(k1), might be the cas of a different put that happened after put(k1,v1), e.g. put(k1,v2) <galderz> mmarkus_, well, that means someone has changed it then <galderz> i think i know what u mean though <galderz> if u have a put and replaceIfUnmodified one after the other, put returning the cas would cut one network call <mmarkus_> not necesarelly <mmarkus_> (that also) <galderz> i'm all ears :) <mmarkus_> considering your example with cas= 12345 (see few lines bellow) <mmarkus_> I do a put that results in cas 12345 <mmarkus_> then immediately after, one does another put(k), resulting in cas 54321 <mmarkus_> then you do an getWithCas and obtain cas=54321 on the server <mmarkus_> now you do a replaceIfUnmodified(k, 54321) <mmarkus_> the cas will match, and the replace operation will take place. But you really wanted it not to take place, as per your example, isn't it? <galderz> well, depends who does the getWithCas <galderz> or more specifically, who did the other put <galderz> what u know is that since u called getWithCas till u do replace, no one else has modified it <galderz> and that's what u do with cas <galderz> u get something, u check the value <mmarkus_> ah, you want to use cas for getting the following gurante: you read data, and based on what you read you want to modify it. Now you only modify it if it hasn't changed in between, correct? <galderz> it matches <galderz> indeed :) <galderz> that's what a cas does :) <galderz> u get something <galderz> u assume that it has X <mmarkus_> I see it now! <galderz> :) <mmarkus_> I was thinking at a different scenario. <mmarkus_> you put data, and at a further point you want to modify it only if it hasn't changed <mmarkus_> but, yeah, this doesn't sound as common as the other scenario, with read * aslak has quit (Quit: Ex-Chat) <mmarkus_> I got it now - thanks for the explanation! * manik_ has quit (Quit: manik_) <galderz> mmarkus_, u're welcome :) <galderz> we'll have to make sure this is explained properly in the docu <galderz> i'll add a jira -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: https://jira.jboss.org/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira