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

[Galder Zamarreno (JIRA)]

     [ https://jira.jboss.org/browse/ISPN-366?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Galder Zamarreno resolved ISPN-366.
-----------------------------------

    Resolution: Done


http://community.jboss.org/docs/DOC-15604

> Document CAS-like operations in Hot Rod clearly
> -----------------------------------------------
>
>                 Key: ISPN-366
>                 URL: https://jira.jboss.org/browse/ISPN-366
>             Project: Infinispan
>          Issue Type: Task
>          Components: Cache Server
>            Reporter: Galder Zamarreno
>            Assignee: Galder Zamarreno
>            Priority: Minor
>             Fix For: 4.1.0.CR2
>
>
> <mmarkus_> galderz: ping
> <mmarkus_> what is the meaning of "replaceIfUnmodified"?
> * amontenegro_ (~amonteneg at 200.111.187.74) has joined #infinispan
> * amontenegro_ is now known as aamonten
> * clebert (~clebert at redhat/jboss/clebert) has joined #infinispan
> * baileyje (~baileyje at redhat/jboss/baileyje) has joined #infinispan
> * clebert (~clebert at 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/java/org/infinispan/server/memcached/commands/Value.java
> <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/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira