Re: [Neo4j] lock or not?
Got it. i really appreciate your help. On Wed, Sep 14, 2011 at 11:14 PM, McKinley wrote: > I mean that if you are not running the REST server or high availability then > you can assume that even if you only put the read lock in the Java > thread/object world, the database will not change. No other process exists > that could change it. You do not need to bother with a read lock in the > database. > > Cheers, > > McKinley > > On Wed, Sep 14, 2011 at 3:07 PM, Linan Wang wrote: > >> McKinley >> Thank you very much for the detailed explanation. however, I don't get >> the part about "only one JVM will access database". >> neo4j doesn't support multiple JVMs has write access to the same db, right? >> >> On Wed, Sep 14, 2011 at 9:14 PM, McKinley wrote: >> > If a second thread reads that there is no node with external_id 123 in >> > between the time that a first thread finds no node and elects to create >> it, >> > you will get 2 nodes with external_id 123. So yes, you need to introduce >> a >> > lock and synchronize. >> > ___ > Neo4j mailing list > User@lists.neo4j.org > https://lists.neo4j.org/mailman/listinfo/user > -- Best regards Linan Wang ___ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user
Re: [Neo4j] lock or not?
I mean that if you are not running the REST server or high availability then you can assume that even if you only put the read lock in the Java thread/object world, the database will not change. No other process exists that could change it. You do not need to bother with a read lock in the database. Cheers, McKinley On Wed, Sep 14, 2011 at 3:07 PM, Linan Wang wrote: > McKinley > Thank you very much for the detailed explanation. however, I don't get > the part about "only one JVM will access database". > neo4j doesn't support multiple JVMs has write access to the same db, right? > > On Wed, Sep 14, 2011 at 9:14 PM, McKinley wrote: > > If a second thread reads that there is no node with external_id 123 in > > between the time that a first thread finds no node and elects to create > it, > > you will get 2 nodes with external_id 123. So yes, you need to introduce > a > > lock and synchronize. > ___ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user
Re: [Neo4j] lock or not?
McKinley Thank you very much for the detailed explanation. however, I don't get the part about "only one JVM will access database". neo4j doesn't support multiple JVMs has write access to the same db, right? On Wed, Sep 14, 2011 at 9:14 PM, McKinley wrote: > If a second thread reads that there is no node with external_id 123 in > between the time that a first thread finds no node and elects to create it, > you will get 2 nodes with external_id 123. So yes, you need to introduce a > lock and synchronize. > > You can create read locks in the graph database but you probably will not > need to do that. In your case you are reading from an index and I am not > sure what locks are available for indexes. If you are running an embedded > database and you can be sure that only your one JVM will access the database > then you can just lock in the Java thread/object space and elect that other > threads must wait to read if the node exists until the first thread has > created it. If you need to communicate concurrency via the database, you > will have to do all the same locking in the Java space and add a read lock > on some control node in Neo4j. > > If you want to create a read lock on a node you can do the following: > > lockManager = graphDb.getConfig().getLockManager(); > lockManager.getReadLock(someNode); > ... > lockManager.releaseReadLock(someNode, null); > > Threading and synchronization in Java will perform great. Depending on your > model, most of the time your node will exist and two or more threads will > just queue up and get the same reference. That waiting is inefficient by > nature, but if your business rules require it then you have to do it. > > Cheers, > > McKinley > > On Wed, Sep 14, 2011 at 12:45 PM, Linan Wang wrote: > >> hi all, >> I have implemented a function >> UserFactory.get_or_create_with_external_id(long external_id). the >> function basically search the index, see if a node already created and >> has the property "external_id" equals to external_id. if there is such >> a node, return, if not, create one, assign property and index it. >> this function will be called inside of transactions, from unmanaged >> extensions. so it's multi-threaded scenario. >> my question is if i need to mark the function synchronized. my guess >> is true, and i feel it'll hurt performance badly. if my guess is >> wrong, why? thanks >> >> -- >> Best regards >> >> Linan Wang >> ___ >> Neo4j mailing list >> User@lists.neo4j.org >> https://lists.neo4j.org/mailman/listinfo/user >> > ___ > Neo4j mailing list > User@lists.neo4j.org > https://lists.neo4j.org/mailman/listinfo/user > -- Best regards Linan Wang ___ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user
Re: [Neo4j] lock or not?
If a second thread reads that there is no node with external_id 123 in between the time that a first thread finds no node and elects to create it, you will get 2 nodes with external_id 123. So yes, you need to introduce a lock and synchronize. You can create read locks in the graph database but you probably will not need to do that. In your case you are reading from an index and I am not sure what locks are available for indexes. If you are running an embedded database and you can be sure that only your one JVM will access the database then you can just lock in the Java thread/object space and elect that other threads must wait to read if the node exists until the first thread has created it. If you need to communicate concurrency via the database, you will have to do all the same locking in the Java space and add a read lock on some control node in Neo4j. If you want to create a read lock on a node you can do the following: lockManager = graphDb.getConfig().getLockManager(); lockManager.getReadLock(someNode); ... lockManager.releaseReadLock(someNode, null); Threading and synchronization in Java will perform great. Depending on your model, most of the time your node will exist and two or more threads will just queue up and get the same reference. That waiting is inefficient by nature, but if your business rules require it then you have to do it. Cheers, McKinley On Wed, Sep 14, 2011 at 12:45 PM, Linan Wang wrote: > hi all, > I have implemented a function > UserFactory.get_or_create_with_external_id(long external_id). the > function basically search the index, see if a node already created and > has the property "external_id" equals to external_id. if there is such > a node, return, if not, create one, assign property and index it. > this function will be called inside of transactions, from unmanaged > extensions. so it's multi-threaded scenario. > my question is if i need to mark the function synchronized. my guess > is true, and i feel it'll hurt performance badly. if my guess is > wrong, why? thanks > > -- > Best regards > > Linan Wang > ___ > Neo4j mailing list > User@lists.neo4j.org > https://lists.neo4j.org/mailman/listinfo/user > ___ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user
[Neo4j] lock or not?
hi all, I have implemented a function UserFactory.get_or_create_with_external_id(long external_id). the function basically search the index, see if a node already created and has the property "external_id" equals to external_id. if there is such a node, return, if not, create one, assign property and index it. this function will be called inside of transactions, from unmanaged extensions. so it's multi-threaded scenario. my question is if i need to mark the function synchronized. my guess is true, and i feel it'll hurt performance badly. if my guess is wrong, why? thanks -- Best regards Linan Wang ___ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user
