Denis
That’s great idea, thank you so much! Just one more question. What happens if client owns a lock is disconnected from Ignite cluster? Will the lock be automatically released? From: Denis Magda [via Apache Ignite Users] [mailto:ml-node+s70518n2458...@n6.nabble.com] Sent: Saturday, January 09, 2016 5:30 PM To: akaptsan Subject: Re: Is there limitation on Semaphores and Queues? Anatoly, 1. If you suppose that a contention between threads will be high in case when a single queue is used I can suggest having a number of queues equal to the number of cache partitions. As you know every cache is divided on a constant number of partitions and every key, stored in a cache, mapped to a particular one. The default number of partitions is 1024 and keys and their values, that are added to the cache, are spread uniformly across the partitions. So you can have 1024 queues that will be concurrently updated or consumed depending on a partition number of an object's key that is locked for now. This is much better then to have a queue per partition and will minimize thread contention that can be high in case of a single queue scenario. A pseudo-code looks like this: ------ // Get a reference to a distributed lock and acquire it. Object objKey = "object_key_to_lock"; Lock lock = locksCache.lock(objKey); lock.lock(); // Lock is acquired you're safe to modify your object. ..... // Get key's partition. int partition = ignite.affinity("locksCache").partition(objKey); // Get system B queue that must receive changes and that corresponds to partition retrieved above. IgniteQueue sysBQueue = getQueue(partition); // Put data to queue. ... -------- Finally answering on your question So I will have much less opened semaphores and queues. But semaphore and queue will be created and closed almost on every change If change rate is low then it's ok. Otherwise I would avoid frequent creation/deletion of these kind of objects. In any case if you go for a solution with queue per partition above then you don't need to care about queues deletion. 2. It's up to you how to implement your custom persistent storage. In its implementation you can send updates to one DB synchronously while to the other one in async fashion. BTW, you've mentioned that it's ok to deliver updates asynchronously from one system to another. Probably you'll be interested in GridGain's data center replication functionality. [1] It's not free but does it job perfectly. [1] https://gridgain.readme.io/docs/data-center-replication Regards, Denis On Sat, Jan 9, 2016 at 1:19 PM, akaptsan <[hidden email]> wrote: Denis 1. One queue per table would prevent concurrent multithreaded modification of objects in this table. I can’t afford have table-level synchronization, it must be row level 2. It not suitable. Replication must be asynchronous: - update and commit on 1-st database - send change message to 2-nd - apply change to 2-nd database Would it be better to modify my algorithm of applying changes: - Create or lock object’s semaphore - Process all messages from object’s queue - Close the queue - Close the semaphore So I will have much less opened semaphores and queues. But semaphore and queue will be created and closed almost on every change Regards, Anatoly From: Denis Magda [via Apache Ignite Users] [mailto:[hidden email] <http://user/SendEmail.jtp?type=node&node=2456&i=0> ] Sent: Saturday, January 09, 2016 12:55 PM To: akaptsan Subject: Re: Is there limitation on Semaphores and Queues? Anatoly, I see. So the rest of the options I have in my mind are the following. 1) Queues & semaphores per object. This solution doesn't look scalable cause the more objects you have the more resources (queues & semaphores) you're allocating. If you still want to go for queues I would have one per table/cache and not per object. Instead of semaphores I would use plain Ignite distributed locks [1] per table/cache. In this case you will have per table/cache synchronization and this can affect performance but probably the difference won't be so much or even better in your use case, you just need to measure. 2) As another one approach you can do all the modifications through Ignite cluster transactionally [2] thus guaranteeing that there won't be concurrent updates of an object. Ignites caches, that will be updated, will work with your own implementation of a custom persistent storage [3] that will send updates to both DBs and a transaction that is started at Ignite's cache layer will be committed only if DBs are updated as well. Hope this helps to come to a satisfactory solution on your side. [1] https://ignite.apache.org/releases/1.5.0.final/javadoc/org/apache/ignite/IgniteCache.html#lock(K) [2] https://apacheignite.readme.io/docs/transactions [3] https://apacheignite.readme.io/docs/persistent-store Regards, Denis On Sat, Jan 9, 2016 at 11:30 AM, akaptsan <[hidden email] <http://user/SendEmail.jtp?type=node&node=2455&i=0> > wrote: Denis Thank you so much for helping me! But none of your suggestions really work for me L. I can’t now switch my application to Ignate as primary storage. But I do have this idea in remote plans. I can’t use native RDBMS replication, because databases are not 100% identical (in fact probably less then 50% of DB objects are identical). I do need application-level capturing and applying of changes. And main reason why I can’t use Kafka or similar messaging system is: I need to eliminate simultaneous change of one object in the two DBs. If it happened, then the object state in DBs would be inconsistent. That’s why I want to use Ignite semaphore From: Denis Magda [via Apache Ignite Users] [mailto:[hidden email] <http://user/SendEmail.jtp?type=node&node=2454&i=0> ] Sent: Saturday, January 09, 2016 9:26 AM To: akaptsan Subject: Re: Is there limitation on Semaphores and Queues? Anatoly, Please properly subscribe to the user list (this way we will not have to manually approve your emails and you will get answers on your questions quicker). All you need to do is send an email to “ [hidden email] <http://user/SendEmail.jtp?type=node&node=2454&i=1> ” and follow simple instructions in the reply Since the content in the databases is almost identical and you have a load balancer that directs queries to one of the systems I would review current architecture design and probably switch to the following one. You can still have a single Ignite in-memory cluster that will persist data [1] in one or several databases depending on your requirements and configuration. In such a design all business model related queries will go directly to the cluster and you don't need to care about replication cause everything will be already there. Ignite supports ANSI-99 SQL so your SQL queries should work fine as well[2]. If your systems are web applications then you may want to use web session clustering in addition [3]. On the other hand, if you can't switch to such an architecture then basing on your description you are trying to implement a replication between the databases and probably solutions like Kafka Connect [4] should work perfectly fine for you. Also different RDBMS vendors provide native tools for replication between their databases. As an example replication between Oracle databases can be done using Oracle Golden Gate product. I don't get why you need to lock an Ignite object when you're applying changes, stored in queues, from one database to another that's why consider that Kafka or RDBMS native replication tool will be enough. Does any of suggestions work for you? [1] https://apacheignite.readme.io/docs/persistent-store [2] https://apacheignite.readme.io/docs/sql-queries [3] https://apacheignite.readme.io/docs/web-session-clustering [4] http://kafka.apache.org/documentation.html _____ If you reply to this email, your message will be added to the discussion below: http://apache-ignite-users.70518.x6.nabble.com/Is-there-limitation-on-Semaphores-and-Queues-tp2415p2453.html To unsubscribe from Is there limitation on Semaphores and Queues?, click here. <http://apache-ignite-users.70518.x6.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml> NAML _____ View this message in context: RE: Is there limitation on Semaphores and Queues? <http://apache-ignite-users.70518.x6.nabble.com/Is-there-limitation-on-Semaphores-and-Queues-tp2415p2454.html> Sent from the Apache Ignite Users mailing list archive <http://apache-ignite-users.70518.x6.nabble.com/> at Nabble.com. _____ If you reply to this email, your message will be added to the discussion below: http://apache-ignite-users.70518.x6.nabble.com/Is-there-limitation-on-Semaphores-and-Queues-tp2415p2455.html To unsubscribe from Is there limitation on Semaphores and Queues?, click here. <http://apache-ignite-users.70518.x6.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml> NAML _____ View this message in context: RE: Is there limitation on Semaphores and Queues? <http://apache-ignite-users.70518.x6.nabble.com/Is-there-limitation-on-Semaphores-and-Queues-tp2415p2456.html> Sent from the Apache Ignite Users mailing list archive <http://apache-ignite-users.70518.x6.nabble.com/> at Nabble.com. _____ If you reply to this email, your message will be added to the discussion below: http://apache-ignite-users.70518.x6.nabble.com/Is-there-limitation-on-Semaphores-and-Queues-tp2415p2458.html To unsubscribe from Is there limitation on Semaphores and Queues?, click here <http://apache-ignite-users.70518.x6.nabble.com/template/NamlServlet.jtp?macro=unsubscribe_by_code&node=2415&code=YWthcHRzYW5AbWFpbC5ydXwyNDE1fC0xNzU0MjEzMTg3> . <http://apache-ignite-users.70518.x6.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml> NAML -- View this message in context: http://apache-ignite-users.70518.x6.nabble.com/Is-there-limitation-on-Semaphores-and-Queues-tp2415p2481.html Sent from the Apache Ignite Users mailing list archive at Nabble.com.
