Val, I don't really like the idea of automatic redeployment of services when classes change. Different nodes may detect these changes at different moments in time, so there won't be any guarantee, that all nodes have the same version. And if redeployment fails, then there won't be a way to notify user code about it. Also service fields may change between versions, so already deployed services won't be able to be deserialized, using new classes.
I think, it would be better if user could trigger redeployment manually. It would solve the mentioned problems and let the user redeploy services, even when only their field parameters change without implementation changes. What do you think? Denis чт, 5 апр. 2018 г. в 22:37, Denis Magda <dma...@apache.org>: > Val, > > Sounds like a great solution. I'm totally for it. > > -- > Denis > > On Thu, Apr 5, 2018 at 12:32 PM, Valentin Kulichenko < > valentin.kuliche...@gmail.com> wrote: > > > Denis, > > > > This is why I'm suggesting to use DeploymentSpi for this. The way I see > > this is that instead of deploying classes on local classpath, user can > > deploy them in the storage that SPI points to. If class is updated in the > > storage, Ignite detects this and automatically restarts the service. This > > is a very simple and straightforward approach that doesn't required a lot > > of changes on our side and allows to reuse existing implementation of > > DeploymentSpi. > > > > -Val > > > > On Thu, Apr 5, 2018 at 12:13 PM, Denis Magda <dma...@gridgain.com> > wrote: > > > > > > > > > > There is no need to deserialize services on the coordinator. It > should > > > only > > > > be able to calculate the assignments. > > > > *LazyServiceConfiguration *should be used to deliver the service > > > > configurations, just like it is done right now. > > > > > > > > > Can that configuration be tweaked over the time requiring to update the > > > class on all the nodes (if, for instance, someone wants to deploy the > > next > > > version of a service)? Just want to be sure we don't need to restart > the > > > cluster nodes (that won't be used for service deployments) on > > > services-related configurational changes. > > > > > > -- > > > Denis > > > > > > On Thu, Apr 5, 2018 at 8:18 AM, Denis Mekhanikov < > dmekhani...@gmail.com> > > > wrote: > > > > > > > Denis, > > > > There is no need to deserialize services on the coordinator. It > should > > > only > > > > be able to calculate the assignments. > > > > *LazyServiceConfiguration *should be used to deliver the service > > > > configurations, just like it is done right now. > > > > > > > > Val, > > > > Usage of DeploymentSpi is a good idea, I didn't think about this > > > > possibility. > > > > This is a viable alternative to peer-class-loading, not that > > > user-friendly > > > > though. > > > > But if peer-class-loading is that hard to implement, then I vote for > > > > DeploymentSpi. > > > > As far as I understand, it won't require us to do any additional > > changes > > > in > > > > Ignite, but will make users think about using a proper DeploymentSpi. > > > > Please correct me, if I'm wrong. > > > > It would be good, though, to add some examples on service > redeployment, > > > > when implementation class changes. > > > > > > > > Denis > > > > > > > > чт, 5 апр. 2018 г. в 2:33, Valentin Kulichenko < > > > > valentin.kuliche...@gmail.com>: > > > > > > > > > I don't think peer class loading is even possible for services. I > > > believe > > > > > we should reuse DeploymentSpi [1] for versioning. > > > > > > > > > > [1] https://apacheignite.readme.io/docs/deployment-spi > > > > > > > > > > -Val > > > > > > > > > > On Wed, Apr 4, 2018 at 12:52 PM, Denis Magda <dma...@gridgain.com> > > > > wrote: > > > > > > > > > > > Sorry, that was me who renamed the IEP to "Oil Change in Service > > > Grid". > > > > > Was > > > > > > writing this email after the renaming. Like that title more > because > > > > it's > > > > > > fun and highlights what we're intended to do - cleaning of our > > > service > > > > > grid > > > > > > engine and powering it up with new "liquid" (new communication > and > > > > > > deployment approach not available before). > > > > > > > > > > > > Denis > > > > > > > > > > > > > > > > > > > This message contains serialized service instance and its > > > > > configuration. > > > > > > > It is delivered to the coordinator node first, that calculates > > the > > > > > > service > > > > > > > deployment assignments and adds this information to the > message. > > > > > > > > > > > > > > > > > > I would consider using a NodeFilter first to decide where a > service > > > can > > > > > be > > > > > > potentially deployed. Otherwise, we would require service > classes > > to > > > > be > > > > > on > > > > > > every node (every node might become a coordinator) which is not > the > > > > > desired > > > > > > requirement. > > > > > > > > > > > > > > > > > > As for the peer-class-loading, I would backup up Dmitriy here. > > Let's > > > at > > > > > > least not to focus on this task for now. We should design > services > > > > > > versioning in the right way first and support it. > > > > > > > > > > > > -- > > > > > > Denis > > > > > > > > > > > > > > > > > > > > > > > > On Wed, Apr 4, 2018 at 12:20 PM, Dmitriy Setrakyan < > > > > > dsetrak...@apache.org> > > > > > > wrote: > > > > > > > > > > > > > Here is the correct link: > > > > > > > https://cwiki.apache.org/confluence/display/IGNITE/IEP- > > > > > > > 17%3A+Oil+Change+in+Service+Grid > > > > > > > > > > > > > > I have looked at the tickets there, and I believe that we > should > > > not > > > > > > > support peer-deployment for services. It is very hard and I do > > not > > > > > think > > > > > > we > > > > > > > should even try. > > > > > > > > > > > > > > I am proposing closing this ticket as Won't Fix - > > > > > > > https://issues.apache.org/jira/browse/IGNITE-975 > > > > > > > > > > > > > > D. > > > > > > > > > > > > > > On Wed, Apr 4, 2018 at 5:39 AM, Denis Mekhanikov < > > > > > dmekhani...@gmail.com> > > > > > > > wrote: > > > > > > > > > > > > > > > Vyacheslav, > > > > > > > > > > > > > > > > I've just posted my first draft of the IEP: > > > > > > > > https://cwiki.apache.org/confluence/display/IGNITE/IEP- > > > > > > > 17%3A+Service+grid+ > > > > > > > > improvements > > > > > > > > It's not finished yet, but you can get the idea from it. > > > > > > > > If you have some thoughts on your mind, please let me know, > > I'll > > > > add > > > > > > them > > > > > > > > to the IEP. > > > > > > > > > > > > > > > > Denis > > > > > > > > > > > > > > > > ср, 4 апр. 2018 г. в 13:09, Vyacheslav Daradur < > > > > daradu...@gmail.com > > > > > >: > > > > > > > > > > > > > > > > > Denis, thanks for the link. > > > > > > > > > > > > > > > > > > I looked through the task and I think that understand your > > > > redesign > > > > > > > point > > > > > > > > > now. > > > > > > > > > > > > > > > > > > Do you have a clear plan or IEP for the whole redesign? > > > > > > > > > > > > > > > > > > I'm interested in this component and I'd like to take part > in > > > the > > > > > > > > > development. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > On Mon, Apr 2, 2018 at 2:55 PM, Denis Mekhanikov < > > > > > > > dmekhani...@gmail.com> > > > > > > > > > wrote: > > > > > > > > > > Vyacheslav, > > > > > > > > > > > > > > > > > > > > Service deployment design, based on replicated utility > > cache > > > > has > > > > > > > proven > > > > > > > > > to > > > > > > > > > > be unstable and deadlock-prone. > > > > > > > > > > You can find a list of JIRA issues, connected to it, in > my > > > > > previous > > > > > > > > > letter. > > > > > > > > > > > > > > > > > > > > The intention behind it is similar to the binary metadata > > > > > redesign, > > > > > > > > that > > > > > > > > > > happened in the following ticket: IGNITE-4157 > > > > > > > > > > <https://issues.apache.org/jira/browse/IGNITE-4157> > > > > > > > > > > This change in service deployment procedure will > eliminate > > > need > > > > > for > > > > > > > > > another > > > > > > > > > > internal replicated cache > > > > > > > > > > and make service deployment more reliable on unstable > > > topology. > > > > > > > > > > > > > > > > > > > > Denis > > > > > > > > > > > > > > > > > > > > вт, 27 мар. 2018 г. в 23:21, Vyacheslav Daradur < > > > > > > daradu...@gmail.com > > > > > > > >: > > > > > > > > > > > > > > > > > > > >> Hi, Denis Mekhanikov! > > > > > > > > > >> > > > > > > > > > >> As far as I know, Ignite services are based on > IgniteCache > > > and > > > > > we > > > > > > > have > > > > > > > > > >> all its features. We can use listeners or continuous > > queries > > > > for > > > > > > > > > >> deployment synchronizations. > > > > > > > > > >> > > > > > > > > > >> Why do you want using the discovery layer for that? > > > > > > > > > >> > > > > > > > > > >> One more thing: we can use baseline approach for > services, > > > > that > > > > > > > means > > > > > > > > > >> *IgniteService.deploy()* returns ready to work service > > after > > > > > > > > > >> deployment on baseline nodes and deploy to other nodes > on > > > > > demand, > > > > > > > for > > > > > > > > > >> example when deployed service's loading will be hight. > > > > > > > > > >> > > > > > > > > > >> About versioning, maybe there is sense to extend public > > API: > > > > > > > > > >> IgniteServices.service(name, *version*)? > > > > > > > > > >> > > > > > > > > > >> At first deployment, we can compute service's hashcode > > (just > > > > for > > > > > > an > > > > > > > > > >> example) and store it, after new deployment request for > > > > services > > > > > > > with > > > > > > > > > >> an existing name we will compute new service's hashcode > > and > > > > > > compare > > > > > > > > > >> them if they have different hashcodes that we will > deploy > > > new > > > > > > > service > > > > > > > > > >> as service with a different version. > > > > > > > > > >> > > > > > > > > > >> > > > > > > > > > >> On Fri, Mar 23, 2018 at 10:03 PM, Denis Magda < > > > > > dma...@apache.org> > > > > > > > > > wrote: > > > > > > > > > >> > Denis, > > > > > > > > > >> > > > > > > > > > > >> > Thanks for the extensive analysis. There is a vast > room > > > for > > > > > > > > > optimizations > > > > > > > > > >> > on the service grid side. > > > > > > > > > >> > > > > > > > > > > >> > Yakov, Sam, Alex G., > > > > > > > > > >> > > > > > > > > > > >> > How do you like the idea of the usage of discovery > > > protocol > > > > > for > > > > > > > the > > > > > > > > > >> service > > > > > > > > > >> > grid system messages exchange? Any pitfalls? > > > > > > > > > >> > > > > > > > > > > >> > > > > > > > > > > >> > -- > > > > > > > > > >> > Denis > > > > > > > > > >> > > > > > > > > > > >> > > > > > > > > > > >> > On Fri, Mar 23, 2018 at 8:01 AM, Denis Mekhanikov < > > > > > > > > > dmekhani...@gmail.com > > > > > > > > > >> > > > > > > > > > > >> > wrote: > > > > > > > > > >> > > > > > > > > > > >> >> Igniters, > > > > > > > > > >> >> > > > > > > > > > >> >> I'd like to start a discussion on Ignite service grid > > > > > redesign. > > > > > > > > > >> >> We have a number of problems in our current > > architecture, > > > > > that > > > > > > > have > > > > > > > > > to > > > > > > > > > >> be > > > > > > > > > >> >> addressed. > > > > > > > > > >> >> > > > > > > > > > >> >> Here are the most severe ones: > > > > > > > > > >> >> > > > > > > > > > >> >> One of them is lack of guarantee, that service is > > > > > successfully > > > > > > > > > deployed > > > > > > > > > >> and > > > > > > > > > >> >> ready for work by the time, when > > > *IgniteService.deploy*()* > > > > > > > methods > > > > > > > > > >> return. > > > > > > > > > >> >> Furthermore, if an exception is thrown from > > > *Service.init() > > > > > > > > *method, > > > > > > > > > >> then > > > > > > > > > >> >> the deploying side is not able to receive it, or even > > > > > > understand, > > > > > > > > > that > > > > > > > > > >> >> service is in unusable state. > > > > > > > > > >> >> So, you may end up in such situation, when you > > deployed a > > > > > > service > > > > > > > > > >> without > > > > > > > > > >> >> receiving any errors, then called a service's method, > > and > > > > > hung > > > > > > > > > >> indefinitely > > > > > > > > > >> >> on this invocation. > > > > > > > > > >> >> JIRA ticket: > > > > > https://issues.apache.org/jira/browse/IGNITE-3392 > > > > > > > > > >> >> > > > > > > > > > >> >> Another problem is locking during service deployment > on > > > > > > unstable > > > > > > > > > >> topology. > > > > > > > > > >> >> This issue is caused by missing updates in continuous > > > query > > > > > > > > > listeners on > > > > > > > > > >> >> the internal cache. > > > > > > > > > >> >> It is hard to reproduce, but it happens sometimes. We > > > > > shouldn't > > > > > > > > allow > > > > > > > > > >> such > > > > > > > > > >> >> possibility, that deployment methods hang without > > saying > > > > > > > anything. > > > > > > > > > >> >> JIRA ticket: > > > > > https://issues.apache.org/jira/browse/IGNITE-6259 > > > > > > > > > >> >> > > > > > > > > > >> >> I think, we should change the deployment procedure to > > > make > > > > it > > > > > > > more > > > > > > > > > >> >> reliable. > > > > > > > > > >> >> Moving from operating over internal replicated > service > > > > cache > > > > > to > > > > > > > > > sending > > > > > > > > > >> >> custom discovery events seems to be a good idea. > > > > > > > > > >> >> Service deployment may trigger a discovery event, > that > > > will > > > > > > make > > > > > > > > > chosen > > > > > > > > > >> >> nodes deploy the service, and the same event will > > notify > > > > > other > > > > > > > > nodes > > > > > > > > > >> about > > > > > > > > > >> >> the deployed service instances. > > > > > > > > > >> >> It will eliminate the need for distributed > transactions > > > on > > > > > the > > > > > > > > > internal > > > > > > > > > >> >> replicated system cache, and make the service > > deployment > > > > > > protocol > > > > > > > > > more > > > > > > > > > >> >> transparent. > > > > > > > > > >> >> > > > > > > > > > >> >> There are a few points, that should be taken into > > account > > > > > > though. > > > > > > > > > >> >> > > > > > > > > > >> >> First of all, we can't wait for services to be > deployed > > > and > > > > > > > > > initialised > > > > > > > > > >> in > > > > > > > > > >> >> the discovery thread. > > > > > > > > > >> >> So, we need to make notification about service > > deployment > > > > > > result > > > > > > > > > >> >> asynchronous, presumably over communication protocol. > > > > > > > > > >> >> I can think of a procedure similar to the current > > > exchange > > > > > > > > protocol, > > > > > > > > > >> when > > > > > > > > > >> >> service deployment is initialised with an initial > > > discovery > > > > > > > > message, > > > > > > > > > >> >> followed by asynchronous notifications from the > hosting > > > > > servers > > > > > > > > over > > > > > > > > > >> >> communication. And finally, one more discovery > message > > > will > > > > > > > notify > > > > > > > > > all > > > > > > > > > >> >> nodes about the service deployment result and > location > > of > > > > the > > > > > > > > > deployed > > > > > > > > > >> >> service instances. Coordinator will be responsible > for > > > > > > collecting > > > > > > > > of > > > > > > > > > the > > > > > > > > > >> >> deployment results in this scheme. > > > > > > > > > >> >> > > > > > > > > > >> >> Another problem is failover in case, when some nodes > > fail > > > > > > during > > > > > > > > > >> deployment > > > > > > > > > >> >> or further work. > > > > > > > > > >> >> The following cases should be handled: > > > > > > > > > >> >> > > > > > > > > > >> >> 1. coordinator failure during deployment; > > > > > > > > > >> >> 2. failure of nodes, that were chosen to host the > > > > service, > > > > > > > > during > > > > > > > > > >> >> deployment; > > > > > > > > > >> >> 3. failure of nodes, that contain deployed > services, > > > > after > > > > > > the > > > > > > > > > >> >> deployment. > > > > > > > > > >> >> > > > > > > > > > >> >> The first case may be resolved by either continuation > > of > > > > > > > deployment > > > > > > > > > >> with a > > > > > > > > > >> >> new coordinator, or by cancelling it. > > > > > > > > > >> >> The second case will require another node to be > chosen > > > and > > > > > > > > notified. > > > > > > > > > >> Maybe > > > > > > > > > >> >> another discovery message will be needed. > > > > > > > > > >> >> The third case will require redeployment, so > > coordinator > > > > > should > > > > > > > > track > > > > > > > > > >> >> topology changes and redeploy failed services. > > > > > > > > > >> >> > > > > > > > > > >> >> Another good improvement would be service versioning. > > > This > > > > > > matter > > > > > > > > was > > > > > > > > > >> >> already discussed in another thread: > > > > > > > > > >> >> > > > > > > > > > >> > > > > > > > > > http://apache-ignite-developers.2346864.n4.nabble. > > > > > > > > com/Service-versioning- > > > > > > > > > >> >> td20858.html > > > > > > > > > >> >> Let's resume this discussion and state the final > > decision > > > > > here. > > > > > > > > > >> >> This feature is closely connected to peer class > > loading, > > > > > which > > > > > > is > > > > > > > > not > > > > > > > > > >> >> working for services currently. > > > > > > > > > >> >> So, service versioning should be implemented along > with > > > > peer > > > > > > > class > > > > > > > > > >> loading. > > > > > > > > > >> >> JIRA ticket for versioning: > > > > > > > > > >> >> https://issues.apache.org/jira/browse/IGNITE-6069 > > > > > > > > > >> >> Peer class loading: https://issues.apache.org/ > > > > > > > > jira/browse/IGNITE-975 > > > > > > > > > >> >> > > > > > > > > > >> >> Please share your thoughts. Constructive criticism is > > > > highly > > > > > > > > > >> appreciated. > > > > > > > > > >> >> > > > > > > > > > >> >> Denis > > > > > > > > > >> >> > > > > > > > > > >> > > > > > > > > > >> > > > > > > > > > >> > > > > > > > > > >> -- > > > > > > > > > >> Best Regards, Vyacheslav D. > > > > > > > > > >> > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > -- > > > > > > > > > Best Regards, Vyacheslav D. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >
