Can you guys go into details on what will happen during a rolling upgrade exactly?
Gwen On Thu, Apr 7, 2016 at 11:21 AM, Ashish Singh <asi...@cloudera.com> wrote: > Hello Harsha, > > On Thu, Apr 7, 2016 at 11:03 AM, Harsha <m...@harsha.io> wrote: > > "My only ask is to have this in 0.10. As Jay pointed out, right now > > there > > are not many implementations out there, we might want to fix it ASAP." > > > > Probably there aren't many implementations but there are lot of users > > using these implementations in production clusters. > > Isn't this going to break the rolling upgrade? > > It will and it is a concern, in my previous mail I have mentioned this as > an issue if we choose to go this route. However, if we actually decide to > do this, I would say it is better to do it sooner than later, as fewer > implementations will be affected. Below is excerpt from my previous mail. > > Increase scope of KIP-50 to move authorizer and related classes to a > separate package. The new package will have java interface. This will allow > implementations to not depend on kafka core and just on authorizer package, > make authorization interface follow kafka’s coding standards and will allow > java implementations to be cleaner. We can either completely drop scala > interface, which might be a pain for existing implementations, or we can > have scala interface wrap java interface. Later allows a cleaner > deprecation path for existing scala authorizer interface, however it may or > may not be feasible as Kafka server will have to somehow decide which > interface it should be looking for while loading authorizer implementation, > this can probably be solved with a config or some reflection. If we choose > to go this route, I can dig deeper. > > If we go with option 2 and commit on getting this in ASAP, preferably in > 0.10, there will be fewer implementations that will be affected. > > and also moving to Java , > > a authorizer implementation going to run inside a KafkaBroker and I > > don't see why this is necessary to move to clients package. > > Are we planning on introducing common module to have it independent of > > broker and client code? > > > Yes, I think that would take away the requirement of depending on Kafka > core from authorizer implementations. Do you suggest otherwise? > > > > -Harsha > > > > On Thu, Apr 7, 2016, at 10:52 AM, Ashish Singh wrote: > > > We might want to take a call here. Following are the options. > > > > > > 1. Let KIP-50 be the way it is, i.e., just add getDescription() to > > > existing scala authorizer interface. It will break binary > > > compatibility > > > (only when using CLI and/or AdminCommand from >= 0.10 against > > > authorizer > > > implementations based on 0.9.). If we go this route, it is a simpler > > > change > > > and existing implementations won’t have to change anything on their > > > end. > > > 2. Increase scope of KIP-50 to move authorizer and related classes > to > > > a > > > separate package. The new package will have java interface. This > will > > > allow > > > implementations to not depend on kafka core and just on authorizer > > > package, > > > make authorization interface follow kafka’s coding standards and > will > > > allow > > > java implementations to be cleaner. We can either completely drop > > > scala > > > interface, which might be a pain for existing implementations, or we > > > can > > > have scala interface wrap java interface. Later allows a cleaner > > > deprecation path for existing scala authorizer interface, however it > > > may or > > > may not be feasible as Kafka server will have to somehow decide > which > > > interface it should be looking for while loading authorizer > > > implementation, > > > this can probably be solved with a config or some reflection. If we > > > choose > > > to go this route, I can dig deeper. > > > > > > If we decide to go with option 1, I think it would be fair to say that > > > scala authorizer interface will be around for some time, as there will > be > > > more implementations relying on it. If we go with option 2 and commit > on > > > getting this in ASAP, preferably in 0.10, there will be fewer > > > implementations that will be affected. > > > > > > *Another thing to notice is that scala authorizer interface is not > > > annotated as unstable.* > > > > > > > > > On Wed, Apr 6, 2016 at 9:41 AM, Ashish Singh <asi...@cloudera.com> > > wrote: > > > > > > > I see value in minimizing breaking changes and I do not oppose the > > idea of > > > > increasing scope of KIP-50 to move auth interface to java. > > > > > > > > As authorizer implementations do not really need to depend on Kafka > > core, > > > > I would suggest that we keep authorizer interface and its components > > in a > > > > separate package. I share the concern that right now using Resource, > > > > Operation, etc, in java implementations is messy. I had to deal with > > lot of > > > > it while writing Apache Sentry plugin. > > > > > > > > My only ask is to have this in 0.10. As Jay pointed out, right now > > there > > > > are not many implementations out there, we might want to fix it ASAP. > > I can > > > > only speak of Sentry integration and I think 0.10 will be the best > for > > such > > > > a change, as I should be able to adopt the changes in Sentry > > integration > > > > before a lot of users start using it. > > > > > > > > On Wed, Apr 6, 2016 at 9:25 AM, Ismael Juma <ism...@juma.me.uk> > wrote: > > > > > > > >> It is small, but breaks binary compatibility. > > > >> > > > >> Ismael > > > >> > > > >> On Wed, Apr 6, 2016 at 5:20 PM, Grant Henke <ghe...@cloudera.com> > > wrote: > > > >> > > > >> > KIP-50 as defined is very small. I don't see any harm in putting > it > > in > > > >> as > > > >> > is and then tackling the follow up work. > > > >> > > > > >> > > > > >> > On Wed, Apr 6, 2016 at 11:16 AM, Ismael Juma <ism...@juma.me.uk> > > wrote: > > > >> > > > > >> > > Thanks Grant. I wonder if KIP-50 should just be done as part of > > this > > > >> > work. > > > >> > > > > > >> > > Ismael > > > >> > > > > > >> > > On Wed, Apr 6, 2016 at 5:12 PM, Grant Henke < > ghe...@cloudera.com> > > > >> wrote: > > > >> > > > > > >> > > > My work with KIP-4 found that many of the Scala classes used > in > > the > > > >> > > > Authorizer interface are needed in the Clients package when > > adding > > > >> the > > > >> > > > various ACL requests and responses. I also found that we don't > > have > > > >> > > > standard Exceptions defined for the authorizer interface. This > > means > > > >> > that > > > >> > > > when I add the Authorizer calls to the broker and wire > > protocols all > > > >> > > > exceptions will be reported as an "Unknown Error" back to the > > user > > > >> via > > > >> > > the > > > >> > > > wire protocol. > > > >> > > > > > > >> > > > I have written more about it on the KIP-4 wiki and created > > jiras to > > > >> > track > > > >> > > > those issues (See below). I think we should wrap up this KIP > as > > is > > > >> and > > > >> > > > tackle the Java/Exception changes as a part of those > jiras/kips. > > > >> > > > > > > >> > > > - KIP-4 "Follow Up Changes" > > > >> > > > < > > > >> > > > > > > >> > > > > > >> > > > > >> > > > https://cwiki.apache.org/confluence/display/KAFKA/KIP-4+-+Command+line+and+centralized+administrative+operations#KIP-4-Commandlineandcentralizedadministrativeoperations-FollowUpChangesfollow-up-changes > > > >> > > > > > > > >> > > > - KAFKA-3509 < > > https://issues.apache.org/jira/browse/KAFKA-3509>: > > > >> > > > Provide > > > >> > > > an Authorizer interface using the Java client enumerator > > classes > > > >> > > > - KAFKA-3507 < > > https://issues.apache.org/jira/browse/KAFKA-3507>: > > > >> > > Define > > > >> > > > standard exceptions for the Authorizer interface > > > >> > > > > > > >> > > > Thank you, > > > >> > > > Grant > > > >> > > > > > > >> > > > On Wed, Apr 6, 2016 at 10:58 AM, Jay Kreps <j...@confluent.io> > > > >> wrote: > > > >> > > > > > > >> > > > > Hey Ismael, > > > >> > > > > > > > >> > > > > Yeah I think this is a minor cleanliness thing. Since this > is > > kind > > > >> > of a > > > >> > > > > power user interface I don't feel strongly either way. > > > >> > > > > > > > >> > > > > My motivation with Scala is just that we've tried to move to > > > >> having > > > >> > the > > > >> > > > > public interfaces be Java, and as a group we definitely > > struggled > > > >> a > > > >> > lot > > > >> > > > > with understanding and maintaining Scala compatibility in > the > > > >> older > > > >> > > > > clients. > > > >> > > > > > > > >> > > > > -Jay > > > >> > > > > > > > >> > > > > On Tue, Apr 5, 2016 at 11:46 PM, Ismael Juma < > > ism...@juma.me.uk> > > > >> > > wrote: > > > >> > > > > > > > >> > > > > > Hi Jay, > > > >> > > > > > > > > >> > > > > > On Wed, Apr 6, 2016 at 3:48 AM, Jay Kreps < > j...@confluent.io > > > > > > >> > wrote: > > > >> > > > > > > > > >> > > > > > > Given that we're breaking compatibility anyway should > we: > > > >> > > > > > > > > > >> > > > > > > > > >> > > > > > We are not breaking source compatibility since the new > > method > > > >> has a > > > >> > > > > default > > > >> > > > > > implementation. I take it that you mean binary > > compatibility? > > > >> > > > > > > > > >> > > > > > > > > >> > > > > > > 1. Remove the get prefix on this method and the existing > > one > > > >> > which > > > >> > > > > > violate > > > >> > > > > > > our own code style guidelines (Oops! Kind of sad we went > > > >> through > > > >> > > the > > > >> > > > > > whole > > > >> > > > > > > KIP process and no one even flagged this) > > > >> > > > > > > > > > >> > > > > > > > > >> > > > > > I did flag this during the discussion and Ashish said he > > would > > > >> > change > > > >> > > > it > > > >> > > > > if > > > >> > > > > > other people felt that it should be changed. > > > >> > > > > > > > > >> > > > > > > > > >> > > > > > > 2. Move the interface out of scala to be a normal Java > > > >> interface > > > >> > > > > > > > > > >> > > > > > > This breaks source compatibility but probably what we > > should > > > >> have > > > >> > > > done > > > >> > > > > > > originally I suspect. Probably there are few enough > > > >> > implementations > > > >> > > > of > > > >> > > > > > this > > > >> > > > > > > that it is better to just rip the bandaid off. > > > >> > > > > > > > > > >> > > > > > > > > >> > > > > > Can you please explain the motivation? It did come up in > > > >> previous > > > >> > > > > > discussions that some things like Operation and > ResourceType > > > >> should > > > >> > > be > > > >> > > > in > > > >> > > > > > the clients library, but not Authorizer itself. Are we > > saying > > > >> that > > > >> > > any > > > >> > > > > > pluggable interface should be in Java so that users can > > > >> implement > > > >> > it > > > >> > > > > > without including the Scala library? > > > >> > > > > > > > > >> > > > > > Grant, you originally suggested that some things would > have > > to > > > >> be > > > >> > in > > > >> > > > the > > > >> > > > > > Java side as well. Can you please elaborate on this? > > > >> > > > > > > > > >> > > > > > Ismael > > > >> > > > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > -- > > > >> > > > Grant Henke > > > >> > > > Software Engineer | Cloudera > > > >> > > > gr...@cloudera.com | twitter.com/gchenke | > > > >> linkedin.com/in/granthenke > > > >> > > > > > > >> > > > > > >> > > > > >> > > > > >> > > > > >> > -- > > > >> > Grant Henke > > > >> > Software Engineer | Cloudera > > > >> > gr...@cloudera.com | twitter.com/gchenke | > > linkedin.com/in/granthenke > > > >> > > > > >> > > > > > > > > > > > > > > > > -- > > > > > > > > Regards, > > > > Ashish > > > > > > > > > > > > > > > > -- > > > > > > Regards, > > > Ashish > > > > -- > > Regards, > Ashish >
