- Cross-platform binary objects are totally possible, all those thin clients support them. - User attributes can be useful, no objections here
However, I don't think we should allow arbitrary objects in user attributes. Let's make them string only, much less to worry about. And using attributes for authentication still seems weird and dirty. On Thu, Jan 23, 2020 at 12:40 PM Dmitrii Ryabov <[email protected]> wrote: > > Even if compact footer is disabled ? > Footer is checked in postWrite - much later class descriptor check. > > чт, 23 янв. 2020 г., 12:23 Alexei Scherbakov <[email protected] > >: > > > чт, 23 янв. 2020 г. в 12:17, Dmitrii Ryabov <[email protected]>: > > > > > > The protocol must be language-agnostic. If we add some features > there, > > > let's make sure they are usable from anywhere. > > > > > > That's why I want to allow primitives only. Any language can send > numbers > > > and strings. > > > > > > > In general it's possible to have cross-platform complex data structures, > > for example see protobuf. > > > > > > > > > > Binary marshaller, before packing object to byte[], will try to use > > > discovery processor and send message containing class descriptor. But > > thin > > > clients don't have discovery. Furthermore, if we write binary > marshaller > > > without class descriptor synchronization, we can get objects with > > different > > > class versions and corresponding exceptions. > > > > > > > Even if compact footer is disabled ? > > > > > > > > > > But we can say users to declare their classes in > > > META-INF/classnames.properties and current binary marshaller will works > > > good. > > > > > > > This approach doesn't looks like cross-platform. > > > > > > > > > > чт, 23 янв. 2020 г., 12:13 Alex Plehanov <[email protected]>: > > > > > > > Hello, > > > > > > > > User attributes also (besides authentication) can be used to pass > some > > > info > > > > about an application that uses a client and then display this > > information > > > > in monitoring tools. Other vendors use such approach (Oracle DB, for > > > > example, have DBMS_APPLICATION_INFO package, PostgreeSQL have > > > > application_name connection property and application information > > > available > > > > later in system views). > > > > > > > > About allowed data types: we should definitely limit attribute types > to > > > > only primitive types. Thin client binary marshaller can't send > > > information > > > > about custom types before the handshake. > > > > > > > > > > > > ср, 22 янв. 2020 г. в 21:39, Pavel Tupitsyn <[email protected]>: > > > > > > > > > I've looked through the PR more closely, trying to understand the > use > > > > case, > > > > > and there are some Java-specific things going on (left a comment). > > > > > Please keep in mind that we have thin clients in Python, Node.js, > > C++, > > > > C#. > > > > > The protocol must be language-agnostic. If we add some features > > there, > > > > > let's make sure they are usable from anywhere. > > > > > > > > > > On Wed, Jan 22, 2020 at 9:21 PM Pavel Tupitsyn < > [email protected] > > > > > > > > wrote: > > > > > > > > > > > The approach with UserAttributes map looks dirty to me and raises > > > > > > questions: > > > > > > > > > > > > - Why is UserAttributes property related to authentication? > > > > > > - UserAttributes name implies that users can put there anything > > they > > > > > want, > > > > > > but what for? What are those additional use cases? > > > > > > > > > > > > I think we should focus on a specific problem at hand and avoid > > > > > > unnecessary future-proofing. > > > > > > What are current and potential future custom authenticators and > > what > > > > kind > > > > > > of data do they need? > > > > > > > > > > > > On Wed, Jan 22, 2020 at 6:28 PM Nikita Amelchev < > > > [email protected]> > > > > > > wrote: > > > > > > > > > > > >> I think we should add this. It will provide an extra level of > > > > security. > > > > > >> This approach is used in many products, for example in AWS > (MFA). > > > [1] > > > > > >> > > > > > >> [1] > > > > > >> > > > > > > > > > > > > > > > https://docs.aws.amazon.com/general/latest/gr/aws-sec-cred-types.html#multi-factor-authentication > > > > > >> > > > > > >> ср, 22 янв. 2020 г. в 18:13, Andrey Kuznetsov < > [email protected] > > >: > > > > > >> > > > > > > >> > Hi, Pavel! > > > > > >> > > > > > > >> > Sometimes single authentication factor is not enough. > Attributes > > > > > >> proposed > > > > > >> > allow to add extra factors flexibly. > > > > > >> > > > > > > >> > ср, 22 янв. 2020 г., 17:39 Pavel Tupitsyn < > [email protected] > > >: > > > > > >> > > > > > > >> > > Token can be sent instead of a password (like git works with > > > > GitHub > > > > > >> > > tokens). > > > > > >> > > > > > > > >> > > For now I don't see a reason to include attributes into the > > > > > handshake > > > > > >> > > message. > > > > > >> > > > > > > > >> > > On Wed, Jan 22, 2020 at 5:32 PM Ilya Kasnacheev < > > > > > >> [email protected] > > > > > >> > > > > > > > > >> > > wrote: > > > > > >> > > > > > > > >> > > > Hello! > > > > > >> > > > > > > > > >> > > > One does not send security certificate as attribute. The > > only > > > > way > > > > > to > > > > > >> > > obtain > > > > > >> > > > peer security certificate is to ask SSL engine to provide > > it. > > > > > >> > > > > > > > > >> > > > Nevertheless, I can see how it can be useful with e.g. > > > Kerberos, > > > > > >> which is > > > > > >> > > > token-based IIRC. > > > > > >> > > > > > > > > >> > > > Regards, > > > > > >> > > > -- > > > > > >> > > > Ilya Kasnacheev > > > > > >> > > > > > > > > >> > > > > > > > > >> > > > ср, 22 янв. 2020 г. в 17:20, Dmitrii Ryabov < > > > > > [email protected] > > > > > >> >: > > > > > >> > > > > > > > > >> > > > > This map is something like user object from > > > > > `SecurityCredentials`. > > > > > >> > > > > Sometimes login and password are not enough for security > > > > checks. > > > > > >> For > > > > > >> > > > > example, we can send security certificate and validate > it > > > > inside > > > > > >> > > > > authenticator. > > > > > >> > > > > > > > > > >> > > > > ср, 22 янв. 2020 г., 17:16 Igor Sapego < > > [email protected] > > > >: > > > > > >> > > > > > > > > > >> > > > > > Hi Dmitrii, > > > > > >> > > > > > > > > > > >> > > > > > Can you please explain your use case? > > > > > >> > > > > > I'm not sure I'm getting what is the motivation of > this > > > > > change. > > > > > >> > > > > > > > > > > >> > > > > > Best Regards, > > > > > >> > > > > > Igor > > > > > >> > > > > > > > > > > >> > > > > > > > > > > >> > > > > > On Wed, Jan 22, 2020 at 5:11 PM Pavel Tupitsyn < > > > > > >> [email protected] > > > > > >> > > > > > > > > >> > > > > > wrote: > > > > > >> > > > > > > > > > > >> > > > > > > Hi Dmitrii, > > > > > >> > > > > > > > > > > > >> > > > > > > Honestly, I could not grasp the problem, can you > > explain > > > > it > > > > > >> in more > > > > > >> > > > > > detail? > > > > > >> > > > > > > What do we solve by adding a map with arbitrary > stuff > > to > > > > the > > > > > >> client > > > > > >> > > > > > > protocol handshake? > > > > > >> > > > > > > > > > > > >> > > > > > > On Wed, Jan 22, 2020 at 5:02 PM Dmitrii Ryabov < > > > > > >> > > > [email protected]> > > > > > >> > > > > > > wrote: > > > > > >> > > > > > > > > > > > >> > > > > > > > Hello, Igniters! > > > > > >> > > > > > > > > > > > > >> > > > > > > > I want to add the possibility of sending user > > defined > > > > > >> attributes > > > > > >> > > > from > > > > > >> > > > > > > thin > > > > > >> > > > > > > > clients. And check them inside custom > authenticator > > > > during > > > > > >> > > > handshake > > > > > >> > > > > > [1]. > > > > > >> > > > > > > > > > > > > >> > > > > > > > There is an issue in hardcoded binary writer for > > JDBC > > > > and > > > > > >> > > > > > `IgniteClient`. > > > > > >> > > > > > > > This writer searches for a classes in the JDK and > > > > > >> > > > > > > > META-INF/classnames.properties, and tries to sync > > > > > >> notdeclared > > > > > >> > > > classes > > > > > >> > > > > > > with > > > > > >> > > > > > > > cluster. But fails because current classloading > uses > > > > > >> discovery. > > > > > >> > > > > > > > > > > > > >> > > > > > > > I'd like to keep this writer and allow only > > primitive > > > > > types > > > > > >> and > > > > > >> > > > > > `String` > > > > > >> > > > > > > > for user attributes to prevent unexpected fails. I > > > think > > > > > it > > > > > >> is > > > > > >> > > > better > > > > > >> > > > > > > than > > > > > >> > > > > > > > changing writer to one with heavy classloading. > > > > > >> > > > > > > > > > > > > >> > > > > > > > Is it ok to restrict thin attributes to primitives > > and > > > > > >> 'String'? > > > > > >> > > > > > > > > > > > > >> > > > > > > > [1] > > > https://issues.apache.org/jira/browse/IGNITE-12049 > > > > > >> > > > > > > > > > > > > >> > > > > > > > > > > > >> > > > > > > > > > > >> > > > > > > > > > >> > > > > > > > > >> > > > > > > > >> > > > > > >> > > > > > >> > > > > > >> -- > > > > > >> Best wishes, > > > > > >> Amelchev Nikita > > > > > >> > > > > > > > > > > > > > > > > > > > > > > > > -- > > > > Best regards, > > Alexei Scherbakov > > >
