Hi Zike and Michael,

I agree that we can move this method to `ClientBuilderImpl`. Generally
users should not set this description, the client version string is
used to identify a specific implementation of the Pulsar API. If there
is a strong demand to configure the description for the API layer
(`PulsarClient`), we can open a new PIP later.

So I changed the PIP to add this method for `ClientBuilderImpl`. And
still, there is a length limit for it.

Thanks,
Yunze

On Mon, Mar 13, 2023 at 3:47 PM Zike Yang <z...@apache.org> wrote:
>
> > I'd rather use the "description" term, which indicates the
> client version has extra description in addition to the client version
> string.
>
> The `description` term and 64 length limit all make sense to me.
>
> > Is there any way we can avoid giving regular users easy access to this
> field via the ClientBuilder while still letting libraries add their
> own suffix?
>
> I think we need to consider this problem. Is it possible to expose
> this field to the `ClientBuilderImpl` instead of the `ClientBuilder`?
> This will make users not easy to access this field. The `description`
> still seems confusing for the user. Are there any strong cases where
> the field `description` makes sense to the general user and not just
> the library developer?
>
> Thanks,
> Zike Yang
>
> On Mon, Mar 13, 2023 at 9:31 AM Yunze Xu <y...@streamnative.io.invalid> wrote:
> >
> > >  I think we
> > > should consider putting a character limit on the field to prevent
> > > descriptions that are too long.
> >
> > Good suggestion. A long description string is unnecessary and could be
> > used as malicious code. What do you think of limiting the length to
> > 64? I think it's long enough.
> >
> > Thanks,
> > Yunze
> >
> > On Fri, Mar 10, 2023 at 12:52 PM Michael Marshall <mmarsh...@apache.org> 
> > wrote:
> > >
> > > Great discussion. Derivative clients are an important consideration
> > > for our discussion around capturing the version information.
> > >
> > > Is there any way we can avoid giving regular users easy access to this
> > > field via the ClientBuilder while still letting libraries add their
> > > own suffix? We cannot prevent a custom library from serializing
> > > whatever they would like in the field, but making this field easily
> > > available to application code could be confusing and might decrease
> > > the value of the version information. In my mind, the goal of the
> > > version string is to give a weak signal that helps operators debug
> > > client related issues.
> > >
> > > If we do leave this field exposed to the application code, I think we
> > > should consider putting a character limit on the field to prevent
> > > descriptions that are too long.
> > >
> > > Thanks,
> > > Michael
> > >
> > > On Thu, Mar 9, 2023 at 8:28 PM Yunze Xu <y...@streamnative.io.invalid> 
> > > wrote:
> > > >
> > > > I've updated this proposal to retain the original client version
> > > > string. I'd rather use the "description" term, which indicates the
> > > > client version has extra description in addition to the client version
> > > > string.
> > > >
> > > > Thanks,
> > > > Yunze
> > > >
> > > > On Fri, Mar 10, 2023 at 10:11 AM Yunze Xu <y...@streamnative.io> wrote:
> > > > >
> > > > > Hi Zike,
> > > > >
> > > > > Good suggestion. I agree with this approach. Maybe we can name it as
> > > > > `subVersionString` to indicate that?
> > > > >
> > > > > Thanks,
> > > > > Yunze
> > > > >
> > > > > On Thu, Mar 9, 2023 at 3:14 PM Zike Yang <z...@apache.org> wrote:
> > > > > >
> > > > > > Hi Yunze,
> > > > > >
> > > > > > > I have changed this proposal to just add a config to 
> > > > > > > `ClientBuilder`.
> > > > > >
> > > > > > I propose to add a field named `clientVersionSuffix` rather than the
> > > > > > `clientVersion`. As I said before:
> > > > > > https://lists.apache.org/thread/g0128l85fkcmw4821188mjjznxbo4lhd
> > > > > >
> > > > > > This is helpful for debugging. Especially for the case of the Nodejs
> > > > > > client in which users can compile the C++ client on their own. This
> > > > > > way, we can know exactly which underlying C++ client version the 
> > > > > > user
> > > > > > uses.
> > > > > >
> > > > > > Thanks,
> > > > > > Zike Yang
> > > > > >
> > > > > > On Wed, Mar 8, 2023 at 5:17 PM Yunze Xu 
> > > > > > <y...@streamnative.io.invalid> wrote:
> > > > > > >
> > > > > > > Hi,
> > > > > > >
> > > > > > > I have changed this proposal to just add a config to 
> > > > > > > `ClientBuilder`.
> > > > > > > And here is the demo implementation:
> > > > > > > https://github.com/BewareMyPower/pulsar/pull/21/files
> > > > > > >
> > > > > > > PTAL again.
> > > > > > >
> > > > > > > Thanks,
> > > > > > > Yunze
> > > > > > >
> > > > > > > On Sat, Mar 4, 2023 at 10:39 PM Yunze Xu <y...@streamnative.io> 
> > > > > > > wrote:
> > > > > > > >
> > > > > > > > Hi Enrico,
> > > > > > > >
> > > > > > > > Thanks for your suggestion. It makes sense to me. I will think 
> > > > > > > > again
> > > > > > > > and modify this proposal.
> > > > > > > >
> > > > > > > > Hi Tison,
> > > > > > > >
> > > > > > > > I mentioned the C++ client because the initial motivation is to 
> > > > > > > > solve
> > > > > > > > the issue for the Python client and Node.js client. But after 
> > > > > > > > thinking
> > > > > > > > for a while, I believe it's more general for clients of other
> > > > > > > > languages, including Java. And this proposal is only for the 
> > > > > > > > Java
> > > > > > > > client.
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > > Yunze
> > > > > > > >
> > > > > > > > On Sat, Mar 4, 2023 at 1:42 PM tison <wander4...@gmail.com> 
> > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > I agree with Enrico that it's better to have a config option.
> > > > > > > > >
> > > > > > > > > Also, we cannot simply replace the PulsarVersion call with the
> > > > > > > > > DynamicPulsarVersion call because the client version string 
> > > > > > > > > is now
> > > > > > > > > constructed as:
> > > > > > > > >
> > > > > > > > > String.format("Pulsar-Java-v%s", PulsarVersion.getVersion())
> > > > > > > > >
> > > > > > > > > It's a config of client version string, not pulsar version.
> > > > > > > > >
> > > > > > > > > Moreover, in your proposal, you mention the case of client 
> > > > > > > > > c++ at first,
> > > > > > > > > but don't talk about it later. Is the scope of this proposal 
> > > > > > > > > in the Java
> > > > > > > > > client only?
> > > > > > > > >
> > > > > > > > > Best,
> > > > > > > > > tison.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Enrico Olivelli <eolive...@gmail.com> 于2023年3月4日周六 06:38写道:
> > > > > > > > >
> > > > > > > > > > Yunze,
> > > > > > > > > >
> > > > > > > > > > Il Ven 3 Mar 2023, 12:31 Yunze Xu 
> > > > > > > > > > <y...@streamnative.io.invalid> ha
> > > > > > > > > > scritto:
> > > > > > > > > >
> > > > > > > > > > > Hi all,
> > > > > > > > > > >
> > > > > > > > > > > Based on the previous discussion [1], I created a 
> > > > > > > > > > > proposal to support
> > > > > > > > > > > configuring client version at SDK level:
> > > > > > > > > > > https://github.com/apache/pulsar/issues/19705
> > > > > > > > > > >
> > > > > > > > > > > I've added more explanations in the motivation part, 
> > > > > > > > > > > let's use this
> > > > > > > > > > > PIP as a subsequent discussion of [1].
> > > > > > > > > > >
> > > > > > > > > > > BTW, there is a PR [2] in the pulsar-client-cpp repo 
> > > > > > > > > > > because the
> > > > > > > > > > > motivation is more meaningful for the C++ client.
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > I understand well this problem, we have it for the cited 
> > > > > > > > > > clients but I also
> > > > > > > > > > see the same issue for other libraries based on the Java 
> > > > > > > > > > client, like the
> > > > > > > > > > official Apache Pulsar Reactive client.
> > > > > > > > > >
> > > > > > > > > > I also see this problem in Startlight for JMS that is a JMS 
> > > > > > > > > > client for
> > > > > > > > > > Pulsar that is based on the Java client.
> > > > > > > > > >
> > > > > > > > > > While I agree on the problem and on the solution I think 
> > > > > > > > > > that a static
> > > > > > > > > > field is not enough, we have some problems:
> > > > > > > > > >
> > > > > > > > > > 1) there may be multiple usages of the Java client in the 
> > > > > > > > > > same JVM, and you
> > > > > > > > > > want each client to report correctly its version
> > > > > > > > > >
> > > > > > > > > > 2) we would need to use the Java security Manager in order 
> > > > > > > > > > to prevent
> > > > > > > > > > malicious code to modify the version or some other 
> > > > > > > > > > mechanism to prevent
> > > > > > > > > > overriding the version.
> > > > > > > > > >
> > > > > > > > > > I believe that in the case of the Java client is is easier 
> > > > > > > > > > to add a
> > > > > > > > > > configuration entry to the Pulsar Client Configuration. 
> > > > > > > > > > That would become a
> > > > > > > > > > field in the JavaClient. So each instance can declare its 
> > > > > > > > > > version and also
> > > > > > > > > > malicious code won't be able ti easily tweak the version 
> > > > > > > > > > (because it won't
> > > > > > > > > > be a simple static method call)
> > > > > > > > > >
> > > > > > > > > > Enrico
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > > [1] 
> > > > > > > > > > > https://lists.apache.org/thread/n59k537fhthjnzkfxtc2p4zk4l0cv3mp
> > > > > > > > > > > [2] https://github.com/apache/pulsar-client-cpp/pull/208
> > > > > > > > > > >
> > > > > > > > > > > Thanks!
> > > > > > > > > > >
> > > > > > > > > >

Reply via email to