Do not publish to the org.postgresql package namespace
lukaseder opened this issue · 9 comments
I'm submitting a ...
- bug report
- feature request
Describe the issue
This driver is currently copying the official PostgreSQL JDBC driver and publishing it under the same package namespace under org.postgresql:
https://github.com/yugabyte/jdbc-yugabytedb/tree/master/jdbc-yugabytedb/src/main/java/org/postgresql
This has always been discouraged in the Java ecosystem for various reasons:
- You don't own that namespace. This is important for both legal (trademark) and security reasons. If anyone on the web would start publishing to this very popular namespace, whom to trust?
- You're altering the contents of that namespace for your own purposes, without pushing changes upstream. Now the behaviour of the namespace is undefined, depending on which implementation is loaded.
- You're creating split packages (starting with JDK 9 and the module system, this is not only discouraged, but forbidden).
When someone needs to include the official driver and yours in the same classpath, or worse, the same module path, then there will be namespacing conflicts between the two drivers. At the latest when pgjdbc decides to publish itself as a Java 9 module, then you won't be able to continue this practice.
Driver Version?
42.2.7-yb-6-SNAPSHOT
Java Version?
N/A
OS Version?
N/A
PostgreSQL Version?
N/A
To Reproduce
See: https://github.com/yugabyte/jdbc-yugabytedb/tree/master/jdbc-yugabytedb/src/main/java/org/postgresql
Expected behaviour
You have two acceptable options:
- Depend on the official driver and extend it via its extension points (less powerful, of course, if you need to patch stuff)
- Fork the official driver and re-publish it under your own namespace, e.g.
com.yugabyte.org.postgresql
The latter is called "shading" and there's a handy Maven plugin for that: https://maven.apache.org/plugins/maven-shade-plugin
Logs
N/A
+1 to not using the org.postgresql namespace. Shading your driver would be fine.
I think the pgjdbc project would be open to adding hooks to enable something like this. It seems that the need is real as this project is not the first to do it.
Hi @lukaseder and others,
Sorry for a delayed response here. Was caught up in few things.
Coming to the issue:
As you all know we just published the beta version of this driver sometime ago. It is a work in progress and any comments on that is very welcome at this stage, so first of all thanks for your inputs.
Making the changes on the fork of the official driver is definitely the way forward and was already in our plan. We were just trying to conduct some experiments on an existing copy of the source code, which was there for some historical reasons, but rest assured the next release will be from a fork and not a copy.
However we would really like to explore this ( below ) suggestion further:
Depend on the official driver and extend it via its extension points (less powerful, of course, if you need to patch stuff)
Just to be clear, I believe you are talking about new extension points (java SPI?) which can be added in the upstream jdbc driver for such purposes. Right?
We would like to understand the proposal better and will be happy to collaborate also on this.
'shading' may not be a preferred option for us as it might break compatibilities with other database tools.
Also, our plan is to extend the driver for a couple of more things specific to YugabyteDB, in future. And even they seem like generic requirements.
- Data aware query routing - The driver detects which server is the best server for any read/write op to go to rather than the default one.
- Automatic failover to a healthy server in case connectivity to one is lost for any reason.
It will be great to talk about these as well along with the in-built 'load-balancing' feature.
BTW the current implementation of load-balance is a very simple approach we have taken and it is almost like an extension approach. It will be great if you can find out sometime and we can discuss this.
Looking forward to your response on this.
I have to say that using the same URL and namespace as the pgjdbc driver is not making me happy. As @lukaseder pointed out there are legal and trademark issues here. Not to mention the mess this would make in an application.
As for extending the driver. We have discussed it, and would entertain proposals.
Just to be clear, I believe you are talking about new extension points (java SPI?) which can be added in the upstream jdbc driver for such purposes. Right?
Yes that's what I had in mind
'shading' may not be a preferred option for us as it might break compatibilities with other database tools.
I take such "breakages" any time over split packages in Maven Central, if you don't own the namespace:
- Few people depend on pgjdbc API directly, most use only JDBC API
- If pgjdbc API is needed, then it's easy to search/replace the package name in a project from
org.postgresqltocom.yugabyteor whatever - While package renames can be fixed easily in client code, split packages cannot, once you're on the module path. There really isn't a good solution to the problem you've created for users who have both drivers on the classpath/module path.
- Data aware query routing - The driver detects which server is the best server for any read/write op to go to rather than the default one.
JDBC 4.3 has sharding support, maybe you mean this? In any case, that's probably interesting to implement in a data source / connection pool, rather than the JDBC driver directly. This can definitely be implemented in any namespace, so I don't think an SPI is required.
- Automatic failover to a healthy server in case connectivity to one is lost for any reason.
Sounds again like a job for a data source / connection pool, rather than the JDBC driver directly.
Honestly I believe all of these things do not belong in the driver. There are a number of pools that implement read/write splitting of statements. There are a number of tools that implement failover. Using the driver seems to be the wrong tool IMO
Ok. We will be having the driver packaged under yugabyte namespace and we will use a different url. Thanks.
Ok. We will be having the driver packaged under yugabyte namespace and we will use a different url. Thanks.
Apparently the driver is being distributed using the same URL and namespace still.
No it is not. This repo clearly mentions in the README:
YugabyteDB JDBC Driver [Deprecated] - Please consider using the new YugabyteDB JDBC Driver from https://github.com/yugabyte/pgjdbc
We are releasing our new driver from this fork https://github.com/yugabyte/pgjdbc
Please have a look at the docs and the README there. Both the url and driver class has changed.
The url starts from jdbc:yugabytedb and Driver class is com.yugabyte.Driver
We will be completely removing it very soon.
Excellent, thanks!