+1. At $dayjob we have a legacy feature that uses phoenix-pig, but I believe that usage is scheduled for deprecation soon and we can maintain it in our internal fork until then. Pig hasn't had a release in 6 years and last I heard doesn't support Hadoop 3; no reason to keep supporting it.
Geoffrey On Tue, Apr 18, 2023 at 1:37 AM Istvan Toth <st...@apache.org> wrote: > Hi! > > We've never had a connectors release, because of multiple unsolved > problems. > Some, like java 11/17 support are relatively straightforward and don't > really need discussion, but some are more impactful. > > I propose the following plan, which should give us a chance to have a > release in the foreseeable future: > Disclosure: at $dayjob, we only support the Spark and HBase connectors, and > those are the ones we can dedicate resources to. > > *- Drop the connectors for Phoenix 4.x* > 4.x is EOL, and it complicates the project structure, build time, etc. > We've never had a release for 4.x either. > > *- Drop the Kafka connector* > It has CVEs, and only works with an ancient Kafka version. > I have also seen zero developer or user interest in it. > If someone volunteers to update and maintain it, we can always add it back > later > > > *- Drop the Pig connector*This doesn't have critical problems, but I have > seen zero interest in it. > The shaded artifact doesn't use maven-shade-plugin, and I suspect that it > would have classpath conflict issues. > Fixing up the shading to be on par with the rest of the connectors would be > a non-trivial amount of work. > If someone volunteers to update and maintain it, we can always add it back > later. > > *- Re-shade the hive 3 connector for hbase-shaded* > Hbase in Hive 3 is very broken, we already need to replace the shipped > HBase jars anyway. > To avoid conflict with the included hbase jars, we want to avoid > duplicating them. > The solution is to omit the Hbase and Hadoop JARs from the shaded > connector, and change the relocations > to handle the binary incompatibilities between the shaded and non-shaded > HBase API. > We already do this for Spark, and this is also how the Hive 4 connector > will have to work. > (This already works well at $dayjob ) > > This would leave us with only three connectors, but those would at least be > released, and easier to support: > Spark 2 > Spark 3 > Hive 3 > > Please share your thoughts! > > Istvan >