Near future

  • Clean up the binary-or-not decision. There should be possible to have partitions with different versions. That may mean lazy typeout generations.

  • Lazy value type cache, per-conn binary I/O decision.

Just thoughts

  • Drop plproxy.get_cluster_config()…

  • RUN ON ALL: sort by?

  • RUN ON ALL: ignore errors?

  • RUN ON ANY: if one con failed, try another

  • Streaming for big resultsets, to avoid loading them fully in memory. This needs also backend changes.

  • integrate with memcache: 

    set_object(id, data)
CACHE SET object(id) = data;
    get_object(id)
CACHE GET object(id)