3 ms·

Curious why pgduck_server is a totally separate process?

• dkdcio 11mo ago
  from the README: > This separation also avoids the threading and memory-safety limitations that would arise from embedding DuckDB directly inside the Postgres process, which is designed around process isolation rather than multi-threaded execution. Moreover, it lets us interact with the query engine directly by connecting to it using standard Postgres clients.
  • beoberha 11mo ago
    Thanks! Didn’t scroll down far enough
• rmnclmnt 11mo ago
  The README explains it: > This separation also avoids the threading and memory-safety limitations that would arise from embedding DuckDB directly inside the Postgres process, which is designed around process isolation rather than multi-threaded execution. Moreover, it lets us interact with the query engine directly by connecting to it using standard Postgres clients.
• pgguru 11mo ago
  What has been pointed out from the README; also: - Separation of concerns, since with a single external process we can share object store caches without complicated locking dances between multiple processes. - Memory limits are easier to reason about with a single external process. - Postgres backends end up being more robust, as you can restart the pgduck_server process separately.