YuweiXIAO
YuweiXIAO
I can reproduce the issue. The backtrace: ``` postgres=# copy test to '/tmp/test.parquet postgres'# '; COPY 1 ``` Appending a new line to the filename will trick the server to...
> Will this have local iceberg table and/or columnar storage support? The integration introduces columnar storage support to pg_duckdb (ps. it is not Iceberg tables). However, the upstream DuckLake plans...
> Hi @YuweiXiao! Interesting patch, to make this bit more usable, could data_path be configurable(=now hardcoded to /tmp/ducklake) Yeah sure. Plan to have a guc to control the path and...
> @YuweiXiao Are you still actively working on this patch? Sorry, no. Given that DuckLake is currently in active development, so gonna wait till a stable release of it. In...
+1 in my setup - driver 42.7.2 - pg version 17 (neon)
I found it is caused by Neon proxy, which drops the search_path start up params. Direct connection to pg should work.
the related of code for processing startup params https://github.com/neondatabase/neon/blob/main/proxy/src/compute/mod.rs#L216
This might be expected. The layout of the parquet file needs multiple random IOs when reading. Like, checking the magic number, file footer, row group meta, and pages in row...
maybe different schemas if you use Postgres as metadata backend, check out https://github.com/duckdb/ducklake/issues/193