PyAirbyte icon indicating copy to clipboard operation
PyAirbyte copied to clipboard

Snowflake is slow

Open aaronsteers opened this issue 1 year ago • 1 comments

Relative to other cache types, Snowflake is pretty slow right now. It appears the main issue is that there are lots of calls which have a round-trip time of .8s-1.5s. These add up to runtimes close to 30-40 seconds to do a simple load.

In the context of a 10-20 minute data sync, these slowdowns may not be noticeable, but they are very noticeable in our integration tests.

For now, I've already moved all Snowflake integration tests into their own integration test file, separating them from the Postgres and DuckDB tests which are much faster.

I'm also attaching a perf trace log from viztracer, for discussion and future reference.

This test ran in ~10 seconds. You can see several calls to the database taking in the .8-1.9s range:

Image

result_test_faker_read_to_snowflake_170789807552213.json.zip

aaronsteers avatar Feb 14 '24 17:02 aaronsteers

Moving to backlog.

bindipankhudi avatar Apr 22 '24 17:04 bindipankhudi