:wave: I am seeing the following after upgrading t...
# general
g
šŸ‘‹ I am seeing the following after upgrading to
0.6.14
, looks like the
concat
at micropartition level not usage of
concat
by me right (I am not using it fwiw)? Will share query plan
Copy code
File "/tmp/ray/session_2026-01-08_20-44-16_741313_1/runtime_resources/pip/4858717c93c532512d70d85304fc18c2c24c6c55/virtualenv/lib/python3.12/site-packages/daft/recordbatch/micropartition.py", line 135, in concat
 return MicroPartition._from_pymicropartition(_PyMicroPartition.concat(micropartitions))
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
daft.exceptions.DaftCoreException: DaftError::SchemaMismatch MicroPartition concat requires all schemas to match, ╭──────────────────────┬────────╮
│ column_name ┆ type │
ā•žā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•Ŗā•ā•ā•ā•ā•ā•ā•ā•ā•”
│ __store_cluster_id_l ┆ String │
╰──────────────────────┓────────╯
 vs ╭──────────────────────┬────────╮
│ column_name ┆ type │
ā•žā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•ā•Ŗā•ā•ā•ā•ā•ā•ā•ā•ā•”
│ __store_cluster_id_r ┆ String │
╰──────────────────────┓────────╯
This is using Ray runner, works fine when I test locally with native runner
note: trying to get everything updated to
0.7.x
, maybe it was fixed their
s
Are you using the old Ray runner or the new one? The old Ray runner was removed from
0.7.x
onwards
g
I am using the new ray runner, which is default in 0.6.14, right?
s
Yes, as long as
use_legacy_ray_runner
is not set to true
g
got it, then yes I am using the new ray runner
s
do you mind sharing the output of
df.explain(show_all=True)
? knowing what operations were performed would help narrow down where the
MicroPartition.concat
is used
g
plan.txt
^^ is from my local testing
native == works, using local ray cluster == fail
šŸ‘ 1
to narrow down changeset, I am moving from
0.5.22
to
0.6.14
s
actually, do you mind generating the explain with the ray runner enabled? it will create a slightly different plan because repartitions are removed from the native runner (since they are no-ops)
g
plan_ray.txt
šŸ‘ 1
wait, this is using sort-merge. I changed to use hash join and it works
šŸ‘ 1
s
Ah ok, I believe the problem is coming from our implementation of sort-merge join
āœ… 1
Ok looks like we confirmed it
g
daft party that is an easy change on my side, thanks! actually wanted hash join, but had a default somewhere to sort-merge when I was testing something
šŸ‘ 1
s
Great, we'll look into the problem with sort-merge join. i believe the bug occurs when the left and right table column names are different, so a temporary fix would be to rename to the same name
g
took a stab at a fix, I am still rusty in rust, but I think this captures the gist of what you mention above https://github.com/Eventual-Inc/Daft/pull/6243
ā¤ļø 1