Install cuDF-py against python 3.10 on Databricks #11477
Conversation
Fix on Databricks runtime for : NVIDIA#11394 Enable the udf_cudf_test test case for Databricks-13.3 Rapids 24.10+ drops python 3.9 or below conda packages. ref: https://docs.rapids.ai/notices/rsn0040/ Install cuDF-py packages against python 3.10 and above on Databricks runtime to run UDF cuDF tests, because on DB-13.3 Conda is not installed by default. Signed-off-by: timl <timl@nvidia.com>
NvTimLiu
requested review from
revans2,
tgravescs,
pxLi,
yinqingh and
YanxuanLiu
NvTimLiu
requested review from
jlowe,
GaryShen2008 and
gerashegalov
as code owners
jenkins/databricks/init_cudf_udf.sh
Outdated
| @@ -61,9 +47,41 @@ REQUIRED_PACKAGES=( | |||
| requests | |||
| sre_yield | |||
| ) | |||
| if base=$(conda info --base); then | |||
There was a problem hiding this comment.
Support install cudf-py packages via either Conda or PIP.
There was a problem hiding this comment.
Q: if no conda in PATH this would error out? use detect existence + var assignment in the same expression is not recommended here
can we try to use the the command to check if exist first? this would help make the if expression and else branches more clear
if command -v conda >/dev/null 2>&1; then
There was a problem hiding this comment.
Does any of Databricks Runtime 13.3+ have conda? If no, I think we can drop supporting conda installation.
There was a problem hiding this comment.
Q: if no conda in PATH this would error out?
The if expression will not error out but get 'false' value instead.
can we try to use the the
commandto check if exist first? this would help make the if expression and else branches more clearif command -v conda >/dev/null 2>&1; then
Sounds good to me, let me update it to make the if/else more readable
There was a problem hiding this comment.
Does any of Databricks Runtime 13.3+ have conda?
DB-13.3+ do not install conda by default
If no, I think we can drop supporting conda installation.
I'd like to keep conda installation there for some time, because:
Keep it to maintain compatibility with DB13.3 or earlier conda installations.
We also have the jenkins/databricks/cudf_udf_test.sh script, which is still compatible with conda installations.
Signed-off-by: timl <timl@nvidia.com>
|
build |
|
The pre-merge CI can not cover the udf-cudf python tests, Verified the PR locally, both the init script and udf-cudf tests got PASS:
|
|
failed unrelated cases: test_broadcast_join_with_conditionals in scala213 could be related to latest cudf change |
|
build |
Fix on Databricks runtime for : #11394
Enable the udf_cudf_test test case for Databricks-13.3
Rapids 24.10+ drops python 3.9 or below conda packages. ref: https://docs.rapids.ai/notices/rsn0040/
Install cuDF-py packages against python 3.10 and above on Databricks runtime to run UDF cuDF tests, because on DB-13.3 Conda is not installed by default.