[JNI] Adds HostColumnVector.EventHandler for spillability checks - #13898
Conversation
Signed-off-by: Alessandro Bellina <abellina@nvidia.com>
abellina
added
Spark
abellina
changed the title
|
Tests pass for me locally. I am seeing a CI failure related to a missing nvtx3 header, but I wasn't upmerged. Trying again: |
|
Failed again, cuDF jni isn't compiling in CI, but it runs fine locally :( It seems the |
| * @param cv - a reference to the HostColumnVector we are closing | ||
| * @param refCount - the updated ref count for this HostColumnVector at | ||
| * the time of invocation |
There was a problem hiding this comment.
| * @param cv - a reference to the HostColumnVector we are closing | |
| * @param refCount - the updated ref count for this HostColumnVector at | |
| * the time of invocation | |
| * @param cv Reference to the HostColumnVector we are closing | |
| * @param refCount The updated ref count for this HostColumnVector at | |
| * the time of invocation |
| * `onClosed` is invoked with the updated `refCount` during `close`. | ||
| * The last invocation of `onClosed` will be with `refCount=0`. | ||
| * | ||
| * @note the callback is invoked with this `HostColumnVector`'s lock held. |
There was a problem hiding this comment.
| * @note the callback is invoked with this `HostColumnVector`'s lock held. | |
| * @note The callback is invoked with this `HostColumnVector`'s lock held. |
There was a problem hiding this comment.
We tend to favor starting with lowercase and we also favor indenting the description so that it reads as one block, but I do think the hyphen is discouraged. How about a compromise? (a0944ee)
There was a problem hiding this comment.
It's not very important since we are not using the generated docds. But if we are the users who read the generated docs, we will see they are ugly with bad style: Sentences start without first letter being Cap.
There was a problem hiding this comment.
and sentences starting without cap. letter is not good IMO.
There was a problem hiding this comment.
Since our code styles on this are all over the place, I am going to suggest filing a follow on if you feel strongly about it. We should then in that issue fix all the comments and make sure that the pre commit hooks exercise them, so we have something consistent and easy to test.
…udf into add_host_column_event_handler
|
I am trying a suggestion from @robertmaynard here 6fb3f1b to get the conda environment setup for the java test run with cuda 12, which is the one that is failing because the nvtx header we need is in a different conda package now. |
revans2
left a comment
revans2
left a comment
There was a problem hiding this comment.
Looks good, just curious about the deps.
|
@robertmaynard that worked! thanks! Also I do need an approval from |
|
@raydouglass would you mind taking a look at this from the ops-codeowner perspective? The change that prompted this review request was: 6fb3f1b |
|
I'll wait for https://github.com/rapidsai/cudf/pull/13901/files#diff-5475a6d76de4c506ee92cf6f941bba30a6a07b5881cfea531d42a6ec035095a6 then re-test and rollback my change here likely. |
|
Removed |
|
/merge |
5 participants
This change adds the exact same API for event handling that was added to the device-side
ColumnVectorhere #13279. We are going to need this to makeHostColumnVector, or a batch of them, spillable in this spark-rapids feature: NVIDIA/cudf-spark#8882.