Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[BUG] Failure at renumbering when passing empty dataframe #2217

Closed
jnke2016 opened this issue Apr 11, 2022 · 0 comments · Fixed by #2224
Closed

[BUG] Failure at renumbering when passing empty dataframe #2217

jnke2016 opened this issue Apr 11, 2022 · 0 comments · Fixed by #2224
Labels
? - Needs Triage Need team to review and classify bug Something isn't working
Milestone
22.06

Comments

@jnke2016
Copy link
Contributor

Some partitions have their column values set to <NA> after shuffling when passing empty dataframe as described below.

    major_vertices  minor_vertices value
0               3               5  <NA>
1               2               0  <NA>
2               2               1  <NA>
3               4               5  <NA>

There is also an OOM/hang reported when calling renumbering after shuffling. There is no hang/error when the datasets is unweighted

Steps/Code to reproduce bug
Build PR #2216 and run any MG algos with the following datasets on 4+ GPUs

"src" = [0, 1, 1, 2, 2, 2, 3, 4, 5]
"dst" = [1, 3, 4, 0, 1, 3, 5, 5, 6]
"value" = [0.1, 2.1, 1.1, 5.1, 3.1, 4.1, 7.2, 3.2, 1.2]
@jnke2016 jnke2016 added ? - Needs Triage Need team to review and classify bug Something isn't working labels Apr 11, 2022
@jnke2016 jnke2016 mentioned this issue Apr 11, 2022
Merged
@ChuckHastings ChuckHastings mentioned this issue Apr 13, 2022
Merged
@ChuckHastings ChuckHastings added this to the 22.06 milestone Apr 13, 2022
rapids-bot bot pushed a commit that referenced this issue Apr 14, 2022
@ChuckHastings
Address some MNMG issues in cython.cu (#2224)
6a8faa0 
Closes #2217 

The issue identifies some issues if the partitioning ends up with some empty partitions.  There were two issues here:
 1. The `call_shuffle` method determines whether a graph has weights by comparing the weight pointer to `nullptr`.  But if an MNMG partition is empty then the weight pointer will be `nullptr` even if the graph has weights.  Added a parameter to identify if the graph is weighted.
 2. The `renumber_helper` method was missing an else and was overwriting the correct value with an invalid value if the return from shuffling on a gpu had no entries.

These are corrected in the PR.

Authors:
  - Chuck Hastings (https://github.com/ChuckHastings)

Approvers:
  - Seunghwa Kang (https://github.com/seunghwak)
  - Joseph Nke (https://github.com/jnke2016)
  - Rick Ratzel (https://github.com/rlratzel)

URL: #2224
rapids-bot bot pushed a commit that referenced this issue Apr 20, 2022
@jnke2016
Enable MG support for small datasets (#2216)
ab5a1f7 
This PR enables MG support for very small datasets where the number of partitions is smaller than the number of workers

Dependent on issue #2217 to be resolved before merging
closes #2196

Authors:
  - Joseph Nke (https://github.com/jnke2016)

Approvers:
  - Rick Ratzel (https://github.com/rlratzel)

URL: #2216
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
? - Needs Triage Need team to review and classify bug Something isn't working
Projects
None yet
Milestone
22.06
Development

Successfully merging a pull request may close this issue.

Address some MNMG issues in cython.cu ChuckHastings/cugraph
2 participants
@ChuckHastings @jnke2016