Repository navigation
fix(gfql): resolve SIGSEGV with engine="cudf" on RAPIDS 25.02 (#977) - #1066
Merged
Merged
Conversation
lmeyerov
marked this pull request as ready for review
April 6, 2026 07:41
…t) with pandas bridge (#977) cudf Series.map(dict/Series) triggers numba JIT which SIGSEGVs on RAPIDS 25.02. Introduce safe_map_series() in Engine.py that bridges through pandas for cudf. Apply at all 7 affected call sites: hop.py (4), chain.py (1), df_executor.py (2), pipeline.py (1). Includes DGX smoke test + unit tests for safe_map_series + cudf guard regression tests. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
edge_map is only assigned inside the edge_hop_col branch; line 974 used it unconditionally which raised UnboundLocalError when edge_hop_col is None. Previously masked by SIGSEGV on 25.02. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
…p label_node_hops
…ba SIGSEGV on 25.02
…ated on 25.02+26.02 (#977)
… on lookup, cleaner cudf check
…ngine.py conventions
…ngine.py conventions
lmeyerov
force-pushed
the
feat/issue-977-cudf-sigsegv
branch
from
April 6, 2026 08:18
d41a24c to
527ace2
Compare
<<<<<<< HEAD line was left from incomplete conflict resolution during rebase; caused SyntaxError in CI test collection. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
Gap tests: - safe_map_series: cudf + pd.Series mapping path (Engine.py:399-400) - hop: missing_mask guard when edge_hop_col=None (hop.py:973) - hop: empty edge_map_df else branch (hop.py:961-965) DRY: extract _mk_abc_chain() helper in test_hop.py — 5 label tests refactored CHANGELOG: fix call-site count (8, not 7); add #977 Tests entry (23 tests); correct file list (no gfql_unified.py — that fix was in #1061) Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
New tests: - output_max_hops path (hop.py:834-891 fallback_map try/except + line-1011 post-filter) - label_seeds=True with implicit starting_nodes (hop.py:843-859 both-branches path) - numeric (int) node IDs through safe_map_series int→int hop labeling (hop.py:938) - undirected label_seeds=False clears seed hop (hop.py:987-1000) DRY: extract _mk_abcd_chain() helper — multi_hop_ordering + output_max_hops refactored Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
…hs (#977) New smoke tests cover: - safe_map_series + pd.Series mapping (Engine.py:399-400) - output_max_hops filtering (hop.py:834-891 fallback_map + line-1011 post-filter) - label_seeds=True seed rows from node_hop_records (hop.py:843-859) - label_seeds=False + undirected clears seed hop (hop.py:987-1000) Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
…umns to_pandas() on integer cudf Series SIGSEGVs on RAPIDS 25.02 (same numba path as the original #977 bug). String columns are safe. Switch all numeric hop value reads to to_arrow().to_pylist() to avoid the numba JIT trigger. Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #977.
Problem
graph.gfql(query, engine="cudf")crashes with SIGSEGV (returncode 139) on RAPIDS 25.02. Root cause traced to two distinct numba JIT trigger paths:Series.map(dict/Series)— triggers numba JIT on RAPIDS 25.02Series.to_pandas()on numerical columns — triggersnumba_cuda.as_cuda_array→safe_cuda_api_call→ SIGSEGV(A third path,
Series.apply(lambda), was fixed in #1061 viamarker.notna().)Root cause confirmed via Python
faulthandler:Fix
Introduce
safe_map_series(series, mapping)inEngine.pythat for cudf:to_arrow().to_pylist()(bypasses numba path).map()for pandasApplied at 8 call sites:
hop.py(4×),chain.py(1×),df_executor.py(2×),pipeline.py(1×).The
filter_by_dict.pycudf guard (lines 75–81) was already safe — converts to pandas before.apply().Validation
graphistry/test-rapids-official:25.02-cuda12-gfql): all smoke tests PASSgraphistry/test-rapids-official:26.02-cuda13-gfql): all smoke tests PASSTests added
test_safe_map_series.py— 16 tests: pandas/cudf paths for dict, pd.Series, cudf.Series, cudf+pd.Series, missing keys, empty mapping, non-default index, NaN in series, mixed value typestest_hop.py— 9 new label tests: reverse, undirected, node-only, edge-only, empty result, empty result with edge hops, multi-hop ordering, missing-mask guard (hop.py:973),_mk_abc_chain()helper DRYtest_chain.py— chain integration test forlabel_node_hopspropagation throughcombine_stepstest_lowering.py— label-filter regression, single-hop regression, multi-column ORDER BY testgraphistry/tests/compute/gfql/cypher/_dgx_977_smoke.py— DGX smoke script (pandas + cudf paths)