GH-43202: [C++][Compute] Detect and explicit error for offset overflow in row table#43226
GH-43202: [C++][Compute] Detect and explicit error for offset overflow in row table#43226pitrou merged 5 commits intoapache:mainfrom
Conversation
|
|
zanmato1984
added
the
Critical Fix
|
Hi @pitrou @westonpace , would you like to take a look? Thanks. |
github-actions
bot
added
awaiting committer review
|
@github-actions crossbow submit -g cpp |
This comment was marked as outdated.
This comment was marked as outdated.
|
@github-actions crossbow submit -g cpp |
|
Revision: ae34ac4 Submitted crossbow builds: ursacomputing/crossbow @ actions-72e104db5c
|
| { | ||
| row_encoder.PrepareEncodeSelected(0, num_rows - 1, columns); | ||
| ASSERT_OK(row_encoder.EncodeSelected(&row_table, static_cast<uint32_t>(num_rows - 1), | ||
| row_ids.data())); |
There was a problem hiding this comment.
Should we drop/clear the row_table after this to release a bit of memory?
There was a problem hiding this comment.
Thanks for looking!
Every time an encoder encodes into a row table, the row table gets effectively cleared. So the memory is only hold til the next EncodeSelected call. Clearing it here doesn't make much difference I guess(?) - the code in between is relatively trivial.
|
After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit cf5d7c7. There were no benchmark performance regressions. 🎉 The full Conbench report has more details. It also includes information about 3 possible false positives for unstable benchmarks that are known to sometimes produce them. |
2 participants
Rationale for this change
There are two places for the offset in a row table to possibly overflow: 1) Encoding columns into a row table; 2) Appending to a row table from another row table.
They are particularly bad because they can cause silent wrong result for the computation.
What changes are included in this PR?
Detect such overflows in the aforementioned places and report an explicit error when an overflow is detected.
Are these changes tested?
UT included.
Are there any user-facing changes?
User code that used to be fake passing could see an explicit error after this change.
This PR contains a "Critical Fix".