Tim Armstrong has posted comments on this change. Change subject: IMPALA-3317: fix crash in sorter when spilling zero-length strings ......................................................................
Patch Set 1: (6 comments) http://gerrit.cloudera.org:8080/#/c/2780/1/be/src/runtime/sorter.cc File be/src/runtime/sorter.cc: Line 477: } else { > how about: DCHECK_EQ(var_len_blocks_.back(), cur_var_len_block) to support Done Line 564: cur_sorted_var_len_block->Allocate<uint8_t>(total_var_len); > same dcheck Done Line 732: // offset in the lower 32 bits. > Add: see CopyVarLenDataConvertOffset() Done. I'm going to refactor some of this in a follow-up patch so that the encode/decode methods are adjacent to each other. Line 827: block_size_ > Let's also dcheck that the calculation to get block_offset fits in 32-bits Done http://gerrit.cloudera.org:8080/#/c/2780/1/testdata/workloads/functional-query/queries/QueryTest/large-sorts.test File testdata/workloads/functional-query/queries/QueryTest/large-sorts.test: Line 3: set mem_limit=512m; > Can you please keep these set statements together with the query ? The quer I was seeing some weird behaviour. There is actually a bug in the regex that implements the reset that I fixed. For some reason num_nodes doesn't seem to be taking effect when this test file is executed (maybe because exec_single_node_rows_threshold was set) so I changed it to set it in the test vector. http://gerrit.cloudera.org:8080/#/c/2780/1/tests/query_test/test_queries.py File tests/query_test/test_queries.py: Line 86 > why is this deleted? oops -- To view, visit http://gerrit.cloudera.org:8080/2780 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: I5f64593e94d367d6b6efb61a8b86e35516f18839 Gerrit-PatchSet: 1 Gerrit-Project: Impala Gerrit-Branch: cdh5-trunk Gerrit-Owner: Tim Armstrong <[email protected]> Gerrit-Reviewer: Dan Hecht <[email protected]> Gerrit-Reviewer: Matthew Jacobs <[email protected]> Gerrit-Reviewer: Michael Ho <[email protected]> Gerrit-Reviewer: Tim Armstrong <[email protected]> Gerrit-HasComments: Yes
