[IMPALA-11223] ASM files from different fragments conflict when using asm_module_dir

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 4.1.0
• Fix Version/s: Impala 4.3.0
• Component/s: Backend
• Epic Color: ghx-label-3
• Assignee: Quanlong Huang
• Reporter: Joe McDonnell
• Created: 04/Apr/22 19:34
• Updated: 14/Apr/23 23:08
• Resolved: 14/Apr/23 23:08

Description

For debugging codegen, it is useful to be able to inspect the generated assembly. The asm_module_dir startup parameter directs Impala to dump the codegen assembly to files in a directory. It currently dumps the assembly for a query into '${query_id}.asm', but there are actually multiple codegen operations going on in a single query and the output from one will overwrite the output from previous ones. I added a debug statement to the dumping code, and it gets called multiple times:

I0404 12:25:34.453413  6527 codegen-symbol-emitter.cc:58] 574df02d4b904ee9:9fa6181b00000001] Writing disassembly to: /data/Impala/logs/asm_module_dir/574df02d4b904ee9:9fa6181b00000000.asm
I0404 12:25:34.463084  6528 codegen-symbol-emitter.cc:58] 574df02d4b904ee9:9fa6181b00000000] Writing disassembly to: /data/Impala/logs/asm_module_dir/574df02d4b904ee9:9fa6181b00000000.asm
...
I0404 12:25:34.491320  6529 codegen-symbol-emitter.cc:58] 574df02d4b904ee9:9fa6181b00000004] Writing disassembly to: /data/Impala/logs/asm_module_dir/574df02d4b904ee9:9fa6181b00000000.asm

We should fix the filenames so that these collisions do not occur. One option would be to use the fragment id rather than the query id.