[ https://issues.apache.org/jira/browse/PARQUET-1995?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17298527#comment-17298527 ]
Tham edited comment on PARQUET-1995 at 3/10/21, 6:42 AM: --------------------------------------------------------- [~emkornfield] the arrow version I'm using is 0.17.1. Sorry, I don't have the data when crash occured but I can share the scheme overall: * there are 152 columns: first column is double and others are float. * float columns may contain null value alternatively. * one row group has ~7680 rows. When our application crashed, it tried to close the parquet file, and I found an error message printed in some log file: {code:java} [2021-03-05T15:03:49.166-04:00][WARNING][70000e71a000] parquet::arrow::FileWriter::Close(): "Column 18 had 7680 while previous column had 0" {code} Any idea? was (Author: thamha): [~emkornfield] the arrow version I'm using is 0.17.1. Sorry, I don't have the data when crash occured but I can share the scheme overall: * there are 152 columns: some are double and some are float. * some columns contain null value alternatively. When our application crashed, it tried to close the parquet file, and I found an error message printed in some log file: {code:java} [2021-03-05T15:03:49.166-04:00][WARNING][70000e71a000] parquet::arrow::FileWriter::Close(): "Column 18 had 7680 while previous column had 0" {code} Any idea? > [C++][Parquet] Crash at parquet::TypedColumnWriterImpl<>::WriteBatchSpaced > -------------------------------------------------------------------------- > > Key: PARQUET-1995 > URL: https://issues.apache.org/jira/browse/PARQUET-1995 > Project: Parquet > Issue Type: Bug > Reporter: Tham > Priority: Major > Attachments: parquet_crash_WriteBatchSpaced.txt > > > I got this crash from our customer machine: > [^parquet_crash_WriteBatchSpaced.txt] , it crashes continuously every time my > application opens a parquet file and write the first row group. It happens > only on one machine and runs fine on others until now. I cannot reproduce it. > I access the parquet file on only one thread, so you can eliminate the case > of race condition. Do you have any idea about this crash report? -- This message was sent by Atlassian Jira (v8.3.4#803005)