• unknown's avatar
    BUG#31833 - ORDER BY leads to wrong result when ARCHIVE, BLOB and · 13ce2d77
    unknown authored
                table cache is full
    
    After reading last record from freshly opened archive table
    (e.g. after flush table, or if there is no room in table cache),
    the table is reported as crashed.
    
    The problem was that azio wrongly invalidated azio_stream when it
    meets EOF.
    
    
    mysql-test/r/archive.result:
      A test case for BUG#31833.
    mysql-test/t/archive.test:
      A test case for BUG#31833.
    storage/archive/azio.c:
      After azread() successfuly read and inflated data, it calls
      check_header() function. According to the comment it is done
      to detect concatenated .az files.
      
      When we read last record, there are no more bytes left at the
      current offset, all further my_read() calls will return 0. In
      this case check_header() wrongly sets s->z_err to Z_ERRNO,
      indicating that azio_stream is broken.
      
      Following is original condition from gzio:
      len = (uInt)fread(s->inbuf + len, 1, Z_BUFSIZE >> len, s->file);
      if (len == 0 && ferror(s->file)) s->z_err = Z_ERRNO;
      
      As fread() returns 0 on both EOF and error, the condition states:
      Invalidate gzio_stream if we got an error from last fread().
      
      Applied the same logic to azio.
      
      Note that a test case contains FLUSH TABLE t1 prior to SELECT. It is
      needed because azio doesn't flush buffers immediately. Thus we may
      azread() last record from in-memory buffer. When we read from
      in-memory buffer, EOF is detected by different branch of code in
      azread() and we never enter check_header() in this case.
    13ce2d77
azio.c 24.3 KB