Working with Memory and Performance

By default xlsxwriter holds all cell data in memory. This is to allow future features where formatting is applied separately from the data.

The effect of this is that for large files xlsxwriter can consume a lot of memory and it is even possible to run out of memory.

Fortunately, this memory usage can be reduced almost completely by setting the Workbook:new() 'constant_memory' property:

workbook = Workbook:new(filename, {constant_memory = true})

The optimisation works by flushing each row after a subsequent row is written. In this way the largest amount of data held in memory for a worksheet is the amount of memory required to hold a single row of data.

Since each new row flushes the previous row, data must be written in sequential row order when 'constant_memory' mode is on:

-- With 'constant_memory' you must write data in row column order.
for row = 0, row_max do
  for col = 0, col_max do
    worksheet:write(row, col, some_data)
  end
end

-- With 'constant_memory' the following would only write the first column.
for col = 0, col_max do  -- !!
  for row = 0, row_max do
    worksheet:write(row, col, some_data)
  end
end

Another optimisation that is used to reduce memory usage is that cell strings aren’t stored in an Excel structure call “shared strings” and instead are written “in-line”. This is a documented Excel feature that is supported by most spreadsheet applications. One known exception is Apple Numbers for Mac where the string data isn’t displayed.

The trade-off when using 'constant_memory' mode is that you won’t be able to take advantage of any features that manipulate cell data after it is written. Currently there aren’t any such features.

For larger files 'constant_memory' mode also gives an increase in execution speed, see below.