Tuple waste size reported by memtx looks untrustworthy #10217
Description
Reproduced with Tarantool 3.2.0-entrypoint-113-ga0cd213bc715
Reproducer:
local json = require('json')
box.cfg{log_level = 'warn'}
box.schema.space.create('test')
box.space.test:create_index('pk')
local function print_stats(header)
print(header, json.encode(box.space.test:stat().tuple.memtx))
end
for i = 1, 100 do
box.space.test:insert({i, string.rep('x', 100 * 1024)})
end
print_stats('after insert')
for i = 1, 100 do
box.space.test:delete({i})
end
print_stats('after delete')
os.exit(0)Output:
after insert {"waste_size":4390012,"data_size":10240700,"header_size":1000,"field_map_size":0}
after delete {"waste_size":3981312,"data_size":0,"header_size":0,"field_map_size":0}
Note that after deleting all tuples from the space, the waste size is still reported to be around 4 MB while it should be 0. Also, 4 MB looks like a lot of waste for 10 MB data. Looks like something's broken with tuple waste accounting.