Large TRIM size and high iodepth causes cause OOM during replay #2056
Description
When there is combination of high iodepth and TRIM large range, the memory footprint goes sky high and then OOM.
When I replay a trace, there is a TRIM command inside with 4GB range. It can replay when iodepth=1 but it will go OOM when iodepth is larger.
read_blktrace() -> queue_trace() -> handle_trace_discard()
static bool handle_trace_discard() {
if (t->bytes > bs[DDIR_TRIM])
// bs[DDIR_TRIM] = 4GB now
bs[DDIR_TRIM] = t->bytes;
and then propagate to td->o.max_bs
bool read_blktrace(struct thread_data* td) {
td->o.max_bs[DDIR_TRIM] = max(td->o.max_bs[DDIR_TRIM], rw_bs[DDIR_TRIM]);
and then it will allocate 4GB * iodepth buffers
int init_io_u_buffers(struct thread_data *td)
{
max_units = td->o.iodepth;
max_bs = td_max_bs(td);
// 4GB * iodepth => 4GB * 1024
td->orig_buffer_size = (unsigned long long) max_bs
* (unsigned long long) max_units;
Should we modify td_max_bs() to remove DDIR_TRIM ?
static inline unsigned long long td_max_bs(struct thread_data *td)
{
unsigned long long max_bs;
max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
// just return max_bs here
return max(td->o.max_bs[DDIR_TRIM], max_bs);
}