jenkins test make check

Rados approved: https://tracker.ceph.com/projects/rados/wiki/MAIN#httpstrellocomcPB1LsGKm1970-wip-yuri8-testing-2024-03-11-1138-old-wip-yuri8-testing-2024-03-06-1329-old-wip-yuri8-testing-2024-03-04-1226-old-wip-yuri8-test

two questions, look for reply.

1. release_alloc_txc osr->q write op， will execute early than async discard。
2. newly allocate offset,len, will overlap with discard_queue. I can't found search and filter in discard_queue .
3. discard/trim nvme area[offset, len] will lost data finally.
   thanks.

two questions, look for reply.

1. release_alloc_txc osr->q write op， will execute early than async discard。
2. newly allocate offset,len, will overlap with discard_queue. I can't found search and filter in discard_queue .
3. discard/trim nvme area[offset, len] will lost data finally.
   thanks.

Please see

void BlueStore::_txc_release_alloc(TransContext *txc)
{
  bool discard_queued = false;
  // it's expected we're called with lazy_release_lock already taken!
  if (unlikely(cct->_conf->bluestore_debug_no_reuse_blocks ||
               txc->released.size() == 0 ||
               !alloc)) {
      goto out;
  }
  discard_queued = bdev->try_discard(txc->released);
  // if async discard succeeded, will do alloc->release when discard callback
  // else we should release here
  if (!discard_queued) {
      dout(10) << __func__ << "(sync) " << txc << " " << std::hex
               << txc->released << std::dec << dendl;
      alloc->release(txc->released);
  }

BlueStore doesn't release extents immediately if discards are enabled. Instead it postpones that until relevant discard op is completed
void BlueStore::handle_discard(interval_set<uint64_t>& to_release) { dout(10) << __func__ << dendl; ceph_assert(alloc); alloc->release(to_release); }
Hence there is no way to allocate an extent while it's being discarded.

two questions, look for reply.

1. release_alloc_txc osr->q write op， will execute early than async discard。
2. newly allocate offset,len, will overlap with discard_queue. I can't found search and filter in discard_queue .
3. discard/trim nvme area[offset, len] will lost data finally.
   thanks.

Please see

void BlueStore::_txc_release_alloc(TransContext *txc)
{
  bool discard_queued = false;
  // it's expected we're called with lazy_release_lock already taken!
  if (unlikely(cct->_conf->bluestore_debug_no_reuse_blocks ||
               txc->released.size() == 0 ||
               !alloc)) {
      goto out;
  }
  discard_queued = bdev->try_discard(txc->released);
  // if async discard succeeded, will do alloc->release when discard callback
  // else we should release here
  if (!discard_queued) {
      dout(10) << __func__ << "(sync) " << txc << " " << std::hex
               << txc->released << std::dec << dendl;
      alloc->release(txc->released);
  }

BlueStore doesn't release extents immediately if discards are enabled. Instead it postpones that until relevant discard op is completed void BlueStore::handle_discard(interval_set<uint64_t>& to_release) { dout(10) << __func__ << dendl; ceph_assert(alloc); alloc->release(to_release); } Hence there is no way to allocate an extent while it's being discarded.

thanks. I got it.
BlueFS and BlueStore reset the bitmap for the release_set only when discard_queued is false. Otherwise, the space is queued for discard and the bitmap reset is performed continuously by a background thread