ISAR/libmdbx
3
0
Fork 0
mirror of https://github.com/isar/libmdbx.git synced 2025-01-02 04:24:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

mdbx: remove obsolete current retired-list record from gc before reserve/backlogging free-pages to reduce looping.

Browse Source
This commit is contained in:
Leonid Yuriev 2021-12-25 16:20:48 +03:00
parent b30ccbde9e
commit 2ba90e63b1
1 changed files with 30 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
src/core.c
View File

@ -8679,7 +8679,7 @@ static __always_inline unsigned backlog_size(MDBX_txn *txn) {
* while reclaiming is prohibited. It should be enough to prevent search
* in mdbx_page_alloc() during a deleting, when GC tree is unbalanced. */
static int mdbx_prep_backlog(MDBX_txn *txn, MDBX_cursor *gc_cursor,
const size_t pnl_bytes) {
const size_t pnl_bytes, unsigned *retired_stored) {
const unsigned linear4list = number_of_ovpages(txn->mt_env, pnl_bytes);
const unsigned backlog4cow = txn->mt_dbs[FREE_DBI].md_depth;
const unsigned backlog4rebalance = backlog4cow + 1;
@ -8694,19 +8694,34 @@ static int mdbx_prep_backlog(MDBX_txn *txn, MDBX_cursor *gc_cursor,
pnl_bytes, backlog_size(txn), linear4list, backlog4cow,
backlog4rebalance);
MDBX_val fake_key, fake_val;
fake_key.iov_base = fake_val.iov_base = nullptr;
fake_key.iov_len = sizeof(txnid_t);
fake_val.iov_len = pnl_bytes;
int err = mdbx_cursor_spill(gc_cursor, &fake_key, &fake_val);
if (unlikely(err != MDBX_SUCCESS))
return err;
MDBX_val gc_key, fake_val;
int err;
if (linear4list < 2) {
gc_key.iov_base = fake_val.iov_base = nullptr;
gc_key.iov_len = sizeof(txnid_t);
fake_val.iov_len = pnl_bytes;
err = mdbx_cursor_spill(gc_cursor, &gc_key, &fake_val);
if (unlikely(err != MDBX_SUCCESS))
return err;
}
gc_cursor->mc_flags &= ~C_RECLAIMING;
err = mdbx_cursor_touch(gc_cursor);
mdbx_trace("== after-touch, backlog %u, err %d", backlog_size(txn), err);
if (linear4list > 1 && err == MDBX_SUCCESS) {
if (retired_stored) {
gc_key.iov_base = &txn->mt_txnid;
gc_key.iov_len = sizeof(txn->mt_txnid);
const struct cursor_set_result csr =
mdbx_cursor_set(gc_cursor, &gc_key, &fake_val, MDBX_SET);
if (csr.err == MDBX_SUCCESS && csr.exact) {
*retired_stored = 0;
err = mdbx_cursor_del(gc_cursor, 0);
mdbx_trace("== clear-4linear, backlog %u, err %d", backlog_size(txn),
err);
}
}
err = mdbx_page_alloc(gc_cursor, linear4list,
MDBX_ALLOC_GC | MDBX_ALLOC_CACHE | MDBX_ALLOC_SLOT)
.err;
@ -8768,9 +8783,10 @@ retry_noaccount:
goto bailout;
}
if (retired_stored < MDBX_PNL_SIZE(txn->tw.retired_pages)) {
if (retired_stored != MDBX_PNL_SIZE(txn->tw.retired_pages)) {
rc = mdbx_prep_backlog(txn, &couple.outer,
MDBX_PNL_SIZEOF(txn->tw.retired_pages));
MDBX_PNL_SIZEOF(txn->tw.retired_pages),
&retired_stored);
if (unlikely(rc != MDBX_SUCCESS))
goto bailout;
}
@ -8807,7 +8823,7 @@ retry_noaccount:
continue;
if (unlikely(rc != MDBX_SUCCESS))
goto bailout;
rc = mdbx_prep_backlog(txn, &couple.outer, 0);
rc = mdbx_prep_backlog(txn, &couple.outer, 0, nullptr);
if (unlikely(rc != MDBX_SUCCESS))
goto bailout;
mdbx_tassert(txn,
@ -8848,7 +8864,7 @@ retry_noaccount:
if (cleaned_gc_id > txn->tw.last_reclaimed)
break;
if (cleaned_gc_id < txn->tw.last_reclaimed) {
rc = mdbx_prep_backlog(txn, &couple.outer, 0);
rc = mdbx_prep_backlog(txn, &couple.outer, 0, nullptr);
if (unlikely(rc != MDBX_SUCCESS))
goto bailout;
}
@ -8974,7 +8990,7 @@ retry_noaccount:
key.iov_base = &txn->mt_txnid;
do {
data.iov_len = MDBX_PNL_SIZEOF(txn->tw.retired_pages);
mdbx_prep_backlog(txn, &couple.outer, data.iov_len);
mdbx_prep_backlog(txn, &couple.outer, data.iov_len, &retired_stored);
rc = mdbx_cursor_put(&couple.outer, &key, &data, MDBX_RESERVE);
if (unlikely(rc != MDBX_SUCCESS))
goto bailout;
@ -9286,7 +9302,7 @@ retry_noaccount:
data.iov_len = (chunk + 1) * sizeof(pgno_t);
mdbx_trace("%s: reserve %u [%u...%u) @%" PRIaTXN, dbg_prefix_mode, chunk,
settled + 1, settled + chunk + 1, reservation_gc_id);
mdbx_prep_backlog(txn, &couple.outer, data.iov_len);
mdbx_prep_backlog(txn, &couple.outer, data.iov_len, nullptr);
rc = mdbx_cursor_put(&couple.outer, &key, &data,
MDBX_RESERVE | MDBX_NOOVERWRITE);
mdbx_tassert(txn,