From: Vladimir Davydov <vdavydov.dev@gmail.com>
To: kostja@tarantool.org
Cc: tarantool-patches@freelists.org
Subject: [PATCH v2 1/3] vinyl: make tx_manager_abort_writers_for_ddl more thorough
Date: Thu, 28 Mar 2019 18:52:27 +0300 [thread overview]
Message-ID: <1574fba56e436ce5129fc0a2853afb3f8c946279.1553787937.git.vdavydov.dev@gmail.com> (raw)
In-Reply-To: <cover.1553787937.git.vdavydov.dev@gmail.com>
In-Reply-To: <cover.1553787937.git.vdavydov.dev@gmail.com>
We need to abort all transactions writing to an altered space when
a new index is built. Currently, we use the write set to look up such
transactions, but it isn't quite correct, because a transaction could
yield on disk read before inserting a statement into the write set.
To address this problem, this patch adds vy_tx->last_lsm, which points
to the primary index of the space affected by the last prepared
transaction. Now, tx_manager_abort_writers_for_ddl will look not only
at the write set, but also at this variable to check if it needs to
abort a transaction.
Needed for #3420
---
src/box/vinyl.c | 4 +++-
src/box/vy_tx.c | 8 ++++++--
src/box/vy_tx.h | 15 ++++++++++++++-
3 files changed, 23 insertions(+), 4 deletions(-)
diff --git a/src/box/vinyl.c b/src/box/vinyl.c
index 3ef43e18..24f900cb 100644
--- a/src/box/vinyl.c
+++ b/src/box/vinyl.c
@@ -2433,8 +2433,10 @@ vinyl_engine_begin_statement(struct engine *engine, struct txn *txn)
(void)engine;
struct vy_tx *tx = txn->engine_tx;
struct txn_stmt *stmt = txn_current_stmt(txn);
+ struct index *pk = space_index(stmt->space, 0);
assert(tx != NULL);
- return vy_tx_begin_statement(tx, &stmt->engine_savepoint);
+ return vy_tx_begin_statement(tx, pk != NULL ? vy_lsm(pk) : NULL,
+ &stmt->engine_savepoint);
}
static void
diff --git a/src/box/vy_tx.c b/src/box/vy_tx.c
index 1b8224f4..889e094e 100644
--- a/src/box/vy_tx.c
+++ b/src/box/vy_tx.c
@@ -322,6 +322,7 @@ vy_tx_read_set_free_cb(vy_tx_read_set_t *read_set,
void
vy_tx_create(struct tx_manager *xm, struct vy_tx *tx)
{
+ tx->last_lsm = NULL;
stailq_create(&tx->log);
write_set_new(&tx->write_set);
tx->write_set_version = 0;
@@ -869,13 +870,14 @@ vy_tx_rollback(struct vy_tx *tx)
}
int
-vy_tx_begin_statement(struct vy_tx *tx, void **savepoint)
+vy_tx_begin_statement(struct vy_tx *tx, struct vy_lsm *lsm, void **savepoint)
{
if (tx->state == VINYL_TX_ABORT) {
diag_set(ClientError, ER_TRANSACTION_CONFLICT);
return -1;
}
assert(tx->state == VINYL_TX_READY);
+ tx->last_lsm = lsm;
if (stailq_empty(&tx->log))
rlist_add_entry(&tx->xm->writers, tx, in_writers);
*savepoint = stailq_last(&tx->log);
@@ -1112,7 +1114,9 @@ tx_manager_abort_writers_for_ddl(struct tx_manager *xm, struct vy_lsm *lsm)
{
struct vy_tx *tx;
rlist_foreach_entry(tx, &xm->writers, in_writers) {
- if (tx->state == VINYL_TX_READY &&
+ if (tx->state != VINYL_TX_READY)
+ continue;
+ if (tx->last_lsm == lsm ||
write_set_search_key(&tx->write_set, lsm,
lsm->env->empty_key) != NULL)
vy_tx_abort(tx);
diff --git a/src/box/vy_tx.h b/src/box/vy_tx.h
index aaa31bee..1767b509 100644
--- a/src/box/vy_tx.h
+++ b/src/box/vy_tx.h
@@ -140,6 +140,16 @@ struct vy_tx {
/** Transaction manager. */
struct tx_manager *xm;
/**
+ * Pointer to the primary index LSM tree of the space
+ * affected by the last prepared statement.
+ *
+ * We need it so that we can abort a transaction on DDL
+ * even if it hasn't inserted anything into the write set
+ * yet (e.g. yielded on unique check) and therefore would
+ * otherwise be ignored by tx_manager_abort_writers_for_ddl.
+ */
+ struct vy_lsm *last_lsm;
+ /**
* In memory transaction log. Contains both reads
* and writes.
*/
@@ -325,9 +335,12 @@ vy_tx_rollback(struct vy_tx *tx);
* Return the save point corresponding to the current
* transaction state. The transaction can be rolled back
* to a save point with vy_tx_rollback_statement().
+ *
+ * @lsm is supposed to point to the LSM tree corresponding
+ * to the primary index of the affected space.
*/
int
-vy_tx_begin_statement(struct vy_tx *tx, void **savepoint);
+vy_tx_begin_statement(struct vy_tx *tx, struct vy_lsm *lsm, void **savepoint);
/**
* Rollback a transaction statement.
--
2.11.0
next prev parent reply other threads:[~2019-03-28 15:52 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-03-28 15:52 [PATCH v2 0/3] Fix DML vs DDL race Vladimir Davydov
2019-03-28 15:52 ` Vladimir Davydov [this message]
2019-03-28 16:03 ` [PATCH v2 1/3] vinyl: make tx_manager_abort_writers_for_ddl more thorough Vladimir Davydov
2019-03-28 17:28 ` [tarantool-patches] " Konstantin Osipov
2019-03-28 15:52 ` [PATCH v2 2/3] vinyl: abort affected transactions when space is removed from cache Vladimir Davydov
2019-03-28 15:52 ` [PATCH v2 3/3] Revert "test: skip ddl test for vinyl on travis" Vladimir Davydov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1574fba56e436ce5129fc0a2853afb3f8c946279.1553787937.git.vdavydov.dev@gmail.com \
--to=vdavydov.dev@gmail.com \
--cc=kostja@tarantool.org \
--cc=tarantool-patches@freelists.org \
--subject='Re: [PATCH v2 1/3] vinyl: make tx_manager_abort_writers_for_ddl more thorough' \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox