* [Tarantool-patches] [PATCH 0/2] Fix transaction manager
@ 2020-10-22 16:02 Aleksandr Lyapunov
2020-10-22 16:02 ` [Tarantool-patches] [PATCH 1/2] memtx: fix a bug in unlinking story lists Aleksandr Lyapunov
` (2 more replies)
0 siblings, 3 replies; 4+ messages in thread
From: Aleksandr Lyapunov @ 2020-10-22 16:02 UTC (permalink / raw)
To: tarantool-patches, Nikita Pettik, Alexander V . Tikhonov
There was two issues:
* https://github.com/tarantool/tarantool/issues/5423
* tx_man test faild on OSX builds
Both problems are fixed now.
Branch: alyapunov/gh-5423-dirty-tuple-unref
Aleksandr Lyapunov (2):
memtx: fix a bug in unlinking story lists
memtx: fix a bug in TX that caused deletion of a durty tuple
src/box/memtx_space.c | 1 +
src/box/memtx_tx.c | 67 +++++++++++++++++++++++++++++++++++++++---------
test/box/tx_man.result | 43 +++++++++++++++++++++++++++++++
test/box/tx_man.test.lua | 16 ++++++++++++
4 files changed, 115 insertions(+), 12 deletions(-)
--
2.7.4
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Tarantool-patches] [PATCH 1/2] memtx: fix a bug in unlinking story lists
2020-10-22 16:02 [Tarantool-patches] [PATCH 0/2] Fix transaction manager Aleksandr Lyapunov
@ 2020-10-22 16:02 ` Aleksandr Lyapunov
2020-10-22 16:02 ` [Tarantool-patches] [PATCH 2/2] memtx: fix a bug in TX that caused deletion of a durty tuple Aleksandr Lyapunov
2020-10-22 16:42 ` [Tarantool-patches] [PATCH 0/2] Fix transaction manager Alexander V. Tikhonov
2 siblings, 0 replies; 4+ messages in thread
From: Aleksandr Lyapunov @ 2020-10-22 16:02 UTC (permalink / raw)
To: tarantool-patches, Nikita Pettik, Alexander V . Tikhonov
Tx stories must be linked into correct double-linked list.
Preserve it.
Part of #5423
---
src/box/memtx_tx.c | 13 ++++++++++++-
1 file changed, 12 insertions(+), 1 deletion(-)
diff --git a/src/box/memtx_tx.c b/src/box/memtx_tx.c
index 55748ad..e378054 100644
--- a/src/box/memtx_tx.c
+++ b/src/box/memtx_tx.c
@@ -433,9 +433,20 @@ memtx_tx_story_gc_step()
}
assert(story->tuple == unused);
}
- memtx_tx_story_unlink(story, i);
+
+ if (link->older.is_story) {
+ struct memtx_story *older_story = link->older.story;
+ memtx_tx_story_unlink(story, i);
+ older_story->link[i].newer_story = older_story;
+ } else {
+ memtx_tx_story_unlink(story, i);
+ }
} else {
+ /* Just unlink from list */
link->newer_story->link[i].older = link->older;
+ if (link->older.is_story)
+ link->older.story->link[i].newer_story =
+ link->newer_story;
link->older.is_story = false;
link->older.story = NULL;
link->newer_story = NULL;
--
2.7.4
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Tarantool-patches] [PATCH 2/2] memtx: fix a bug in TX that caused deletion of a durty tuple
2020-10-22 16:02 [Tarantool-patches] [PATCH 0/2] Fix transaction manager Aleksandr Lyapunov
2020-10-22 16:02 ` [Tarantool-patches] [PATCH 1/2] memtx: fix a bug in unlinking story lists Aleksandr Lyapunov
@ 2020-10-22 16:02 ` Aleksandr Lyapunov
2020-10-22 16:42 ` [Tarantool-patches] [PATCH 0/2] Fix transaction manager Alexander V. Tikhonov
2 siblings, 0 replies; 4+ messages in thread
From: Aleksandr Lyapunov @ 2020-10-22 16:02 UTC (permalink / raw)
To: tarantool-patches, Nikita Pettik, Alexander V . Tikhonov
There was a mess in tuple refernce in TX history.
Now it was remade in the following asumptions:
* a clean tuple belongs to space, and the space implicitly holds
a reference to the tuple.
* a dirty tuple belongs to TX manager and a reference is held
in the corresponding story.
Closes #5423
---
src/box/memtx_space.c | 1 +
src/box/memtx_tx.c | 54 ++++++++++++++++++++++++++++++++++++++----------
test/box/tx_man.result | 43 ++++++++++++++++++++++++++++++++++++++
test/box/tx_man.test.lua | 16 ++++++++++++++
4 files changed, 103 insertions(+), 11 deletions(-)
diff --git a/src/box/memtx_space.c b/src/box/memtx_space.c
index d4b18d9..2a43e64 100644
--- a/src/box/memtx_space.c
+++ b/src/box/memtx_space.c
@@ -78,6 +78,7 @@ void
memtx_space_update_bsize(struct space *space, struct tuple *old_tuple,
struct tuple *new_tuple)
{
+ assert(space->vtab->destroy == &memtx_space_destroy);
struct memtx_space *memtx_space = (struct memtx_space *)space;
ssize_t old_bsize = old_tuple ? box_tuple_bsize(old_tuple) : 0;
ssize_t new_bsize = new_tuple ? box_tuple_bsize(new_tuple) : 0;
diff --git a/src/box/memtx_tx.c b/src/box/memtx_tx.c
index e378054..bcd4a37 100644
--- a/src/box/memtx_tx.c
+++ b/src/box/memtx_tx.c
@@ -418,8 +418,8 @@ memtx_tx_story_gc_step()
if (link->newer_story == NULL) {
/*
* We are at the top of the chain. That means
- * that story->tuple is in index. If the story is
- * actually delete the tuple, it must be deleted from
+ * that story->tuple is in index. If the story
+ * actually deletes the tuple, it must be deleted from
* index.
*/
if (story->del_psn > 0 && story->space != NULL) {
@@ -432,6 +432,12 @@ memtx_tx_story_gc_step()
panic("failed to rollback change");
}
assert(story->tuple == unused);
+ } else if (i == 0) {
+ /*
+ * The tuple is left clean in the space.
+ * It now belongs to the space and must be referenced.
+ */
+ tuple_ref(story->tuple);
}
if (link->older.is_story) {
@@ -655,6 +661,14 @@ memtx_tx_history_add_stmt(struct txn_stmt *stmt, struct tuple *old_tuple,
&replaced) != 0)
goto fail;
memtx_tx_story_link_tuple(add_story, replaced, i);
+ if (i == 0 && replaced != NULL && !replaced->is_dirty) {
+ /*
+ * The tuple was clean and thus belonged to
+ * the space. Now tx manager takes ownership
+ * of it.
+ */
+ tuple_unref(replaced);
+ }
add_story_linked++;
struct tuple *visible_replaced = NULL;
@@ -729,15 +743,16 @@ memtx_tx_history_add_stmt(struct txn_stmt *stmt, struct tuple *old_tuple,
collected_conflicts = collected_conflicts->next;
}
- /*
- * We now reference both new and old tuple because the stmt holds
- * pointers to them.
- */
- if (stmt->new_tuple != NULL)
- tuple_ref(stmt->new_tuple);
*result = old_tuple;
- if (*result != NULL)
+ if (*result != NULL) {
+ /*
+ * The result must be a referenced pointer. The caller must
+ * unreference it by itself.
+ * Actually now it goes only to stmt->old_tuple, and
+ * stmt->old_tuple is unreferenced when stmt is destroyed.
+ */
tuple_ref(*result);
+ }
return 0;
fail:
@@ -756,6 +771,10 @@ memtx_tx_history_add_stmt(struct txn_stmt *stmt, struct tuple *old_tuple,
unreachable();
panic("failed to rollback change");
}
+ if (i == 0 && was != NULL && !was->is_dirty) {
+ /* Just rollback previous tuple_unref. */
+ tuple_ref(was);
+ }
memtx_tx_story_unlink(stmt->add_story, i);
@@ -787,6 +806,10 @@ memtx_tx_history_rollback_stmt(struct txn_stmt *stmt)
for (uint32_t i = 0; i < story->index_count; i++) {
struct memtx_story_link *link = &story->link[i];
if (link->newer_story == NULL) {
+ /*
+ * We are at top of story list and thus
+ * story->tuple is in index directly.
+ */
struct tuple *unused;
struct index *index = stmt->space->index[i];
struct tuple *was = memtx_tx_story_older_tuple(link);
@@ -796,6 +819,17 @@ memtx_tx_history_rollback_stmt(struct txn_stmt *stmt)
unreachable();
panic("failed to rollback change");
}
+ if (i == 0 && was != NULL &&
+ !link->older.is_story) {
+ /*
+ * That was the last story in history.
+ * The last tuple now belongs to space
+ * and the space must hold a reference
+ * to it.
+ */
+ tuple_ref(was);
+ }
+
} else {
struct memtx_story *newer = link->newer_story;
assert(newer->link[i].older.is_story);
@@ -813,7 +847,6 @@ memtx_tx_history_rollback_stmt(struct txn_stmt *stmt)
}
stmt->add_story->add_stmt = NULL;
stmt->add_story = NULL;
- tuple_unref(stmt->new_tuple);
}
if (stmt->del_story != NULL) {
@@ -987,7 +1020,6 @@ memtx_tx_history_commit_stmt(struct txn_stmt *stmt)
assert(stmt->del_story->del_stmt == stmt);
assert(stmt->next_in_del_list == NULL);
res -= stmt->del_story->tuple->bsize;
- tuple_unref(stmt->del_story->tuple);
stmt->del_story->del_stmt = NULL;
stmt->del_story = NULL;
}
diff --git a/test/box/tx_man.result b/test/box/tx_man.result
index 3a6375d..a294165 100644
--- a/test/box/tx_man.result
+++ b/test/box/tx_man.result
@@ -826,6 +826,49 @@ s:select{}
| - - [1, 3]
| ...
+s:drop()
+ | ---
+ | ...
+s2:drop()
+ | ---
+ | ...
+
+-- https://github.com/tarantool/tarantool/issues/5423
+s = box.schema.space.create('test')
+ | ---
+ | ...
+i1 = s:create_index('pk', {parts={{1, 'uint'}}})
+ | ---
+ | ...
+i2 = s:create_index('sec', {parts={{2, 'uint'}}})
+ | ---
+ | ...
+
+s:replace{1, 0}
+ | ---
+ | - [1, 0]
+ | ...
+s:delete{1}
+ | ---
+ | - [1, 0]
+ | ...
+collectgarbage()
+ | ---
+ | - 0
+ | ...
+s:replace{1, 1}
+ | ---
+ | - [1, 1]
+ | ...
+s:replace{1, 2 }
+ | ---
+ | - [1, 2]
+ | ...
+
+s:drop()
+ | ---
+ | ...
+
test_run:cmd("switch default")
| ---
| - true
diff --git a/test/box/tx_man.test.lua b/test/box/tx_man.test.lua
index 60fa52b..53b0773 100644
--- a/test/box/tx_man.test.lua
+++ b/test/box/tx_man.test.lua
@@ -236,6 +236,22 @@ s:select{}
tx3:commit()
s:select{}
+s:drop()
+s2:drop()
+
+-- https://github.com/tarantool/tarantool/issues/5423
+s = box.schema.space.create('test')
+i1 = s:create_index('pk', {parts={{1, 'uint'}}})
+i2 = s:create_index('sec', {parts={{2, 'uint'}}})
+
+s:replace{1, 0}
+s:delete{1}
+collectgarbage()
+s:replace{1, 1}
+s:replace{1, 2 }
+
+s:drop()
+
test_run:cmd("switch default")
test_run:cmd("stop server tx_man")
test_run:cmd("cleanup server tx_man")
--
2.7.4
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [Tarantool-patches] [PATCH 0/2] Fix transaction manager
2020-10-22 16:02 [Tarantool-patches] [PATCH 0/2] Fix transaction manager Aleksandr Lyapunov
2020-10-22 16:02 ` [Tarantool-patches] [PATCH 1/2] memtx: fix a bug in unlinking story lists Aleksandr Lyapunov
2020-10-22 16:02 ` [Tarantool-patches] [PATCH 2/2] memtx: fix a bug in TX that caused deletion of a durty tuple Aleksandr Lyapunov
@ 2020-10-22 16:42 ` Alexander V. Tikhonov
2 siblings, 0 replies; 4+ messages in thread
From: Alexander V. Tikhonov @ 2020-10-22 16:42 UTC (permalink / raw)
To: Aleksandr Lyapunov; +Cc: tarantool-patches
Hi Aleksandr, thanks for the patch. As I see it really fixed the issue
[1], and no new degradations occurred. Patchset LGTM.
[1] - https://gitlab.com/tarantool/tarantool/-/pipelines/206407310
On Thu, Oct 22, 2020 at 07:02:31PM +0300, Aleksandr Lyapunov wrote:
> There was two issues:
> * https://github.com/tarantool/tarantool/issues/5423
> * tx_man test faild on OSX builds
>
> Both problems are fixed now.
>
> Branch: alyapunov/gh-5423-dirty-tuple-unref
>
> Aleksandr Lyapunov (2):
> memtx: fix a bug in unlinking story lists
> memtx: fix a bug in TX that caused deletion of a durty tuple
>
> src/box/memtx_space.c | 1 +
> src/box/memtx_tx.c | 67 +++++++++++++++++++++++++++++++++++++++---------
> test/box/tx_man.result | 43 +++++++++++++++++++++++++++++++
> test/box/tx_man.test.lua | 16 ++++++++++++
> 4 files changed, 115 insertions(+), 12 deletions(-)
>
> --
> 2.7.4
>
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2020-10-22 16:42 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-10-22 16:02 [Tarantool-patches] [PATCH 0/2] Fix transaction manager Aleksandr Lyapunov
2020-10-22 16:02 ` [Tarantool-patches] [PATCH 1/2] memtx: fix a bug in unlinking story lists Aleksandr Lyapunov
2020-10-22 16:02 ` [Tarantool-patches] [PATCH 2/2] memtx: fix a bug in TX that caused deletion of a durty tuple Aleksandr Lyapunov
2020-10-22 16:42 ` [Tarantool-patches] [PATCH 0/2] Fix transaction manager Alexander V. Tikhonov
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox