[Tarantool-patches] [PATCH 0/2] Validate result of upserts squash & skip invalid upserts which can't be applied
Nikita Pettik
korablev at tarantool.org
Fri May 22 05:42:06 MSK 2020
On 21 May 00:13, Vladislav Shpilevoy wrote:
> Thanks for the patchset!
>
> I grepped for other vy_apply_upsert() usage cases and found these:
>
> - vy_build_recover_stmt() - should we patch it too?
No, I'd say not. If user wants to skip errors and continue recovery
process one can use force_recovery option.
> - vy_tx_write() - here the upsert is skipped in case of any error, but
> what if it was not a ClientError?
Originally, there was a comment:
/*
* Ignore a memory error, because it is
* not critical to apply the optimization.
*/
So I decided to keep this logic. I guess returning -1 in case of OOM
is also OK. So I can add check here as well. Let me know.
Btw, I've found another one bug concerned with upserts. It is revealed
since upserts can turn out to be invalid not only due to OOM.
Here's the patch (see also as an attachment). It is located on the top
of np/gh-1622-skip-invalid-upserts branch.
Author: Nikita Pettik <korablev at tarantool.org>
Date: Fri May 22 04:38:45 2020 +0300
vinyl: handle invalid upserts in background squash process
When L0 collects more than certain number of upserts (to be more precise
VY_UPSERT_THRESHOLD == 128), background process is launched to squash
them. During this squash, new statements could be inserted into L0.
Until committed prepared statements feature lsn = MAX_LSN as a special
marker. It is assumed that all upserts will be successfully squashed. As
a result statements with given key in L0 will be able to have only
MAX_LSN lsn (except for the squash result). However, it is not so if at
least one upsert is failed to be applied: it will get stack in L0 until
dump and it has normal (i.e. < MAX_LSN) lsn. So the assertion which
verifies that lsn of upsert in L0 is greater than MAX_LSN is wrong.
Moreover, if we execute enough (> 128) invalid upserts, they will
prevent from squashing normal upserts. So let's skip invalid upserts and
don't account them while calculating number of upserts residing in L0.
Follow-up #1622
diff --git a/src/box/vinyl.c b/src/box/vinyl.c
index 55a4c8fb7..b39e3b9d8 100644
--- a/src/box/vinyl.c
+++ b/src/box/vinyl.c
@@ -3539,10 +3539,21 @@ vy_squash_process(struct vy_squash *squash)
if (vy_tuple_compare(result, mem_stmt, lsm->cmp_def) != 0 ||
vy_stmt_type(mem_stmt) != IPROTO_UPSERT)
break;
- assert(vy_stmt_lsn(mem_stmt) >= MAX_LSN);
- vy_stmt_set_n_upserts((struct tuple *)mem_stmt, n_upserts);
- if (n_upserts <= VY_UPSERT_THRESHOLD)
- ++n_upserts;
+ /*
+ * Upsert was not applied in vy_history_apply(),
+ * but it still resides in tree memory. Ignore
+ * such statements and do not account them while
+ * counting upserts. Otherwise invalid upserts will
+ * get stack and won't allow other upserts to squash.
+ */
+ if (vy_stmt_lsn(mem_stmt) >= MAX_LSN) {
+ vy_stmt_set_n_upserts((struct tuple *) mem_stmt,
+ n_upserts);
+ if (n_upserts <= VY_UPSERT_THRESHOLD)
+ ++n_upserts;
+ } else {
+ vy_stmt_set_n_upserts((struct tuple *) mem_stmt, 0);
+ }
vy_mem_tree_iterator_prev(&mem->tree, &mem_itr);
}
diff --git a/test/vinyl/gh-1622-skip-invalid-upserts.result b/test/vinyl/gh-1622-skip-invalid-upserts.result
index b59ed53dc..12ca65c19 100644
--- a/test/vinyl/gh-1622-skip-invalid-upserts.result
+++ b/test/vinyl/gh-1622-skip-invalid-upserts.result
@@ -130,6 +130,140 @@ s:select{}
| - - [2, 13]
| ...
+-- Try to insert too many invalid upserts in a row so that
+-- they trigger squashing procedure.
+--
+fiber = require('fiber')
+ | ---
+ | ...
+s:drop()
+ | ---
+ | ...
+
+s = box.schema.space.create('test', { engine = 'vinyl', field_count = 2 })
+ | ---
+ | ...
+pk = s:create_index('pk')
+ | ---
+ | ...
+s:replace({1, 1})
+ | ---
+ | - [1, 1]
+ | ...
+
+squashed_before = pk:stat().upsert.squashed
+ | ---
+ | ...
+applied_before = pk:stat().upsert.applied
+ | ---
+ | ...
+len_before = s:len()
+ | ---
+ | ...
+upsert_cnt = 129
+ | ---
+ | ...
+for i =1 , upsert_cnt do s:upsert({1, 1}, {{'=', 3, 5}}) end
+ | ---
+ | ...
+-- 1 squash is accounted in vy_squash_process() unconditionally.
+--
+while pk:stat().upsert.squashed ~= squashed_before + 1 do fiber.sleep(0.01) end
+ | ---
+ | ...
+-- Make sure invalid upserts are committed into tree.
+--
+assert(s:len() == len_before + upsert_cnt)
+ | ---
+ | - true
+ | ...
+assert(pk:stat().upsert.applied == applied_before)
+ | ---
+ | - true
+ | ...
+-- Adding new invalid upserts will not result in squash until
+-- we reach VY_UPSERT_THRESHOLD (128) limit again.
+--
+s:upsert({1, 1}, {{'=', 3, 5}})
+ | ---
+ | ...
+assert(pk:stat().upsert.squashed == squashed_before + 1)
+ | ---
+ | - true
+ | ...
+for i = 1, upsert_cnt - 3 do s:upsert({1, 1}, {{'=', 3, 5}}) end
+ | ---
+ | ...
+assert(pk:stat().upsert.squashed == squashed_before + 1)
+ | ---
+ | - true
+ | ...
+s:upsert({1, 1}, {{'=', 3, 5}})
+ | ---
+ | ...
+while pk:stat().upsert.squashed ~= squashed_before + 2 do fiber.sleep(0.01) end
+ | ---
+ | ...
+assert(pk:stat().upsert.squashed == squashed_before + 2)
+ | ---
+ | - true
+ | ...
+assert(pk:stat().upsert.applied == applied_before)
+ | ---
+ | - true
+ | ...
+box.snapshot()
+ | ---
+ | - ok
+ | ...
+-- No one upsert is applied so number of rows should not change.
+--
+assert(s:len() == len_before)
+ | ---
+ | - true
+ | ...
+assert(pk:stat().upsert.squashed == squashed_before + 2)
+ | ---
+ | - true
+ | ...
+assert(pk:stat().upsert.applied == applied_before)
+ | ---
+ | - true
+ | ...
+-- Now let's mix valid and invalid upserts. Make sure that only
+-- valid are accounted during squash/apply.
+--
+for i = 1, upsert_cnt do if i % 2 == 0 then s:upsert({1, 1}, {{'=', 3, 5}}) else s:upsert({1, 1}, {{'+', 2, 1}}) end end
+ | ---
+ | ...
+while pk:stat().upsert.squashed ~= squashed_before + 3 do fiber.sleep(0.01) end
+ | ---
+ | ...
+assert(s:len() == len_before + upsert_cnt)
+ | ---
+ | - true
+ | ...
+assert(pk:stat().upsert.applied == math.floor(upsert_cnt/2) + 1)
+ | ---
+ | - true
+ | ...
+box.snapshot()
+ | ---
+ | - ok
+ | ...
+assert(s:len() == len_before)
+ | ---
+ | - true
+ | ...
+assert(pk:stat().upsert.squashed == squashed_before + 3)
+ | ---
+ | - true
+ | ...
+assert(pk:stat().upsert.applied == math.floor(upsert_cnt/2) + 1)
+ | ---
+ | - true
+ | ...
+
s:drop()
| ---
| ...
diff --git a/test/vinyl/gh-1622-skip-invalid-upserts.test.lua b/test/vinyl/gh-1622-skip-invalid-upserts.test.lua
index eb93393be..27c168fcb 100644
--- a/test/vinyl/gh-1622-skip-invalid-upserts.test.lua
+++ b/test/vinyl/gh-1622-skip-invalid-upserts.test.lua
@@ -47,4 +47,55 @@ s:select{}
box.snapshot()
s:select{}
+-- Try to insert too many invalid upserts in a row so that
+-- they trigger squashing procedure.
+--
+fiber = require('fiber')
+s:drop()
+
+s = box.schema.space.create('test', { engine = 'vinyl', field_count = 2 })
+pk = s:create_index('pk')
+s:replace({1, 1})
+
+squashed_before = pk:stat().upsert.squashed
+applied_before = pk:stat().upsert.applied
+len_before = s:len()
+upsert_cnt = 129
+for i =1 , upsert_cnt do s:upsert({1, 1}, {{'=', 3, 5}}) end
+-- 1 squash is accounted in vy_squash_process() unconditionally.
+--
+while pk:stat().upsert.squashed ~= squashed_before + 1 do fiber.sleep(0.01) end
+-- Make sure invalid upserts are committed into tree.
+--
+assert(s:len() == len_before + upsert_cnt)
+assert(pk:stat().upsert.applied == applied_before)
+-- Adding new invalid upserts will not result in squash until
+-- we reach VY_UPSERT_THRESHOLD (128) limit again.
+--
+s:upsert({1, 1}, {{'=', 3, 5}})
+assert(pk:stat().upsert.squashed == squashed_before + 1)
+for i = 1, upsert_cnt - 3 do s:upsert({1, 1}, {{'=', 3, 5}}) end
+assert(pk:stat().upsert.squashed == squashed_before + 1)
+s:upsert({1, 1}, {{'=', 3, 5}})
+while pk:stat().upsert.squashed ~= squashed_before + 2 do fiber.sleep(0.01) end
+assert(pk:stat().upsert.squashed == squashed_before + 2)
+assert(pk:stat().upsert.applied == applied_before)
+box.snapshot()
+-- No one upsert is applied so number of rows should not change.
+--
+assert(s:len() == len_before)
+assert(pk:stat().upsert.squashed == squashed_before + 2)
+assert(pk:stat().upsert.applied == applied_before)
+-- Now let's mix valid and invalid upserts. Make sure that only
+-- valid are accounted during squash/apply.
+--
+for i = 1, upsert_cnt do if i % 2 == 0 then s:upsert({1, 1}, {{'=', 3, 5}}) else s:upsert({1, 1}, {{'+', 2, 1}}) end end
+while pk:stat().upsert.squashed ~= squashed_before + 3 do fiber.sleep(0.01) end
+assert(s:len() == len_before + upsert_cnt)
+assert(pk:stat().upsert.applied == math.floor(upsert_cnt/2) + 1)
+box.snapshot()
+assert(s:len() == len_before)
+assert(pk:stat().upsert.squashed == squashed_before + 3)
+assert(pk:stat().upsert.applied == math.floor(upsert_cnt/2) + 1)
+
s:drop()
\ No newline at end of file
> On 13/04/2020 23:55, Nikita Pettik wrote:
> > Branch: https://github.com/tarantool/tarantool/tree/np/gh-1622-skip-invalid-upserts
> > Issue: https://github.com/tarantool/tarantool/issues/1622
> >
> > In fact, the first patch in series does not make much sense without
> > second since otherwise dump or compaction processes (and any other
> > operations that read data as well) will fail whenever they are started.
> > But still I've divided fix into two patches for the sake of review
> > simplicity.
> >
> > Nikita Pettik (2):
> > vinyl: validate resulting tuple after upsert is applied
> > vinyl: skip invalid upserts during squash
> >
> > src/box/vy_history.c | 20 ++-
> > src/box/vy_lsm.c | 13 +-
> > src/box/vy_tx.c | 29 ++--
> > src/box/vy_upsert.c | 4 +
> > src/box/vy_write_iterator.c | 34 +++--
> > .../vinyl/gh-1622-skip-invalid-upserts.result | 135 ++++++++++++++++++
> > .../gh-1622-skip-invalid-upserts.test.lua | 50 +++++++
> > 7 files changed, 258 insertions(+), 27 deletions(-)
> > create mode 100644 test/vinyl/gh-1622-skip-invalid-upserts.result
> > create mode 100644 test/vinyl/gh-1622-skip-invalid-upserts.test.lua
> >
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-vinyl-handle-invalid-upserts-in-background-squash-pr.patch
Type: text/x-diff
Size: 8278 bytes
Desc: not available
URL: <https://lists.tarantool.org/pipermail/tarantool-patches/attachments/20200522/a278fa33/attachment.bin>
More information about the Tarantool-patches
mailing list