From: Vladimir Davydov <vdavydov.dev@gmail.com> To: kostja@tarantool.org Cc: tarantool-patches@freelists.org Subject: [PATCH v3 0/3] Fix DML vs DDL race Date: Thu, 28 Mar 2019 19:17:47 +0300 [thread overview] Message-ID: <cover.1553789818.git.vdavydov.dev@gmail.com> (raw) In Vinyl a DML request may yield on disk read while still having a pointer to the target space saved on stack. If the space is altered during this time window, the DML request will most certainly crash while trying to dereference the stale pointer when it resumes. The solution is simple: abort all affected transactions whenever we delete a space from the cache. https://github.com/tarantool/tarantool/issues/3420 https://github.com/tarantool/tarantool/commits/dv/gh-3420-vy-abort-tx-in-ddl Changes in v3: - Keep track of the *space* affected by the last prepared statement rather than the LSM tree. v2: https://www.freelists.org/post/tarantool-patches/PATCH-v2-03-Fix-DML-vs-DDL-race Changes in v2: - Instead of introducing a per transaction space hash, simply remember the pointer to the last prepared index. Thanks to Kostja for the tip. - Drop the patch allowing transactions affecting only replica local spaces outlast a switch to read-only mode, as we can't easily implement it without a per transaction space hash. Anyway, it was out of the scope of this issue. v1: https://www.freelists.org/post/tarantool-patches/PATCH-04-Fix-DML-vs-DDL-race Vladimir Davydov (3): vinyl: make tx_manager_abort_writers_for_ddl more thorough vinyl: abort affected transactions when space is removed from cache Revert "test: skip ddl test for vinyl on travis" rpm/tarantool.spec | 2 - src/box/blackhole.c | 1 + src/box/memtx_space.c | 1 + src/box/schema.cc | 3 ++ src/box/space.c | 6 +++ src/box/space.h | 15 ++++++ src/box/sysview.c | 1 + src/box/vinyl.c | 33 ++++++++++--- src/box/vy_point_lookup.c | 14 ++++++ src/box/vy_read_iterator.c | 13 +++++ src/box/vy_tx.c | 14 ++++-- src/box/vy_tx.h | 15 ++++-- test/vinyl/ddl.skipcond | 6 --- test/vinyl/errinj_ddl.result | 110 +++++++++++++++++++++++++++++++++++++++++ test/vinyl/errinj_ddl.test.lua | 51 +++++++++++++++++++ 15 files changed, 265 insertions(+), 20 deletions(-) delete mode 100644 test/vinyl/ddl.skipcond -- 2.11.0
next reply other threads:[~2019-03-28 16:17 UTC|newest] Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top 2019-03-28 16:17 Vladimir Davydov [this message] 2019-03-28 16:17 ` [PATCH v3 1/3] vinyl: make tx_manager_abort_writers_for_ddl more thorough Vladimir Davydov 2019-03-28 16:17 ` [PATCH v3 2/3] vinyl: abort affected transactions when space is removed from cache Vladimir Davydov 2019-03-28 16:17 ` [PATCH v3 3/3] Revert "test: skip ddl test for vinyl on travis" Vladimir Davydov 2019-03-28 17:53 ` [PATCH v3 0/3] Fix DML vs DDL race 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=cover.1553789818.git.vdavydov.dev@gmail.com \ --to=vdavydov.dev@gmail.com \ --cc=kostja@tarantool.org \ --cc=tarantool-patches@freelists.org \ --subject='Re: [PATCH v3 0/3] Fix DML vs DDL race' \ /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