[Tarantool-patches] [PATCH v14 0/6] qsync: implement packets filtering
Cyrill Gorcunov
gorcunov at gmail.com
Fri Sep 10 18:29:04 MSK 2021
Guys, take a look please, once time permit. The questionable moments:
- use filter disabling procedure for join/recovery: we make it so since
snapshot has promote record which fills initial limbo state
- need more tests to cover all possible scenarios
- I keep filter_confirm_rollback() as is but rereading Vlad's comment
>
> 9. What if rollback is for LSN > limbo's last LSN? It
> also means nothing to do. The same for confirm LSN < limbo's
> first LSN.
>
I presume I need to traverse limbo and test if incoming LSN is
present inside current queue.
Anyway I send this version early to gather more comments, I hope
not that much left to implement to be ready for merging.
previous series https://lists.tarantool.org/tarantool-patches/20210804190752.488147-1-gorcunov@gmail.com/
branch gorcunov/gh-6036-rollback-confirm-14
issue https://github.com/tarantool/tarantool/issues/6036
v6:
- use txn_limbo_terms name for structure
- rebase on fresh sp/gh-6034-empty-limbo-transition branch
- rework filtering chains
v8:
- add ability to disable filtering for local recovery
and join stages
- update tests
v9:
- opencode terms tracking
- fix tests to use wait function since log output might
be deferred by OS
v10:
- rework FILTER_IN and FILTER_PROMOTE chains with more
detailed packets inspection
- preserve old naming for terms manipulations
- require the packet's replica_id to match limbo owner_id
all the time
v11-13: internal
v14:
- use straightforward packet inspection by their type
without more general type routing
- tried to hide locking api inside limbo level but since
journal completion is called from inside of sched fiber
the lock owner get migrated which cause error thus leave
explicit locking instead
- added updating of limbo::confirmed_lsn since we need it
for proper validation
- added new error code to distinguish filter errors from
anything else
- use say_error instead of say_info
- keep disabling of filtration inside initial join/recovery
because we're filling initial limbo state
Cyrill Gorcunov (6):
qsync: track confirmed lsn number on reads
qsync: update confirmed lsn on initial promote request
latch: add latch_is_locked helper
qsync: order access to the limbo terms
qsync: filter incoming synchro requests
test: add replication/gh-6036-rollback-confirm
.../gh-6036-qsync-filter-packets.md | 9 +
src/box/applier.cc | 26 +-
src/box/box.cc | 30 +-
src/box/errcode.h | 1 +
src/box/memtx_engine.cc | 3 +-
src/box/txn_limbo.c | 337 +++++++++++++++---
src/box/txn_limbo.h | 85 ++++-
src/lib/core/latch.h | 11 +
test/box/error.result | 1 +
test/replication/gh-6036-master.lua | 1 +
test/replication/gh-6036-node.lua | 33 ++
test/replication/gh-6036-replica.lua | 1 +
.../gh-6036-rollback-confirm.result | 180 ++++++++++
.../gh-6036-rollback-confirm.test.lua | 92 +++++
14 files changed, 747 insertions(+), 63 deletions(-)
create mode 100644 changelogs/unreleased/gh-6036-qsync-filter-packets.md
create mode 120000 test/replication/gh-6036-master.lua
create mode 100644 test/replication/gh-6036-node.lua
create mode 120000 test/replication/gh-6036-replica.lua
create mode 100644 test/replication/gh-6036-rollback-confirm.result
create mode 100644 test/replication/gh-6036-rollback-confirm.test.lua
base-commit: b0431cf8f47e9d081f6a402bc18edb1d6ad49847
--
2.31.1
More information about the Tarantool-patches
mailing list