Post-mortem analysis of a replication failure is painful as it is, but
with poor logging that we have it is usually nearly to impossible to
understand what happened in the frist place. This patch set targets on
ameliorating the situration by logging all relevant vclocks and ids on
join/subscribe.
Changes in v2:
- Log all replica id changes, because without them vclocks don't make
much sense.
- Print remote and local vclocks on subscribe to replica's vclock.
It may be useful if master's log rotated.
v1:
https://www.freelists.org/post/tarantool-patches/PATCH-replication-improve-joinsubscribe-logging
Vladimir Davydov (3):
vclock: use static buffer to format vclock
box: don't set cluster uuid and instance id in initial snapshot
replication: improve join/subscribe logging
src/box/alter.cc | 3 +-
src/box/applier.cc | 17 ++++++++---
src/box/bootstrap.snap | Bin 1911 -> 1831 bytes
src/box/box.cc | 27 +++++++++++------
src/box/error.cc | 6 ++--
src/box/gc.c | 6 ++--
src/box/lua/upgrade.lua | 7 +++--
src/box/replication.cc | 17 +++++------
src/box/vclock.c | 64 +++++++++++------------------------------
src/box/vclock.h | 5 ++--
src/box/xlog.c | 16 +++--------
test/box-py/bootstrap.result | 5 ++--
test/box-py/bootstrap.test.py | 16 ++---------
test/replication/misc.result | 9 +++++-
test/replication/misc.test.lua | 5 +++-
test/unit/vclock.cc | 3 +-
16 files changed, 91 insertions(+), 115 deletions(-)
--
2.11.0