Roman Puchkovskiy created IGNITE-18077:
------------------------------------------
Summary: Handle 'RAFT log abruptly deleted' scenario
Key: IGNITE-18077
URL: https://issues.apache.org/jira/browse/IGNITE-18077
Project: Ignite
Issue Type: Improvement
Reporter: Roman Puchkovskiy
Fix For: 3.0.0-beta2
The following case is possible: some writes happen to storage (via RAFT
infrastructure), then Ignite is stopped, RAFT log is deleted (or its FS
partition is simply unmounted) and node is started again. When it starts, it
sees no RAFT log, so, according to RAFT semantics, it might think that this is
a shiny fresh node that just entered the RAFT group, reset index to 0 and
request log/snapshot from a leader, while in reality it has some data in its
state machine storage and the correct action would be to remount the FS
partition with RAFT log.
So, we need a special handling for situations when there is no RAFT log (at
all), but the main storage reports that its persistedIndex is non-zero. One
option would be to put the whole Ignite node in a Maintenance node (or its
equivalent).
Please note that the described situation is different from the 'fresh node'
scenario where RAFT log is absent, but main storage's persistedIndex is 0. This
is a normal situation when joining a RAFT group for the first time, no special
handling is needed (it's handled by JRaft).
A design is needed.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
