Ivan Bessonov created IGNITE-20419:
--------------------------------------
Summary: Raft "snapshots" are broken in aipersist
Key: IGNITE-20419
URL: https://issues.apache.org/jira/browse/IGNITE-20419
Project: Ignite
Issue Type: Bug
Reporter: Ivan Bessonov
This code:
{code:java}
storage.runConsistently(locker -> {
storage.lastApplied(maxLastAppliedIndex, maxLastAppliedTerm);
return null;
}); {code}
Here we don't touch any pages. As a result, updated index value may not be
persisted to the storage in the next checkpoint. This causes eventual exception
on restart:
{noformat}
java.lang.IllegalArgumentException: Try to truncate logs before 336, but the
firstLogIndex is 370
at
org.apache.ignite.raft.jraft.util.Requires.requireTrue(Requires.java:92)
~[main/:?]
at
org.apache.ignite.raft.jraft.storage.impl.LogManagerImpl.truncatePrefix(LogManagerImpl.java:997)
~[main/:?]
at
org.apache.ignite.raft.jraft.storage.impl.LogManagerImpl.setSnapshot(LogManagerImpl.java:654)
~[main/:?]
...{noformat}
In checkpointer, we must check for changed meta as well.
Everything here can be tested with unit tests.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
