Kirill Tkalenko created IGNITE-20062:
----------------------------------------
Summary: Temporary fix for StackOverflowError in
Replica#waitForActualState when a peer leaves
Key: IGNITE-20062
URL: https://issues.apache.org/jira/browse/IGNITE-20062
Project: Ignite
Issue Type: Improvement
Reporter: Kirill Tkalenko
Assignee: Kirill Tkalenko
Fix For: 3.0.0-beta2
It has been found that when a peer leaves,
*org.apache.ignite.internal.replicator.Replica#waitForActualState* can generate
*StackOverflowError* that can lead to unpleasant consequences.
For example, as a result of an *StackOverflowError*, we will not be able to
execute *busyLock.leaveBusy();* in
*org.apache.ignite.internal.raft.RaftGroupServiceImpl#sendWithRetry(org.apache.ignite.internal.raft.Peer,
java.util.function.Function<org.apache.ignite.internal.raft.Peer,? extends
org.apache.ignite.network.NetworkMessage>, long,
java.util.concurrent.CompletableFuture<R>)*, because the stack is full and when
entering to *leaveBusy*, *StackOverflowError* will be thrown again, which will
lead to freezing at the node stop.
At the moment I don't know how to fix this properly, while a temporary solution
comes to mind - add thread pool and use it instead of *Runnable::run* in
*Replica#waitForActualState*, you need to properly fix it in 123.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
