[
https://issues.apache.org/jira/browse/RATIS-878?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17095225#comment-17095225
]
runzhiwang edited comment on RATIS-878 at 4/29/20, 8:10 AM:
------------------------------------------------------------
There are 2 options to avoid the infinite restart of LogAppender.
1. If LogAppender try to restart when remove group, we refuse it, else we
restart.
2. If restart times do not exceed the limit i.e. checkCanRestart is true, both
the 2 options can restart GrpcLogAppender again
3. If exceed the limit i.e. checkCanRestart is false, there are 2 options we
can choose:
a. close the pipeline, throw a special exception back to client
b. leader step down, wait all the GrpcLogAppenders to be dead, and trigger a
leader election, and throw a special exception back to client
The pseudo code as follow.
{code:java}
Queue<TimeStamp, LogAppender> queue;
restart() {
if (removeGroup) {
return;
}
if (checkCanRestart()) {
newLogAppender = restartLogAppender();
if (queue.size == 3) {
queue.poll();
}
queue.add(<now, newLogAppender>);
} else {
stop restart, and there are three options we can choose.
a. close the pipeline, throw a special exception back to client
b. If not in remove group, leader step down, wait all the GrpcLogAppenders
to be dead, and trigger a leader election, throw a special exception back
}
}
boolean checkCanRestart() {
if (queue.size() < 3) {
return true;
}
if ((now - queue.peek().TimeStamp > 1 hour) and
(queue.peek().LogAppender.isAlive == false)) {
return true;
}
return false;
}
{code}
was (Author: yjxxtd):
There are 2 options to avoid the infinite restart of LogAppender.
1. If LogAppender try to restart when remove group, we refuse it, else we
restart.
2. If restart times do not exceed the limit i.e. checkCanRestart is true, both
the 2 options can restart GrpcLogAppender again
3. If exceed the limit i.e. checkCanRestart is false, there are 2 options we
can choose:
a. close the pipeline, throw a special exception back to client
b. leader step down, wait all the GrpcLogAppenders to be dead, and trigger a
leader election, and throw a special exception back to client
The pseudo code as follow.
{code:java}
Queue<TimeStamp, LogAppender> queue;
restart() {
if (removeGroup) {
return;
}
if (checkCanRestart()) {
newLogAppender = restartLogAppender();
if (queue.size == 3) {
queue.poll();
}
queue.add(<now, newLogAppender>);
} else {
stop restart, and there are three options we can choose.
a. close the pipeline, throw a special exception back to client
b. If not in remove group, leader step down, wait all the
GrpcLogAppenders to be dead, and trigger a leader election, throw a special
exception back
}
}
boolean checkCanRestart() {
if (queue.size() < 3) {
return true;
}
if ((now - queue.peek().TimeStamp > 1 hour) and
(queue.peek().LogAppender.isAlive == false)) {
return true;
}
return false;
}
{code}
> Infinite restart of LogAppender
> --------------------------------
>
> Key: RATIS-878
> URL: https://issues.apache.org/jira/browse/RATIS-878
> Project: Ratis
> Issue Type: Bug
> Reporter: runzhiwang
> Assignee: runzhiwang
> Priority: Major
>
> The details please refer it here
> [RATIS-840|https://issues.apache.org/jira/browse/RATIS-840].
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
