Gary Gregory created LOG4J2-1562:
------------------------------------
Summary: SocketAppender memoey usage grows unbounded if it cannot
connect to a server
Key: LOG4J2-1562
URL: https://issues.apache.org/jira/browse/LOG4J2-1562
Project: Log4j 2
Issue Type: Bug
Components: Appenders
Affects Versions: 2.6.2
Reporter: Gary Gregory
Assignee: Gary Gregory
SocketAppender memoey usage grows unbounded if it cannot connect to a server.
The socket appender uses a ByteArrayOutputStream as its data sink if a socket
cannot be opened to the server. The appender then tries to reconnect to this
server and then overwrites the output stream. The data in the
ByteArrayOutputStream is lost. The problem is that if the server is never
there, the ByteArrayOutputStream will keep on accumulating log events until the
JVM runs out of memory.
The proposed solution is to replace using a ByteArrayOutputStream with a
NullOutputStream.
As Remko remarks on the dev ML: "Actually supporting message buffering until
reconnect properly is not going to be trivial. (Wouldn't a spool file be better
than a memory buffer? Bounded or unbounded? What to do if we discover an unsent
spool file at startup? How do we handle disconnects while sending events from
the spool file? What about log4j shutdown while sending a large spool file? Or
process crashes?) We're entering Kafka-like territory here..."
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
