[
https://issues.apache.org/jira/browse/LOG4J2-164?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Remko Popma updated LOG4J2-164:
-------------------------------
Description:
Many buffered Appenders have an attribute "immediateFlush" that gives users the
option to follow each write with a flush to ensure that all messages are
available on disk. The drawback of the immediateFlush idiom is the performance
impact of touching the disk on every single event.
If the appender could determine from the current LogEvent whether any more
events are expected, it could avoid unnecessary flushes and only write to disk
when the last event was received (or the buffer is full).
This would work especially nicely when receiving bursts of events - a common
scenario - when you want to avoid adding more overhead to a system that is
already very busy.
Asynchronous loggers and appenders are well-positioned to use such a mechanism:
they can easily call LogEvent.setEndOfBatch(true) if the current event is the
last one in the queue.
The patches I proposed for LOG4J2-151 and LOG4J2-163 already use this mechanism
internally. The FastFileAppender included in these patches checks if a log
event is a specific class, casts the event to that class, and finally calls the
isEndOfBatch method on that event. However I think the idea is generally
applicable to any buffered appender and adding these methods to the LogEvent
interface would benefit users of the existing AsynchAppender and
OutputStreamAppender classes.
was:
Many buffered Appenders have an attribute "immediateFlush" that gives users the
option to follow each write with a flush to ensure that all messages are
available on disk. The drawback of the immediateFlush idiom is the performance
impact of touching the disk on every single event.
If the appender could determine from the current LogEvent whether any more
events are expected, it could avoid unnecessary flushes and only write to disk
when the last event was received (or the buffer is full).
This would work especially nicely when receiving bursts of events - a common
scenario - when you want to avoid adding more overhead to a system that is
already very busy.
Asynchronous loggers and appenders are well-positioned to use such a mechanism:
they can easily call LogEvent.setEndOfBatch(true) if the current event is the
last one in the queue.
The patches I proposed for LOG4J2-151 and LOG4J2-163 already use this mechanism
internally with the FastFileAppender included in these patches, but I think the
idea is generally applicable to any buffered appender.
> Add methods is/setEndOfBatch to LogEvent (buffered appender performance
> improvement)
> ------------------------------------------------------------------------------------
>
> Key: LOG4J2-164
> URL: https://issues.apache.org/jira/browse/LOG4J2-164
> Project: Log4j 2
> Issue Type: Improvement
> Components: Core
> Affects Versions: 2.0-beta4
> Reporter: Remko Popma
>
> Many buffered Appenders have an attribute "immediateFlush" that gives users
> the option to follow each write with a flush to ensure that all messages are
> available on disk. The drawback of the immediateFlush idiom is the
> performance impact of touching the disk on every single event.
> If the appender could determine from the current LogEvent whether any more
> events are expected, it could avoid unnecessary flushes and only write to
> disk when the last event was received (or the buffer is full).
> This would work especially nicely when receiving bursts of events - a common
> scenario - when you want to avoid adding more overhead to a system that is
> already very busy.
> Asynchronous loggers and appenders are well-positioned to use such a
> mechanism: they can easily call LogEvent.setEndOfBatch(true) if the current
> event is the last one in the queue.
> The patches I proposed for LOG4J2-151 and LOG4J2-163 already use this
> mechanism internally. The FastFileAppender included in these patches checks
> if a log event is a specific class, casts the event to that class, and
> finally calls the isEndOfBatch method on that event. However I think the idea
> is generally applicable to any buffered appender and adding these methods to
> the LogEvent interface would benefit users of the existing AsynchAppender and
> OutputStreamAppender classes.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
