On Mar 10, 2010, at 10:37 AM, Stefano Nichele wrote: > Hi all, > looking for perfomance issue in my server application, I found that > AppenderSkeleton.doAppend method is synchronized. > I google a bit and I have found some messages but not a clear anwser to my > doubt: why that method is synchronized ? it seems that just check the > threshold, the filter and that "append" the event but in this way is > serializing all log calls. I'm logging events on the db and also using > AsyncAppender + my own appender I'm not able to obtain good performance since > just one thread at time can append (insert) an event and the AsyncAppender > queue grows pretty quick. > > For sure i'm missing something...could you help me to understand the root > reason for having it synchronized ? I'm thinking to have my owm > implementation of AppenderSkeleton removing synchronization but I'm not sure > if there are drawbacks that im not taking in consideration. > > Thanks a lot in advance > ste > >
log4j 1.2 was designed a long time ago and relies on that big lock to provide thread safety. Other classes in log4j (layouts, appenders and the like) were designed assuming that they would be externally synchronized by that lock and likely not safe if that lock is bypassed. You could either implement the Appender interface or extend AppenderSkeleton but override doAppend (copying and pasting the implementation but without the synchronized modifier, but all appenders, layouts, etc used must be able to operate safely without that lock. Addressing this issue is one of the core design goals for log4j 2.0 (http://issues.apache.org/jira/browse/LOG4J2-3) --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
