Confirming a problem reported on the users@ list. This affects both NIO and APR protocols when file is delivered with sendfile. Note, that the file have to be >48kb to trigger use of sendfile. Steps to reproduce: 1) Configure three connectors: <Connector port="8081" protocol="org.apache.coyote.http11.Http11Protocol" connectionTimeout="20000" redirectPort="8443" /> <Connector port="8082" protocol="org.apache.coyote.http11.Http11NioProtocol" connectionTimeout="20000" redirectPort="8443" /> <Connector port="8083" protocol="org.apache.coyote.http11.Http11AprProtocol" connectionTimeout="20000" redirectPort="8443" /> and configure AccessLogValve in a <Host>: <Valve className="org.apache.catalina.valves.AccessLogValve" directory="logs" prefix="localhost_access_log." suffix=".txt" pattern="%h %l %u %t "%r" %s %b (%D ms)" /> The difference from common pattern is that I added "(%D ms)" at the end. 2) Put some big file (>48k) as webapps/ROOT/file 3) Download it 3 times: http://localhost:8081/file?8081 http://localhost:8082/file?8082 http://localhost:8083/file?8083 4) Access Log: 127.0.0.1 - - [10/Dec/2011:01:05:08 +0300] "GET /file?8081 HTTP/1.1" 200 15027784 (4766 ms) 127.0.0.1 - - [10/Dec/2011:01:05:23 +0300] "GET /file?8082 HTTP/1.1" 200 - (31 ms) 127.0.0.1 - - [10/Dec/2011:01:05:48 +0300] "GET /file?8083 HTTP/1.1" 200 - (16 ms) Note that '-' is printed instead of file size for Nio and Apr connectors. Workarounds: ------------ A). Disable sendfile with useSendfile="false" on a connector B). Add the following pattern to AccessLogValve: "%{org.apache.tomcat.sendfile.start}r %{org.apache.tomcat.sendfile.end}r" That will log the range of bytes sent by sendfile. E.g.: 127.0.0.1 - - [10/Dec/2011:01:05:08 +0300] "GET /file?8081 HTTP/1.1" 200 15027784 (4766 ms) - - 127.0.0.1 - - [10/Dec/2011:01:05:23 +0300] "GET /file?8082 HTTP/1.1" 200 - (31 ms) 0 15027784 127.0.0.1 - - [10/Dec/2011:01:05:48 +0300] "GET /file?8083 HTTP/1.1" 200 - (16 ms) 0 15027784 -------------- Overall, looking at the timing value printed by %D I would say that logging occurs before data are sent. So the value of 0 bytes is correct. It could be possible to log file size from request attributes set by sendfile, but that cannot account for aborted downloads. Though I think that would be better than the current logging of '0'. At least this limitation can be mentioned in documentation for %b and %B patterns in AccessLogValve section in valve.html and in Javadoc.
Since sendfile sends the data asynchronously on a separate thread, the access log can never know exactly how many bytes are sent without destroying the benefits of sendfile. I lean towards logging the bytes intended to be sent (calculated from the request attributes).
Fixed in trunk and 7.0.x and will be included in 7.0.24 onwards.