Kirill Koulechov wrote: > Hi, > > i think i have a rather simple questions for the developers of tomcat. Let > me give short explanation. > > I am working on a research topic, which is generic fault detection and > tolerance for software architectures. i've picked apache as a sample of > c/c++ program and tomcat as a java program. don't get me wrong: i did it > because i beleave that these programs are well known, and developed > professionaly, since i need fault statistics that are not obscured by simply > omitting basic rules of software development. > > i studied the bug database for tomcat4 versions 4.0.x Final, with the > assumption that these have been (fairly) stable versions. i omitted NEW and > UNCONFIRMED bugs, as well as INVALID, WONTFIX, DUPLICATE, WORKSFORME. > besides that, i ignored some IMHO rare plattforms like BeOS, OS/2, Netware, > OS/400 and included only critical, major and normal bugs. > > i did that in the hope to get representative statistics about common > software faults in a big, professionaly developed software product. if i did > something wrong, please correct me. at the end i got 163 bugs. > > now the important stuff. > > after examining some of them, i noticed some amount of bugs (i.e. bugs > #4930, #4047) causing unchecked exceptions, which strike all the way through > the stack, ending in org.apache.catalina.connector.http.HttpProcessor.run() > and then in java.lang.Thread.run(). in my understanding, this stops the > affected thread and all functionality that was done by > org.apache.catalina.connector.http.HttpProcessor.run() is unavailable now. > what is tomcats reaction to this? does the program crash/freeze/hang > whatever, or do these exceptions not affect the runime behaviour? or is only > the active servlet affected? besides that, i wondered what are the effects > of the bugs 3817, 4542, 5201, since the provided stack doesn't go up to > java.lang.Thread.run(). > > > thank you very much in advance, > > i really appreciate your answers because it doesnt help you developing > tomcat in the first place, but just think that you are helping to research > ways to prevent software from crashing at all!
I did some work to improve that on the old HTTP/1.1 connector since then. This connector won't be maintained anymore, unless a critical security problem is found (I hope not). On Coyote HTTP/1.1, I took extra care to make everything rock solid in terms of exception handling and algorithms (as well as fast, but we're talking about reliability here). Feel free to review it. Remy -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
