Hi Any thoughts on this one?
It all boils down to a suggestion to add 1 methods to the org.apache.camel.Message API new method: - tryGetBody(Class type) or overload existing with a boolean to indicate ignore exception and return null - getBody(Class, true) The same applies for the org.apache.camel.spi.TypeConverter interface as we need a method on the line as above new method: - tryConvertTo(Class type) or overload existing with a boolean to indicate ignore exception and return null - convertTo(Class, true) The boolean thing is maybe a nice one as it have the same method name. On Mon, Mar 16, 2009 at 7:32 AM, Claus Ibsen <claus.ib...@gmail.com> wrote: > Hi > > In Camel 1.6.0 and 2.0 we have had a performance issue that could > seriously degrade performance by x2-x10 when you did stress test by > sending > 1000 msg/sec. > > The cause of this is the TypeConverter that will throw a > NoSuchTypeConverterExists when Camel cannot converter to the desired > type. > We do this internally in Camel in some areas of the code where we will > ignore this exception and eg. continue. > > We did a fix for this performance issue in the stream cache but this > was just a fix in once place. Lately this issue surfaced again in > camel-mina when using UDP. > > So I have been experimenting with doing a fix once for all. > > I suggest to: > - add 2 new methods to TypeConverter that returns *null* instread of > throwing the NoSuchTypeConverterExists exception. > I have named them: tryConvertTo. But maybe there is a better name? > - add 1 new method on Message to get the body using the try convert to > instead. > I have named the method: tryGetBody(T). But maybe there is a better name? > - Use tryGetBody internally in Camel where we today do try .. > catch(NoSuchTypeConverterExists) > - Then we can expose the regular getBody(T) for the end users, and let > it act as it does today by throwing the exception > - We can also optimize the DefaultTypeConverter to let it look in the > miss cache first if it have tried to convert this before but could not > > If the tryGetBody, tryConvertTo are you to your liking we can also do: > - add a new method to TypeConvert: boolean canConvertTo(T, value) > - add a new method to Message: boolean canGetBody(T, value) > - then use that method as guard before doing the regular convertTo/getBody > But this requires that you always use this as guard, to be sure that > the NoSuchTypeConverterExists is not thrown > And it also requires that we use the miss cache to not do 2x > convertions, 1 for the boolean check, and then 1 for the actual > convertion > It also requires that the body can be converted multiple times. > > Whether solution 1 or 2 we also need to check the other components and > fix it internally as well, so we dont have a situation like mina with > UDP that was affected by the performance drawback. > > In my experiment I have the first solution up and running. > > Any thoughts? > > I will create a JIRA ticket for this one as well so we wont forget it. > > > -- > Claus Ibsen > Apache Camel Committer > > Open Source Integration: http://fusesource.com > Blog: http://davsclaus.blogspot.com/ > -- Claus Ibsen Apache Camel Committer Open Source Integration: http://fusesource.com Blog: http://davsclaus.blogspot.com/