On Sunday 02 August 2009 00:51:10 Luca Foppiano wrote: > On Tue, 2009-07-28 at 21:00 +0300, makkalot wrote: > > Poll for output in Minions: > > This implementation is a result of an idea that was post on ML a few > > weeks ago. > > The main idea was to be able to get some output during the minion side > > execution. > > Yeah! I think is a great feature for long operations, having a minimun > output is better to have nothing for a while... :) > > > The idea maybe useful for tasks which are kind of long and user wants to > > get some notification about what is going on on the other side. Because > > we are talking about long running operations that stuff is reasonable > > only for async stuff. Here is some example of tracking a long operation > > on overlord : > > [...] > > > Because we use xml-rpc as transport protocol these kind of things are not > > very efficient in func. But the best way > > to do that was (if someone has better ideas please tell) to log all async > > requests in different log files according > > to their minion job ids. When a new async request comes a log object is > > attached to called method and new log file > > is created named with minion job id.Therefore user can use that log > > object to log stuff and overlord stuff can track > > that log file via its (overlord job_id)--(minion job_id) matching db.That > > structure maybe very useful for methods that > > take really long time to finish its stuff. The cons of the method is we > > have lots of separate log files and we have kind > > of high one to many transport communication,polling 100 minions for log > > information will not be very effcient.However it > > will work for single or a few machines very good ! > > > > Comments ,ideas ? > > Sorry for the delay, this implementation seems very nice. I would like > to test it before give suggestion but for me it's ok your approach. > > I'd also like to see the performance on huge number of machines and how > much we lost using it. What I'm thinking is how is it when you run > puppet on a lot of machines (I think puppet is not so fast so we can > breath! :)) Well as i said doing that for lots of machines at once doesnt make too much sense to me . What you can do is if you are using it for web application is to use version that polls single machine at time, so if user wants to see what is going on on some machine can click and poll its logs. Polling for globs "*" is kind of confusing because of lots of outputs around but it is up to you how will use it. And yes it will be slow for 100 machines to be polled for logs at once it is too much traffic around. > > Thank you again to implemented it :) > > Cheers > Luca > > _______________________________________________ > Func-list mailing list > Func-list@xxxxxxxxxx > https://www.redhat.com/mailman/listinfo/func-list _______________________________________________ Func-list mailing list Func-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/func-list
