On Wed, 2006-03-01 at 00:18 -0500, Dan Williams wrote: > On Wed, 2006-03-01 at 06:03 +0100, Ralf Corsepius wrote: > > > Please check to make sure the job did > > > not actually get enqueued. > > How about you fixing your code to not produce false alarms, if you are > > sure these are false alarms? > > To be perfectly fair, the error message doesn't say "The buildsys is > hung", it says just what it means; that the build system didn't give you > a job ID. The only reason that this message attained the connotations > it did, was because I kept asking people both on this list and on > #fedora-extras to report instances of this message. Since that's no > longer the case, I'd like to officially change the connotation that this > error message has to "check the web UI for your job for 5 minutes after > you submit your job, then report." Well, to me this would be a behavioral regression. > If you'd like, I can increase the plague-client timeout to 20s such that > it will almost always give you a job ID. Sorry, I am not familiar with the buildsys to the extend that would be required here, to be able comment on this. > However, that's really just > papering over the problem. Probably - This had been my feeling all the time, because experience tells me, struggling with timeouts always indicate a major design problem (likely synchronization or protocol deficiencies). > In the long run, we need to evaluate how to > deal with sqlite's table locking issues, which is the way the code is > architected as it is, which is why we have to synchronize access to the > database, and which is why you can't get a job ID back immediately. > Alternatively, some re-architecting of how builder status updates are > handled is necessary, which is the secondary cause of this issue. Ralf -- fedora-extras-list mailing list fedora-extras-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-extras-list
