Hey all: Saturday morning was when we started to play catch up with the previous runs we couldn't attempt while we were restoring. Two factors contributed to the duplication of some entries:
1) For feeds without date information (e.g. RSS 0.9x and later versions where dates are omitted), FeedBlitz maintains a high water mark (actually four of them) to prevent resesnds. No database, no saved high water mark, so the posts went out again.
2) Under stress the DB was still flaky while we went through this process. Some tasks failed and our cooperative architecture restarted them. But it can only restart from the last known good point, which can result in resends. If the same batch fails more than once, well.... you see what happens.
I really, really wanted the open source database we're using to have worked out. But it seems to nbe simply not ready for the kind of stress we put it through daily; we're going to have to plan a migration to a proprietary but more robust relational system.
Meanwhile I want to thank everyone for hanging in there.
4 Comments:
I'd just subscribed, set up my first syndication then poof!
I'm glad to know you're busy fixing the problem, and look forward to being able to use the service again.
Multiple problems with this AM's launch, too. Some subscribers got double mailings.. many have dead links. Please repair.
Thankyou for keeping me posted. I am tired of listening to my subscribers whining that they didn't get my post.
Hey all: Saturday morning was when we started to play catch up with the previous runs we couldn't attempt while we were restoring. Two factors contributed to the duplication of some entries:
1) For feeds without date information (e.g. RSS 0.9x and later versions where dates are omitted), FeedBlitz maintains a high water mark (actually four of them) to prevent resesnds. No database, no saved high water mark, so the posts went out again.
2) Under stress the DB was still flaky while we went through this process. Some tasks failed and our cooperative architecture restarted them. But it can only restart from the last known good point, which can result in resends. If the same batch fails more than once, well.... you see what happens.
I really, really wanted the open source database we're using to have worked out. But it seems to nbe simply not ready for the kind of stress we put it through daily; we're going to have to plan a migration to a proprietary but more robust relational system.
Meanwhile I want to thank everyone for hanging in there.
Post a Comment
Note: Only a member of this blog may post a comment.
<< Home