[Merged] AutoDL/v1.8 wish list and issues
This issue recaps wishes made for AutoDL that have found work arounds and have lesser priority. (Lukasz will test)
UI FIXES (Tyler)
- [x] Add server time on web page AND Add UTC after "Competition Ends:" date and in the schedule
- [x] UI change suggested by Andre
- [ ] Results are not showing properly in the admin submission panel.
BUGS (many emails were sent about it with the links to the issues on the forum):
- [ ] Error in computing total execution time (shows more time used than the sum of children)
- [ ] Weird "Submission Failed" (submission OK but denoted failed)
QUEUE PROBLEMS (Eric) Implement diagnosis tools for queueing - Fix the FIFO - elastic cloud
- [ ] Improve admin tools. We should monitor whether there are some jobs stuck waiting for abnormal times. Statistics for queue monitoring see also elastic cloud.
- [x] Check parent scoring program never stuck in submitting mode.
- [ ] GET CANCEL TO WORK: Check cancel behavior works properly Also: When we are in the admin "Submissions" panel and click on a "cancel" button, we are redirected to the main competition page; no acknowledgement cancel was done and we leave the "Submissions" panel page. Also Submissions cannot be dowloaded if canceled or failed.
- [x] Erratic compute time Difference between compute time and duration; runaway time then a program fails. See also this bug report and the fact that jobs of workers that died never get marked as cancelled or failed.
TWO PHASE PROBLEM Two parent phases - Having a final phase and migrating automatically:
- [ ] Button for starting auto-migration (priority not possible for v1.8)
- [ ] Get auto-migration to work properly for multiple-parent-phase competitions when we have children phases.
- (AutoDL Auto-Migration Competition Demo: #2626)
LATER
- [ ] Add new pages in editor
- [ ] Missing parent scoring program execution
- [ ] Add warning when people make a submission in a sub-phase.
- [ ] Check that "If submission beats old score, put submission on leaderboard" works properly (for the moment we show the last submission).
- [ ] Verify that a competitions has the correct boxes checked before launching it. Document this behavior with parent and children phases and look at forward compatibility with v2.
- [ ] Prefix error messages with name of program it is coming from
- [ ] Improve refresh
Issue with 50 phases/datasets: #2703
Issues:
- [ ] Dump in autodl cannot be reloaded #2825
- [ ] Error: [Errno 104] Connection reset by peer #2774
- [ ] Downloading processes block the website #2752
- [ ] Weird compute time #2674
- [ ] Uploading size limit: 2GB #2625
What about this version 1.8? It is used and maintain somewhere?
There is still this instance autodl.lri.fr and maybe also a version maintained by 4thParadigm.
But I think we want to put all the effort in the current develop branch to avoid having again this situation where we have new features separated in several independent and hard to maintain branches.