Sync issue with multiple judgehosts

[jason-huh]

1. Judgehost A, B are running well.
2. Judgehost A is no longer able to run scripts (i.e. lack of disk space, abnormal termination) but Domserver still knows that it is alive. Hence, there occur lots of pending submissions assigned to A.
3. The administrator is aware of the problem with judgehost A and stops it. Domserver does not send further submissions.
4. Judgehost B, the only judgehost which can judge, judges them and the results are sent to DB.
5. After fixed, judgehost A judges them and the results are sent to DB.

Then submission.php will look like the following

http://*.com/domjudge/jury/submission.php?id=107

	ID	start	max runtime	judgehost	result	rejudging
	j107	18:56	0 s		wrong-answer
→	j187	10:56	s	ip-172---*	compiler-error

error: domjudge query error(value SELECT result FROM judging WHERE submitid = 107 AND valid = 1): Query returned too many rows(2)

SELECT * FROM judging WHERE submitid = 107 AND valid = 1

judgingidUnique ID	cidContest ID	submitidSubmission ID being judged	starttimeTime judging started	endtimeTime judging ended, null = still busy	judgehostJudgehost that performed the judging	resultResult string as defined in config.php	verifiedResult verified by jury member?	jury_memberName of jury member who verified this	verify_commentOptional additional information provided by the ve…	validOld judging is marked as invalid when rejudging	output_compileOutput of the compiling the program	seenWhether the team has seen this judging	rejudgingidRejudging ID (if rejudge)	prevjudgingidPrevious valid judging (if rejudge)
107	8	107	1502704590.673028000	1502704595.577489000	NULL	wrong-answer	0	NULL	NULL	1	[BLOB - 578 B]	0	NULL	NULL
187	8	107	1502762175.915671000	1502762176.436800000	ip-172---*	compiler-error	0	NULL	NULL	1	[BLOB - 672 B]	0	NULL	NULL

As you may see the valid field are both updated to 1.

This issue is totally regardless of the two issues I've just created before. Please note that the domjudge version is 5.1.2 and some sensitive information is marked with an asterisk.

[eldering]

Ah, this has also been fixed in current development master branch. Now when a judgedaemon is restarted after a crash (or currently more often: an internal error it reported), any previously running judging is canceled, and will be picked up anew.