扭矩队列问题

Question

我在使用 Torque + Maui 时遇到了麻烦。

问题如下：我有 2 个队列，每个队列有 10 个关联节点。如果我向第一个队列提交 10k 个作业，向第二个队列提交 1 个作业，则第二个队列中的作业仍处于 Q 状态。

[root@localhost bin]# ./showstart 366808
job 366808 requires 2 procs for 1:00:50
Earliest start in         00:00:00 on Fri Dec 18 16:27:21
Earliest completion in     1:00:50 on Fri Dec 18 17:28:11
Best Partition: DEFAULT

[root@localhost bin]# ./checkjob -v 366808
.....
PE:  2.00  StartPriority:  3
job can run in partition DEFAULT (80 procs available.  2 procs required)

[root@localhost bin]# ./diagnose -n
Total Nodes: 20  (Active: 10  Idle: 10  Down: 0)

[root@localhost server_logs]# cat 20151218 | grep 366808
12/18/2015 16:23:47;0008;PBS_Server.44477;Job;366808.localhost.cluster;ready to commit job
12/18/2015 16:23:47;0008;PBS_Server.44477;Job;366808.localhost.cluster;ready to commit job completed
12/18/2015 16:23:47;0008;PBS_Server.44477;Job;366808.localhost.cluster;committing job
12/18/2015 16:23:47;0008;PBS_Server.44477;Job;svr_setjobstate;svr_setjobstate: setting job 366808.localhost.cluster state from TRANSIT-TRANSICM to QUEUED-QUEUED (1-10)
12/18/2015 16:23:47;0100;PBS_Server.44477;Job;366808.localhost.cluster;enqueuing into dev, state 1 hop 1
12/18/2015 16:23:47;0008;PBS_Server.44477;Job;svr_enquejob;jobs queued job id 366808.localhost.cluster for dev
12/18/2015 16:23:47;0008;PBS_Server.44477;Job;req_commit;job_id: 366808.borcluster1.cm.cluster

[root@localhost log]# cat maui.log | grep 366808
12/18 16:25:06 MPBSJobUpdate(366808,366808.localhsot.cluster,TaskList,0)
12/18 16:25:06 INFO:     job '366808' Priority:        1
12/18 16:25:06 INFO:     job '366808' Priority:        1
12/18 16:25:06 MJobPReserve(366808,DEFAULT,ResCount,ResCountRej)
12/18 16:25:06 MJobPReserve(366808,DEFAULT,ResCount,ResCountRej)

我们可以看到，这不是资源问题，因为 Maui 明确表示该作业可以执行。

仅当我使用 qrun 强制执行时，该作业才会执行