 |
 |
Well, for the xrootd configs, I guess we could get away with the master and the supervisors, the workers would then login to the master, and it would parse them out to the supervisors. But usually we just create the xrootd config once, and then put it on all the nodes and restart. We need to restart all the xrootd servers anyway. Douglas On 07/11/2013 07:02 AM, Medernach wrote: > On 07/11/2013 03:53 PM, Douglas Smith wrote: >> Ok, I'll get these test together. I checked on things, and it looks >> like the source production worked without errors, I found the >> bad nodes with the object production. So that went fine, except >> for the lost time in the fileserver load. The source loading into >> mysql is going on now, that should take ~2hours I guess. >> > Excellent news ! This is doing very well. > >> So, then we'll have the meeting and then try the final config and >> see if things can start up. >> > About config, which xrootd files need to be changed, only supervisors ? > >> Douglas >> >> >> On 07/11/2013 06:40 AM, Medernach wrote: >>> On 07/11/2013 02:23 PM, Douglas Smith wrote: >>>> Oh, yes, good point. Lets do that, I'll check on the source >>>> production and make an announcement for that. >>>> >>> About the summer planning : I will be in holydays tomorrow for one week. I will >>> be back the 22nd of July (but busy on power off/restarting our grid Tier2 this >>> day and part of the day after) and available all summer until the 20th of August. >>> >>> I hope the Qserv scale test will do well. About benchmarking queries, the >>> following links have some kind of example queries we think interesting to test : >>> >>> https://dev.lsstcorp.org/trac/wiki/db/Qserv/IN2P3/BenchmarkMarch2013 >>> https://dev.lsstcorp.org/trac/wiki/db/queries/ForPerfTest >>> >>> And also below is some query from our Petasky colleagues which takes quite a >>> long time (but it is Ok and I don't think it is a real problem) : >>> >>> select objectid as id, count(sourceid) as c >>> from Source >>> group by objectid >>> having c > 1000 >>> limit 10 ; >>> >>> Could you please set up a similar wiki page with queries and fill the results we >>> will obtain ? >>> >>> Thanks in advance, >>> -- >>> Emmanuel >>> >>>> Douglas >>>> >>>> On 07/11/2013 12:39 AM, Medernach wrote: >>>>> On 07/11/2013 03:40 AM, Douglas Smith wrote: >>>>>> Well, most of the source production is done now, but 20-30 >>>>>> machines are taking longer, and they are not done yet, although >>>>>> they appear close. >>>>>> >>>>>> I think I'm going to let this go, and do the data loading tomorrow. >>>>>> Perhaps Emmanuel can try to get the Object data loaded on the >>>>>> worker nodes. The list.txt in the master etc dir has all the current >>>>>> node that are working at this point. >>>>>> >>>>>> I'll try and be back up and online by the afternoon France time >>>>>> tomorrow. >>>>>> >>>>> By the way, do we have planned a meeting at 9h00 (Pacific time) ? >>>>> >>>>>> Douglas >>>>>> >>>>>> >>>>>> On 07/10/2013 05:24 AM, Yvan Calas wrote: >>>>>>> On Jul 10, 2013, at 2:22 PM, Douglas Smith <[log in to unmask]> wrote: >>>>>>> >>>>>>>> Well, there is also the fact that the cmsd isn't going to work >>>>>>>> correctly until we define the supervisors also. The master can >>>>>>>> only manage 63 workers, and the rest will just cause it problems. >>>>>>>> >>>>>>>> Maybe we should shut down the cmsd until the data has been >>>>>>>> produced on the nodes. >>>>>>> I killed the cmsd process few minutes ago. >>>>>>> >>>>>>> - ######################################################################## Use REPLY-ALL to reply to list To unsubscribe from the QSERV-L list, click the following link: https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=QSERV-L&A=1