[gold-users] first installation gold (problem mysql)

Scott Jackson scottmo at adaptivecomputing.com
Tue Nov 24 16:43:28 MST 2009


Andrei,

Andrei Volkau wrote:
> On 23 November 2009 18:11:33 you wrote:
>   
>> Volkau,
>>
>> This issue was addressed in gold 2.1.11.0, though it has not been
>> released yet.
>>
>> Gold changes:
>>
>> Mod Release 2.1.11.0
>>
>>   Updated init.d file to support the status subcommand
>>   Disabled g_reservation_acct_where_idx for MySQL since MySQL does not
>>     support WHERE clauses in CREATE INDEX statements.
>>
>> You should be able to simply ignore the error message.
>>
>> Sorry about the problem it has caused.
>>
>> Scott
>>
>> Andrei Volkau wrote:
>>     
>>> Greetings, all
>>>
>>> I carry out the first installation gold--2.1.10.0-2 for work
>>> With maui-3.2.6p21, torque-2.3.6. A configuration with mysql.
>>>
>>> After creation of base, I carry out a step on creation of tables and I
>>> receive an error.
>>>
>>> Further my steps:
>>>
>>> lmaster:/usr/local/gold # mysql -V
>>> mysql  Ver 14.12 Distrib 5.0.67, for suse-linux-gnu (x86_64) using
>>> readline 5.2
>>> lmaster:/usr/local/gold # mysql -u gold -p gold < bank.sql
>>> Enter password:
>>> ERROR 1064 (42000) at line 700: You have an error in your SQL syntax;
>>> check the manual that corresponds to your MySQL server version for the
>>> right syntax to use near 'WHERE g_deleted != 'True'' at line 1
>>>
>>> lmaster:/usr/local/gold # grep 'WHERE g_deleted !=' bank.sql
>>> CREATE INDEX g_reservation_acct_where_idx ON g_reservation_allocation
>>> (g_account) WHERE g_deleted != 'True';
>>>
>>> lmaster:/usr/local/gold #
>>>
>>>
>>> Error at filling base only on this line.
>>> Without this line all passes well.
>>> It is possible to create this index later?
>>>       

I do not believe it is possible to create this index in mysql. This 
index is not essential. It is simply a performance optimization.

>>> Syntax mysql for creation INDEX does not contain WHERE.
>>> It is correct?
>>>
>>> Can be Mysql any more it is supported?
>>>
>>>
>>> Volkau Andrei
>>> system administrator HPC centre of NAS of Belarus
>>> +375 29 623-31-21 ICQ 337588608 http://www.sysnet.by
>>>
>>> _______________________________________________
>>> gold-users mailing list
>>> gold-users at supercluster.org
>>> http://www.supercluster.org/mailman/listinfo/gold-users
>>>       
>
>
> Why Maui dont write about nodes for job?
> Why Maui dont DEFER with Failure answer?
>   
I will try to answer your questions the best I know how, however I am 
not a maui developer and you may want to ask these questions to the 
maui-users mailing list instead.
Nodes for a job are typically not included in the allocation manager 
statistics because this list can become unwieldly (some of our customers 
jobs literally run on tens or hundreds of thousands of nodes). I believe 
we have customers who do include this in their Gold job records. If you 
want to do this, you will need to create an extra NodeList field on your 
Job Object (see the Gold Admin Guide on how to add new fields to Gold 
objects). Then you will have to interact with Gold using prologue and 
epilogue scripts instead of the AMCFG lines (i.e. the interaction will 
be done by the resource manager instead of by Maui). If you want to do 
this, this will take some work on your part.

I really can't tell you why Maui behaves in a certain way. You'll have 
to ask this question on a maui forum or list.
> Why list jobs empty (Web Gui)?
>   
This is not clear to me. I have not seen this before. Perhaps there were 
some errors at some point with reservations or charges between Maui and 
Gold. If the maui and gold logs have not rolled, you may be able to 
learn more by an examination of these logs.

> Where store full information about jobs from maui?
>
> (Ex:
> # JobID              NRQ TRQ UserName GroupName WCLimit  JobState       Class 
> QueueTime DispatchT StartTime CompleteT   Network   Arch  Opsys MC  Mem DC   
> Disk   Features    SQTime Tas TPNd QOS Flags   Account         Command Comm 
> Byp  PSUtil Partition DProcs   DMem  DDisk  DSwap StartDate   EndDate MNode 
> RMName HostList Reservation AppType RES1 RES2
> 2205                   0   8  spectre     users     600 Completed   [batch:1] 
> 1259099148 1259099149 1259099149 1259099167    [NONE] [NONE] [NONE] >=    0M 
>   
>> =      0M    [batch] 1259099148   8    8 [NONE]:DEFAULT [NONE]    [NONE]          
>>     
> [NONE] [NONE]   0    0.00   DEFAULT      1      0M      0M      0M         0 
> 2140000000 node-17 lmaster [NONE] [NONE] [DEFAULT] [NONE] [NONE]
> )
>
>
>   
I know that Moab has the option of storing the nodelist in the events 
files, but I can't answer a question about why maui doesn't do this 
besides the possibility that it was just never engineered to capture 
this information and Adaptive Computing puts all of its engineering 
efforts into its commercial product for obvious reasons. It may be 
possible that you could succeed in getting this added as a feature 
request to through the maui-users list or whatever maui support avenue 
you might have available to you.  I'm sorry I can't be any more help 
here. I can put you in touch with someone who can discuss options for 
Maui support with you if you would like.

Thanks,

Scott

> Ex:
>
> AMCFG[lmaster] TYPE=GOLD
> AMCFG[lmaster] HOST=localhost
> AMCFG[lmaster] PORT=7112
> AMCFG[lmaster] SOCKETPROTOCOL=HTTP
> AMCFG[lmaster] WIREPROTOCOL=XML
> AMCFG[lmaster] FLUSHINTERVAL=12:00:00
> AMCFG[lmaster] APPENDMACHINENAME=TRUE
> AMCFG[lmaster] CHARGEPOLICY=DEBITALLWC
> AMCFG[lmaster] DEFERJOBONFAILURE=TRUE
> AMCFG[lmaster] TIMEOUT=00:00:00:15
>
>
>
> glsjob
>
> Id JobId User    Project  Machine Queue QualityOfService Stage   Charge 
> Processors Nodes WallDuration StartTime EndTime Description
> -- ----- ------- -------- ------- ----- ---------------- ------- ------ ---------- ----- ------------ --------- ------- -----------
> 1                                                                     0                                                     
> 2                                                                     0                                                     
> 3                                                                     0                                                     
> 4  2203  spectre q_batch  lmaster batch DEFAULT          Charge       3 24               
> 5                                  
> 5  2204  spectre q_batch  lmaster batch DEFAULT          Reserve      0 24               
> 345600                             
> 6  2205  spectre q_batch  lmaster batch DEFAULT          Charge       4 8                
> 18 
>
>
> and from log for Job 2204:
>
> grep 07.054  /var/log/gold/goldd.log
> 2009-11-24 23:46:07.054 TRACE Gold::Message::receiveChunk  invoked with 
> arguments: ()
> 2009-11-24 23:46:07.054 DEBUG Gold::Message::receiveChunk  Read message header 
> (POST /SSSRMAP3 HTTP/1.1
> 2009-11-24 23:46:07.054 INFO  Gold::Message::receiveChunk  Read message 
> payload (562, <Envelope component="ClusterScheduler" count="1" name="maui" 
> version="3.2.6p20"><Signature><DigestValue>g3mc4swCwyP5jKacqnlLdCyN9J0=</DigestValue><SignatureValue>ljsvCvY97p0mbcbXCtW9DLS6zA0=</SignatureValue></Signature><Body 
> actor="root"><Request action="Charge" 
> actor="root"><Object>Job</Object><Data><Job><JobId>2205</JobId><UserId>spectre</UserId><QualityOfService>DEFAULT</QualityOfService>
> <MachineName>lmaster</MachineName><Processors>8</Processors><WallDuration>18</WallDuration><Queue>batch</Queue><Type>job</Type></Job></Data></Request></Body></Envelope>).
>
>
>
> lmaster:/usr/local/gold/etc # grep 07.055  /var/log/gold/goldd.log
> 2009-11-24 23:46:07.055 TRACE Gold::Message::unmarshallChunk  invoked with 
> arguments: (<Envelope component="ClusterScheduler" count="1" name="maui" 
> version="3.2.6p20"><Signature><DigestValue>g3mc4swCwyP5jKacqnlLdCyN9J0=</DigestValue><SignatureValue>ljsvCvY97p0mbcbXCtW9DLS6zA0=</SignatureValue></Signature><Body 
> actor="root"><Request action="Charge" 
> actor="root"><Object>Job</Object><Data><Job><JobId>2205</JobId><UserId>spectre</UserId><QualityOfService>DEFAULT</QualityOfService><MachineName>lmaster</MachineName>
> <Processors>8</Processors><WallDuration>18</WallDuration><Queue>batch</Queue><Type>job</Type></Job></Data></Request></Body></Envelope>)
> lmaster:/usr/local/gold/etc #  
>
>
>
>
> Volkau Andrei
> system administrator HPC centre of NAS of Belarus
> +375 29 623-31-21 ICQ 337588608 http://www.sysnet.
>
>
>   



More information about the gold-users mailing list