roller-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Anil Gangolli (JIRA)" <j...@apache.org>
Subject [jira] Commented: (ROL-1760) Scheduled tasks do not run on Oracle (or other DBs with high precision timestamps)
Date Fri, 28 Nov 2008 23:37:19 GMT

    [ https://issues.apache.org/roller/browse/ROL-1760?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14461#action_14461
] 

Anil Gangolli commented on ROL-1760:
------------------------------------


Ceri is correct.  The whole point of this functionality is to allow multiple callers on one
or more hosts to attempt to run the same task (only one should succeed).    The WHERE clause
is there to prevent a caller from claiming the lease while it is still held by another one.
 The issue is the equality comparison and granularity of the stored value.

> Scheduled tasks do not run on Oracle (or other DBs with high precision timestamps)
> ----------------------------------------------------------------------------------
>
>                 Key: ROL-1760
>                 URL: https://issues.apache.org/roller/browse/ROL-1760
>             Project: Roller
>          Issue Type: Bug
>          Components: Database Access & Data Model
>    Affects Versions: 4.0
>         Environment: Solaris 10, Oracle 10gR2, Roller 4.0.1
>            Reporter: Dick Davies
>            Assignee: Roller Unassigned
>
> Rollers JPA config seems to assume that CURRENT_TIMESTAMP is low precision 
> (i.e. hundredths of a second).
> Specifically, I'm not sure the ORM config in TaskLock.orm.xml is correct.
> On databases that have microsecond resolution (Oracle, PostgreSQL, etc.) the 
> JPAThreadManagerImpl.registerLease() method quickly gets into a state where the 
> "TaskLock.updateClient&Timeacquired&Timeleased&LastRunByName&Timeacquired"
NamedUpdate never suceeds,
> and so tasks never run.
> (There was a similar issue with DB2 that resulted in us dumping that , so it's not all
bad :) )
> I turned on JPA debugging and on Oracle 10, we see 
> DEBUG 2008-11-20 15:30:00,064 CommonsLogFactory$LogAdapter:trace - Executing query: [UPDATE
TaskLock t SET t.clientId=?1, t.timeAquired= CURRENT_TIMESTAMP, t.timeLeased= ?2, t.lastRun=
?3 WHERE t.name=?4 AND t.timeAquired=?5 AND ?6 < CURRENT_TIMESTAMP] with parameters: {3=Thu
Aug 07 16:07:00 BST 2008, 2=PingQueueTask, 1=2008-08-07 16:37:00.07, 6=2008-11-20 15:30:00.0,
5=30, 4=devel-roller01}
> DEBUG 2008-11-20 15:30:00,065 CommonsLogFactory$LogAdapter:trace - <t 28169327, conn
12589755> executing prepstmnt 11980159 UPDATE roller_tasklock t0 SET timeleased = ?, client
= ?, lastrun = ?, timeacquired = CURRENT_TIMESTAMP WHERE (t0.name = ? AND t0.timeacquired
= ? AND ? < CURRENT_TIMESTAMP) [params=(int) 30, (String) devel-roller01, (Timestamp) 2008-11-20
15:30:00.0, (String) PingQueueTask, (Timestamp) 2008-08-07 16:07:00.07, (Timestamp) 2008-08-07
16:37:00.07]
>   
> TIMEACQUIRED at that time was '07-AUG-08 04.07.00.069896 PM', and since Oracle checks
down to the microsecond
> when comparing timestamps, this means the WHERE clause never matches and nothing updates.
> The WHERE clause seems overly picky in any case -  
> the schema creation scripts declare the roller_tasklock.name column to be unique, so
'WHERE to.name = ?' is enough
> to find the right row, surely?
> Another fix is to declare the timestamp columns as timestamp(2) in the Oracle 
> (+DB2 + PostgreSQL ? ) creation scripts. We're trying that now to see if it helps.
> Do you want a bug filed?
>  

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message