Like I said, I haven't used nested contexts, but I'm assuming that
data in the parent context is available to the child contexts, but
data in sibling contexts is not available to the child context.

As for not nesting contexts too deep, I'm guessing that's a
performance issue rather than a technical limitation.  This process
doesn't sound like it's an interactive time-sensitive operation, and
you've already said that memory isn't an issue (and data contexts are
not memory intensive in any case).

Years ago, before Cayenne had child contexts, I handled situations
like yours by using a single context and a hashmap.
As I created new objects, I stored them in the hash map by the line
identifier, then pulled them back out later when I had a line that
referred to a previous line.


On Thu, Mar 29, 2018 at 5:50 PM, Ken Anderson
<[email protected]> wrote:
> Mike,
>
> The docs say to not go too deep with nested contexts, so I don't think that's 
> a viable solution.  I'm also not clear what benefit you think this would have 
> over the 2 level plan.
>
> Ken
>
> On 3/29/18, 5:39 PM, "Mike Kienenberger" <[email protected]> wrote:
>
>     Instead of having each line as a child context under a parent context,
>
>     parent
>     - child1 (line 1)
>     - child2 (line 2)
>     - child3 (line 3)
>
>     could you have each line processed as a child context of the previous 
> line?
>
>     - child1 (line 1)
>     -- child2 (line 2)
>     --- child3 (line 3)
>     ---- child4 (line 4)
>
>     If you want to "undo" the current line,
>     child4.rollbackChangesLocally(); and start on the next line.
>
>     Note that I have not done anything with child contexts, but this would
>     be how I'd try to solve it.
>
>
>     On Wed, Mar 28, 2018 at 9:29 PM, Ken Anderson <[email protected]> wrote:
>     > Hugi,
>     >
>     > That’s correct - it’s not like we’re just importing a bunch of records. 
>  Each row in the file could affect the same set of objects.
>     >
>     > So, we did the child context, but obviously if we created an object in 
> a prior child and then saved it to the parent, we won’t be able to easily 
> find it in the next child context.  If you get a localObject in your new 
> child context, it is “hollow”, so not connected to all the other objects 
> floating around in the parent.  We also can’t fire relationships, because 
> those relationships will go to the database instead of the parent context.
>     >
>     > Ken
>     >
>     >> On Mar 28, 2018, at 7:11 PM, Hugi Thordarson <[email protected]> wrote:
>     >>
>     >>> That's exactly what we want to do - save once at the end.  However, 
> we have 2 problems:
>     >>>
>     >>> 1. How do we find the objects that we already created but haven't 
> saved yet
>     >>
>     >> You can go through yur ObjectContext's newObjects() and filter that to 
> your liking—pretty much the same as you'd do with EOF.
>     >>
>     >>> 2. How do we roll back each line if there's an error?  Not a DB 
> error, but the logic gets so far, and then determines that there's no way to 
> continue so we must skip this line.
>     >>
>     >> As you tried yourself, I'd use a child context and commit to the 
> parent once you're sure everything is in place. Can you explain further what 
> was problematic with that (that you need to "access the same objects multiple 
> times")? Do you mean that each row of the file is in some way looking at data 
> from other rows?
>     >>
>     >> Cheers,
>     >> - hugi
>     >>
>     >>
>     >>>
>     >>> Ken
>     >>>
>     >>> On 3/28/18, 6:07 PM, "John Huss" <[email protected]> wrote:
>     >>>
>     >>>   Well, you could just save once at the end.  Why do you need to save
>     >>>   multiple times during the processing?  Validation exceptions and 
> Optimistic
>     >>>   Locking errors could be handled in the save with some custom logic 
> and a
>     >>>   retry.
>     >>>
>     >>>   Or if this isn't a super long process you can use a database 
> transaction to
>     >>>   allow saving multiple times without actually having that data be 
> visible
>     >>>   outside of the transaction.
>     >>>
>     >>>
>     >>>   On Wed, Mar 28, 2018 at 6:56 AM Ken Anderson 
> <[email protected]>
>     >>>   wrote:
>     >>>
>     >>>> All,
>     >>>>
>     >>>> We have a process that reads in a file and, for each line, creates or
>     >>>> edits objects in the object graph.  We only want to commit to the 
> database
>     >>>> once at the end.
>     >>>>
>     >>>> We have a finite set of lines, so memory is not an issue.  We need 
> to save
>     >>>> only once because saving will actually fire triggers that will start 
> doing
>     >>>> other things to the database, which will then lead to optimistic lock
>     >>>> exceptions for us if we have data that overlaps (which we do).
>     >>>>
>     >>>> Please don’t suggest we change how the trigger pattern works – it’s 
> a big
>     >>>> system and we don’t have control over it.
>     >>>>
>     >>>> So, what we’ve toyed with is using a parent/child context 
> arrangement,
>     >>>> where each line is processed in a child, and assuming everything 
> goes OK,
>     >>>> we commit only to the parent.  This works well as long as we don’t 
> need to
>     >>>> access the same objects multiple times, but unfortunately, we do.  
> We can
>     >>>> reach into the parent context’s unsaved objects, but those objects 
> do not
>     >>>> have any relationships since they were built in the child context.  
> This
>     >>>> makes things painful.
>     >>>>
>     >>>> In EOF, I might consider using a single context and undo, but it 
> doesn’t
>     >>>> seem like Cayenne has this kind of functionality.
>     >>>>
>     >>>> Thoughts?  Suggestions?  In EOF, I had once written a layer that
>     >>>> intercepted all queries and tried to find the correct object in 
> unsaved
>     >>>> objects, but I don’t have nearly enough experience with Cayenne to 
> do that.
>     >>>>
>     >>>> Thanks!
>     >>>> Ken
>     >>>>
>     >>>> Confidentiality Notice: This e-mail and accompanying documents 
> contain
>     >>>> confidential information intended for a specific individual and 
> purpose.
>     >>>> This e-mailed information is private and protected by law. If you 
> are not
>     >>>> the intended recipient, you are hereby notified that any disclosure,
>     >>>> copying, or distribution, or the taking of any action based on the 
> contents
>     >>>> of this information, is strictly prohibited.
>     >>>>
>     >>>
>     >>>
>     >>> Confidentiality Notice: This e-mail and accompanying documents 
> contain confidential information intended for a specific individual and 
> purpose. This e-mailed information is private and protected by law. If you 
> are not the intended recipient, you are hereby notified that any disclosure, 
> copying, or distribution, or the taking of any action based on the contents 
> of this information, is strictly prohibited.
>     >>
>     >
>
>
> Confidentiality Notice: This e-mail and accompanying documents contain 
> confidential information intended for a specific individual and purpose. This 
> e-mailed information is private and protected by law. If you are not the 
> intended recipient, you are hereby notified that any disclosure, copying, or 
> distribution, or the taking of any action based on the contents of this 
> information, is strictly prohibited.

Reply via email to