[
https://issues.apache.org/jira/browse/PIG-2923?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dmitriy V. Ryaboy updated PIG-2923:
-----------------------------------
Attachment: bagspill_delayed_register.patch
This patch introduces the following changes:
1) Make SpillableMemoryManager a real singleton (before, it was assumed to be a
singleton, but the assumption was not enforced)
2) deprecates BagFactory's registerBag() method, as it is no longer a Factory's
job to do this. Also removes calls to this method from the default
implementation.
3) gets rid of some redundant code for addAll() methods in various databag
implementations
4) introduces a markSpillableIfNecessary method in AbstractDataBag, which will
register with the SMM if necessary (upon reaching a memory threshold of 100K).
5) adds markSpillableIfNecessary calls to all spillable DataBags' add() methods.
TestDataBag passes, test-commit passes, TestDataBagAccess passes.
While testing this, I discovered the fun fact that calling _size() on a
DistinctDataBag causes it to stop being distinct_. But that's for a different
jira...
> Lazily register bags with SpillableMemoryManager
> ------------------------------------------------
>
> Key: PIG-2923
> URL: https://issues.apache.org/jira/browse/PIG-2923
> Project: Pig
> Issue Type: Improvement
> Reporter: Dmitriy V. Ryaboy
> Assignee: Dmitriy V. Ryaboy
> Attachments: bagspill_delayed_register.patch
>
>
> Currently, all Spillable DataBags get registered by the BagFactory at the
> moment of creation. In practice, a lot of these bags will not get large
> enough to be worth spilling; we can avoid a lot of memory overhead and
> cheapen the process of finding a bag to spill when we do need it, by allowing
> Bags themselves to register when they grow to some respectable threshold.
> Related JIRAs: PIG-2917, PIG-2918
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
