[
https://issues.apache.org/jira/browse/SPARK-1647?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14110942#comment-14110942
]
Giulio De Vecchi edited comment on SPARK-1647 at 8/27/14 10:33 AM:
-------------------------------------------------------------------
Not sure if this make sense, but maybe would be nice to have a kind of "flag"
available within the code that tells me if I'm running in a "normal" situation
or during a recovery.
To better explain this, let's consider the following scenario:
I am processing data, let's say from a Kafka streaming, and I am updating a
database based on the computations. During the recovery I don't want to update
again the database (for many reasons, let's just assume that) but I want my
system to be in the same status as before, thus I would like to "know" if my
code is running for the first time or during a recovery so I can avoid to
update the database again.
More generally I want to know this in case I'm interacting with external
entities.
was (Author: gadv):
Not sure if this make sense, but maybe would be nice to have a kind of "flag"
available within the code that tells me if I'm running in a "normal" situation
or during a recovery.
To better explain this, let's consider the following scenario:
I am processing data, let's say from a Kafka streaming, and I am updating a
database based on the computations. During the recovery I don't want to update
again the database (for many reasons, let's just assume that) but I want my
system to be in the same status as before, so I would be able to "know" if my
code is running for the first time or during a recovery so I can avoid to
update the database again.
More generally I want to know this in case I'm interacting with external
entities.
> Prevent data loss when Streaming driver goes down
> -------------------------------------------------
>
> Key: SPARK-1647
> URL: https://issues.apache.org/jira/browse/SPARK-1647
> Project: Spark
> Issue Type: Bug
> Components: Streaming
> Reporter: Hari Shreedharan
> Assignee: Hari Shreedharan
>
> Currently when the driver goes down, any uncheckpointed data is lost from
> within spark. If the system from which messages are pulled can replay
> messages, the data may be available - but for some systems, like Flume this
> is not the case.
> Also, all windowing information is lost for windowing functions.
> We must persist raw data somehow, and be able to replay this data if
> required. We also must persist windowing information with the data itself.
> This will likely require quite a bit of work to complete and probably will
> have to be split into several sub-jiras.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
