apex-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Yan <da...@datatorrent.com>
Subject Re: END_STREAM tuple at recovery
Date Tue, 08 Dec 2015 01:37:42 GMT
I traced to this piece of code in WindowIdActivatedReservoir.java below.
Chetan, please let me know if you recall the idea behind this.  The commit
message of the EndStreamTuple constructor call says something about making
a recovery test pass.

@Override
public Tuple sweep()
{
  Tuple t;
  while ((t = reservoir.sweep()) != null) {
    if (t.getType() == MessageType.BEGIN_WINDOW && t.getWindowId() >
windowId) {
      reservoir.setSink(sink);
      return (est = new EndStreamTuple(windowId));
    }
    reservoir.remove();
  }

  return null;
}

On Mon, Dec 7, 2015 at 3:02 PM, David Yan <david@datatorrent.com> wrote:

> Anybody have any idea?  Is it safe to treat the END_STREAM tuple just as
> RESET_TUPLE at recovery from a checkpoint?
>
> David
>
> On Fri, Dec 4, 2015 at 10:34 AM, David Yan <david@datatorrent.com> wrote:
>
>> Hi all,
>>
>> At recovery from a checkpoint, I see END_STREAM control tuple as the
>> first tuple sent from upstream.  That seems a little counter intuitive
>> because a stream is actually starting.  Can someone shed some light on this?
>>
>> Thanks!
>>
>> David
>>
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message