asterixdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chris Hillery <chill...@hillery.land>
Subject Re: REST API question
Date Mon, 10 Aug 2015 21:32:49 GMT
That's a good question. At first glance it appears the answer is "no",
unfortunately. I'm not intimately familiar with that code, but it looks
like it might be possible to support it. It would be easy enough to have a
flag that told the insert (and other DMLs) to return immediately with some
kind of key. Then I think it would be possible to have another "is this job
done yet" HTTP endpoint that accepted that key, the way that queryResults
does today. That would allow the user to poll for the response.

Ceej
aka Chris Hillery

On Mon, Aug 10, 2015 at 2:18 PM, Yingyi Bu <buyingyi@gmail.com> wrote:

> >>I would suggest that for long-running queries, using the deferred-results
> >>APIs makes more sense than counting on a TCP connection to stay up.
>
> That makes sense to me.
> Is the option also available for DMLs (inserts and deletes) which do not
> have any
> returned results?
> A user may want to how long a "long running" DML takes and whether it
> succeeds or not.
>
> Best,
> Yingyi
>
>
>
> On Mon, Aug 10, 2015 at 2:06 PM, Chris Hillery <chillery@hillery.land>
> wrote:
>
> > My first guess would be that Jetty (in the AsterixDB server) is timing
> out
> > the HTTP connection as idle. The default timeout is either 5 minutes or
> 30
> > seconds, depending on the Jetty version. We can configure that, but I'm
> not
> > sure having it set to multiple hours is a reasonable out-of-the-box
> > experience.
> >
> > I would suggest that for long-running queries, using the deferred-results
> > APIs makes more sense than counting on a TCP connection to stay up.
> >
> > Ceej
> > aka Chris Hillery
> >
> > On Mon, Aug 10, 2015 at 1:09 PM, Yingyi Bu <buyingyi@gmail.com> wrote:
> >
> > > Right, probably it is a timing issue --- but I'm not sure it's a server
> > > side problem or a client side problem..
> > >
> > > Best,
> > > Yingyi
> > >
> > > On Mon, Aug 10, 2015 at 12:56 PM, Ian Maxon <imaxon@uci.edu> wrote:
> > >
> > > > I've had this happen before as well, it's annoying to have to either
> > > scour
> > > > the logs or look in the Hyracks adminconsole to see what really
> > happened.
> > > > Is it possible that we are timing out the HTTP connection
> incorrectly?
> > > >
> > > > - Ian
> > > >
> > > > On Mon, Aug 10, 2015 at 11:24 AM, Yingyi Bu <yingyib@ics.uci.edu>
> > wrote:
> > > >
> > > > > Hi guys,
> > > > >
> > > > >     I ran an insert DDL from the REST API.  From the cc/nc logs,
> it
> > > > looks
> > > > > the insert job finished without any exceptions in about 1.5 hours.
> > > > Also, a
> > > > > simple count aggregation on the target dataset of the insert
> > statement
> > > > > returns the correct result.  However, my HTTP client program hangs
> > > > > forever.  Smaller insert jobs do not have that problem.  Does
> anyone
> > > have
> > > > > similar experience or know what's going on?
> > > > >     Thanks!
> > > > >
> > > > > Best,
> > > > > Yingyi
> > > > >
> > > >
> > >
> >
>

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