Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Flume >> mail # user >> hdfs.idleTimeout ,what's it used for ?

Copy link to this message
Re: hdfs.idleTimeout ,what's it used for ?
I lined up why it was happening in FLUME-1850

He has hourly rolls, a 4000 interval and a 900 idle.

After an hour 400 remains on the interval. So the interval gets
triggered first, which triggers close, which cancels all timers
including the idleTimeout. Thus the entry in sfWriters remains. His
memory dump confirms this(he has a huge sfWriters map in memory after 30
days). I also confirmed this behaviour of rollInterval when developing
the idleTimeout feature.

You're right  about the limit on the size of sfWriters. With a limit of
5000, even if the closed ones stay in the list, they shouldn't be that
big  since buffers should be cleaned up.

idleTimeout will indeed result in more files if you don't have a steady
stream of files. It is most useful with a steady stream of data and time
bucketed data. In such situations, I might even recommend not using
rollInterval at all and having a short idleTimeout(or if you're not in a
rush to get your file closed, give it a comfortably long timeout)

On 01/18/2013 11:19 AM, Connor Woodson wrote:
> Whether idleTimeout is lower or higher than rollInterval is a
> preference; set it before, and assume you get one message right on the
> turn of the hour, then you will have some part of that hour without
> any bucket writers; but if you get another message at the end of the
> hour, you will end up with two files instead of one. Set it
> idleTimeout to be longer and you will get just one file, but also (at
> worst case) you will have twice as many bucketwriters open; so it all
> depends on how many files you want/how much memory you have to spare.
> - Connor
> An aside:
> bucketwriters, after being closed by rollInterval, aren't really a
> memory leak; they just are very rarely useful to keep around (your
> path could rely on hostname, and you could use a rollinterval, and
> then those bucketwriters will still remain useful). And they will get
> removed eventually; by default after you've created your 5001st
> bucketwriter, the first (or whichever was used longest ago) will be
> removed.
> And I don't think that's the cause behind 1850 as he did have an
> idleTimeout set at 15 minutes.
> On Thu, Jan 17, 2013 at 6:08 PM, Juhani Connolly
> <mailto:[EMAIL PROTECTED]>> wrote:
>     It's also useful if you want files to get promptly closed and
>     renamed from the .tmp or whatever.
>     We use it with something like 30seconds setting(we have a constant
>     stream of data) and hourly bucketing.
>     There is also the issue that files closed by rollInterval are
>     never removed from the internal linkedList so it actually causes a
>     small memory leak(which can get big in the long term if you have a
>     lot of files and hourly renames). I believe this is what is
>     causing the OOM Mohit is getting in FLUME-1850
>     So I personally would recommend using it(with a setting that will
>     close files before rollInterval does).
>     On 01/18/2013 06:38 AM, Bhaskar V. Karambelkar wrote:
>         Ah I see. Again something useful to have in the flume user guide.
>         On Thu, Jan 17, 2013 at 3:29 PM, Connor Woodson
>         <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>> wrote:
>             the rollInterval will still cause the last 01-17 file to
>             be closed
>             eventually. The way the HDFS sink works with the different
>             files is each
>             unique path is specified by a different BucketWriter
>             object. The sink can
>             hold as many objects as specified by hdfs.maxOpenWorkers
>             (default: 5000),
>             and bucketwriters are only removed when you create the
>             5001th writer (5001th
>             unique path). However, generally once a writer is closed
>             it is never used
>             again (all of your 1-17 writers will never be used again).
>             To avoid keeping