As per my experience, in normal case lock wont be held for 60 seconds. How many writes/sec per node you are doing? Seems like there is some hotspotting in your use case or cluster might need some tuning/tweaking. Have you verified that your writes/reads are evenly spread out. Do u have a time component as prefix of your rowkey?
On Sun, Oct 23, 2016 at 7:01 PM, Manjeet Singh <[email protected]> wrote: > Anil its written it can hold lock upto 60 second. In my case my job get > stuck and many update for same rowkey cause fir bead health of hbase in cdh > 5.8 > > On 24 Oct 2016 06:26, "anil gupta" <[email protected]> wrote: > > Writes/Updates usually takes few milliseconds in HBase. So, in normal cases > lock wont be held for seconds. > > On Sun, Oct 23, 2016 at 12:57 PM, Manjeet Singh < > [email protected]> > wrote: > > > Anil all information are correct I am talking about suppose I didn't set > > any version and I have very simple requirement to update if I found xyz > > record and if I hv few ETL process which are responsible for aggregate > the > > data which is very common. ... why my hbase stuck if I try to update same > > rowkey... its mean its hold the lock for few second.... > > > > On 24 Oct 2016 00:46, "anil gupta" <[email protected]> wrote: > > > > > Writes within a HBase row are atomic. Now, whichever write becomes the > > > latest write(with the help of timestamp value) will prevail as the > > default > > > value. If you set versions to more than 1 in column family, then you > will > > > be able to see both the values if you query for multiple versions. > > > > > > HTH, > > > Anil Gupta > > > > > > On Sun, Oct 23, 2016 at 12:02 PM, Manjeet Singh < > > > [email protected]> > > > wrote: > > > > > > > Till now what i understand their is no update > > > > > > > > if two different thread try to update same record what happen > > > > > > > > first record insert with some version > > > > second thread comes and change the version and its like a new insert > > with > > > > some version > > > > this process called MVCC > > > > > > > > If I am correct how hbase support MVCC mean any configuration for > > > handlling > > > > multiple thread at same time? > > > > > > > > On Mon, Oct 24, 2016 at 12:24 AM, Manjeet Singh < > > > > [email protected]> > > > > wrote: > > > > > > > > > No I don't have 50 clients? I want to understand internal working > of > > > > Hbase > > > > > in my usecase I have bulk update operation from spark job we have 7 > > > > > different kafka pipeline and 7 spark job > > > > > it might happen that my 2 0r 3 spark job have same rowkey for > update > > > > > > > > > > > > > > > > > > > > On Mon, Oct 24, 2016 at 12:20 AM, Dima Spivak < > [email protected] > > > > > > > > wrote: > > > > > > > > > >> If your typical use case sees 50 clients simultaneously trying to > > > update > > > > >> the same row, then a strongly consistent data store that writes to > > > disk > > > > >> for > > > > >> fault tolerance may not be for you. That said, such a use case > seems > > > > >> extremely unusual to me and I'd ask why you're trying to update > the > > > same > > > > >> row in such a manner. > > > > >> > > > > >> On Sunday, October 23, 2016, Manjeet Singh < > > > [email protected]> > > > > >> wrote: > > > > >> > > > > >> > Hi Dima, > > > > >> > > > > > >> > I didn't get ? point is assume I have 50 different client all > > having > > > > >> same > > > > >> > rowkey all want to update on same rowkey at same time now just > > tell > > > > what > > > > >> > will happen? who will get what value? > > > > >> > > > > > >> > Thanks > > > > >> > Manjeet > > > > >> > > > > > >> > On Mon, Oct 24, 2016 at 12:12 AM, Dima Spivak < > > > [email protected] > > > > >> > <javascript:;>> wrote: > > > > >> > > > > > >> > > Unless told not to, HBase will always write to memory and > append > > > to > > > > >> the > > > > >> > WAL > > > > >> > > on disk before returning and saying the write succeeded. > That's > > by > > > > >> design > > > > >> > > and the same write pattern that companies like Apple and > > Facebook > > > > have > > > > >> > > found works for them at scale. So what's there to solve? > > > > >> > > > > > > >> > > On Sunday, October 23, 2016, Manjeet Singh < > > > > >> [email protected] > > > > >> > <javascript:;>> > > > > >> > > wrote: > > > > >> > > > > > > >> > > > Hi All, > > > > >> > > > > > > > >> > > > I have read below mention blog and it also said Hbase holds > > the > > > > >> lock on > > > > >> > > > rowkey level > > > > >> > > > https://blogs.apache.org/hbase/entry/apache_hbase_ > > > > >> > internals_locking_and > > > > >> > > > (0) Obtain Row Lock > > > > >> > > > (1) Write to Write-Ahead-Log (WAL) > > > > >> > > > (2) Update MemStore: write each cell to the memstore > > > > >> > > > (3) Release Row Lock > > > > >> > > > > > > > >> > > > > > > > >> > > > SO question is how to solve this if I have very frequent > > update > > > on > > > > >> > Hbase > > > > >> > > > > > > > >> > > > Thanks > > > > >> > > > Manjeet > > > > >> > > > > > > > >> > > > On Wed, Aug 17, 2016 at 9:54 AM, Manjeet Singh < > > > > >> > > [email protected] <javascript:;> > > > > >> > > > <javascript:;>> > > > > >> > > > wrote: > > > > >> > > > > > > > >> > > > > Hi All > > > > >> > > > > > > > > >> > > > > Can anyone help me about how and in which version of Hbase > > > > support > > > > >> > > Rowkey > > > > >> > > > > lock ? > > > > >> > > > > I have seen article about rowkey lock but it was about .94 > > > > >> version it > > > > >> > > > said > > > > >> > > > > that if row key not exist and any update request come and > > that > > > > >> rowkey > > > > >> > > not > > > > >> > > > > exist then in this case Hbase hold the lock for 60 sec. > > > > >> > > > > > > > > >> > > > > currently I am using Hbase 1.2.2 version > > > > >> > > > > > > > > >> > > > > Thanks > > > > >> > > > > Manjeet > > > > >> > > > > > > > > >> > > > > > > > > >> > > > > > > > > >> > > > > -- > > > > >> > > > > luv all > > > > >> > > > > > > > > >> > > > > > > > >> > > > > > > > >> > > > > > > > >> > > > -- > > > > >> > > > luv all > > > > >> > > > > > > > >> > > > > > > >> > > > > > > >> > > -- > > > > >> > > -Dima > > > > >> > > > > > > >> > > > > > >> > > > > > >> > > > > > >> > -- > > > > >> > luv all > > > > >> > > > > > >> > > > > >> > > > > >> -- > > > > >> -Dima > > > > >> > > > > > > > > > > > > > > > > > > > > -- > > > > > luv all > > > > > > > > > > > > > > > > > > > > > -- > > > > luv all > > > > > > > > > > > > > > > > -- > > > Thanks & Regards, > > > Anil Gupta > > > > > > > > > -- > Thanks & Regards, > Anil Gupta > -- Thanks & Regards, Anil Gupta
