hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Simon Kelly <simongdke...@gmail.com>
Subject Re: Pre-split table using shell
Date Wed, 13 Jun 2012 07:01:31 GMT
Thanks Mike, that's pretty much the same reaction I had before. We should
be getting another 8Gb shortly but that's the limit for those servers and
while that's still not a lot I think we'll manage for now.

Unfortunately I'm not the decision maker when it comes to these things so
I'm just doing my best with what I've got.

On the point of hashing, considering that we have a 4% std deviation in the
regions do you still think we need to hash the UUID? Or is your concern
that we might get hot spotting if a set of UUID's happen to be close
together?

Thanks again for the help.
Simon

On 12 June 2012 18:09, Michael Segel <michael_segel@hotmail.com> wrote:

> ?Inferred sigh of despair?  Was it that obvious? :-)
>
> I'm not sure what hardware you're running on so its hard to say.
>
> Here's the problem... On each DN, you're running a DN and a RS. Assuming
> that you're not going to run a TT or do any M/R to push/pull data in and
> out of HBase.
> You don't have a lot of memory to play with.
>
> I guess you could make the heap size a max of 4GB...
>
> Its really tight.
>
> In general, I'd recommend at least 4GB per physical core. Some are looking
> at 8GB.
>
> The problem is that with too little memory, if you hit swap, you can cause
> a cascading failure taking down your entire instance.
>
> Sorry, I tend to be a bit paranoid and try to make the servers as robust
> as budgets allow.
>
> Getting back to your initial problem...
> Hash the keys and I think then you'll be ok.
>
>
> HTH
>
> -Mike
>
> On Jun 12, 2012, at 9:37 AM, Simon Kelly wrote:
>
> > No, this isn't on EC2 and yes, its (supposed to be) production. Please
> > elaboration on your inferred sigh of dispair....
> >
> > On 12 June 2012 15:48, Michael Segel <michael_segel@hotmail.com> wrote:
> >
> >> Ok...
> >>
> >> Please tell me that this isn't a production system.
> >>
> >> Is this on EC2?
> >>
> >> On Jun 12, 2012, at 6:55 AM, Simon Kelly wrote:
> >>
> >>> Thanks Michael
> >>>
> >>> I'm 100% sure its not the UUID distribution that's causing the problem.
> >> I'm
> >>> going to try us the API to create the table and see if that changes
> >> things.
> >>>
> >>> The reason I want to pre-split the table is that HBase doesn't handle
> the
> >>> initial load to a single regionserver and I can't start the system off
> >>> slowly and allow a few splits to happen before fully loading it. Its
> 100%
> >>> or nothing. I'm also stuck with only 8Gb of RAM per server and only 5
> >>> servers so I need to try and get as much as I can from the get go.
> >>>
> >>> Simon
> >>>
> >>> On 12 June 2012 13:37, Michael Segel <michael_segel@hotmail.com>
> wrote:
> >>>
> >>>> Ok,
> >>>> Now that I'm awake, and am drinking my first cup of joe...
> >>>>
> >>>> If you just generate UUIDs you are not going to have an even
> >> distribution.
> >>>> Nor are they going to be truly random due to how the machines are
> >>>> generating their random numbers.
> >>>> But this is not important in solving your problem....
> >>>>
> >>>> There is a set of UUIDs which are hashed and then truncated back down
> >> to a
> >>>> 128 bit string.
> >>>> You can generate the UUID, take a hash (SHA-1 or MD5) and then
> truncate
> >> it
> >>>> to 128 bits.
> >>>> This would generate a more random distribution across your splits.
> >>>>
> >>>> I'm also a bit curious about why you're pre-splitting in the first
> >> place.
> >>>> I mean I understand why people do it, but its a short term fix and I
> >>>> wonder how much pain you feel.
> >>>>
> >>>> Of course YMMV based on your use case.
> >>>>
> >>>> Hash your key and you'll be ok.
> >>>>
> >>>>
> >>>>
> >>>> On Jun 12, 2012, at 4:41 AM, Simon Kelly wrote:
> >>>>
> >>>>> Yes, I'm aware that UUID's are designed to be unique and not evenly
> >>>>> distributed but I wouldn't expect a big gap in their distribution
> >> either.
> >>>>>
> >>>>> The other thing that is really confusing me is that the regions
> splits
> >>>>> aren't lexicographical sorted. Perhaps there is a problem with the
> way
> >>>> I'm
> >>>>> specifying the splits in the split file. I haven't been able to
find
> >> any
> >>>>> docs on what format the splits keys should be in so I've used what's
> >>>>> produced by Bytes.toStringBinary. Is that correct?
> >>>>>
> >>>>> Simon
> >>>>>
> >>>>> On 12 June 2012 10:23, Michael Segel <michael_segel@hotmail.com>
> >> wrote:
> >>>>>
> >>>>>> UUIDs are unique but not necessarily random and even in random
> >>>> samplings,
> >>>>>> you may not see an even distribution except over time.
> >>>>>>
> >>>>>>
> >>>>>> Sent from my iPhone
> >>>>>>
> >>>>>> On Jun 12, 2012, at 3:18 AM, "Simon Kelly" <simongdkelly@gmail.com>
> >>>> wrote:
> >>>>>>
> >>>>>>> Hi
> >>>>>>>
> >>>>>>> I'm getting some unexpected results with a pre-split table
where
> some
> >>>> of
> >>>>>>> the regions are not getting any data.
> >>>>>>>
> >>>>>>> The table keys are UUID (generated using Java's UUID.randomUUID()
)
> >>>> which
> >>>>>>> I'm storing as a byte[16]:
> >>>>>>>
> >>>>>>> key[0-7] = uuid most significant bits
> >>>>>>> key[8-15] = uuid least significant bits
> >>>>>>>
> >>>>>>> The table is created via the shell as follows:
> >>>>>>>
> >>>>>>> create 'table', {NAME => 'cf'}, {SPLITS_FILE => 'splits.txt'}
> >>>>>>>
> >>>>>>> The splits.txt is generated using the code here:
> >>>>>>> http://pastebin.com/DAExXMDz which generates 32 regions
split
> >> between
> >>>>>> x00
> >>>>>>> and xFF. I have also tried with 16 byte regions keys (x00x00...
to
> >>>>>>> xFFxFF...).
> >>>>>>>
> >>>>>>> As far as I understand this should distribute the rows evenly
> across
> >>>> the
> >>>>>>> regions but I'm getting a bunch of regions with no rows.
I'm also
> >>>>>> confused
> >>>>>>> as the the ordering of the regions since it seems the start
and end
> >>>> keys
> >>>>>>> aren't really matching up correctly. You can see the regions
and
> the
> >>>>>>> requests they are getting here: http://pastebin.com/B4771g5X
> >>>>>>>
> >>>>>>> Thanks in advance for the help.
> >>>>>>> Simon
> >>>>>>
> >>>>
> >>>>
> >>
> >>
>
>

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