Hi Kenny,

So that looks pretty good.  It looks like it solved your problem, correct?
 If so, we'll probably change the block cache default to disabled.

- Doug

On Tue, Aug 21, 2012 at 12:44 AM, Kenny F. <[email protected]> wrote:

> Hi Doug,
>
> Here some results:
> RangeServer works for about 24 hours without crash.
>
> "Block Cache Fill", "Block Cache Hit Rate", "Block Cache Max Memory" are
> zero.
>
> "Virtual Memory" slowly rose up, but it still under the "cap" level, so my
> be it fall down but a bit later.
> At the picture we see restart ~10AM.
>
> Or you already found the reason of crash.)
>
> Regards,
> Kenny F.
>
>
>
> On Friday, August 17, 2012 7:58:05 PM UTC+3, Doug Judd wrote:
>
>> Hi Kenny,
>>
>> I think I may know what's going on here.  Can you try your test again
>> with the following changes:
>>
>> 1. Revert the Hypertable.RangeServer.**MemoryLimit changes.  In other
>> words, leave the RangeServer memory properties at their defaults (60% of
>> physical RAM).
>> 2. Add the following property to your hypertable.cfg to disable the block
>> cache:
>>
>> Hypertable.RangeServer.**BlockCache.MaxMemory=0
>>
>> I suspect that the RangeServer is fighting with the OS file cache for
>> RAM.  Disabling the block cache is not as bad as you might think.  The OS
>> will cache the compressed blocks in its file cache.
>>
>> - Doug
>>
>> On Fri, Aug 17, 2012 at 7:24 AM, Kenny F. <[email protected]> wrote:
>>
>>> ))
>>> I hope this issue help you to locate the problem.
>>> Hypertable's RangeServers use ALL memory(not all FREE memory) when have
>>> no memory limit (or high memory limit).
>>> It is easy to watch "Block Cache Max Memory" at RangeServer Statistics
>>> (monitoring).
>>> This issue appeared between Hypertable 0.9.5.0 and Hypertable 0.9.5.4.
>>> I hope this bug will be fix.
>>>
>>>
>>> On Thursday, August 16, 2012 4:19:53 PM UTC+3, Christoph Rupp wrote:
>>>
>>>> Well, by default the RangeServer uses 60%, but that memory is used for
>>>> caching CellStores.
>>>>
>>>> If there's additional memory required to fetch data for a single
>>>> request, then it can exceed the limit for a short while. It seems that this
>>>> might have happened here, If you have very large scans with lots of data
>>>> (and in addition some of the memory was allocated by apache).
>>>>
>>>> But i am glad that we worked it out :)
>>>>
>>>> bye
>>>> Christoph
>>>>
>>>> 2012/8/16 Kenny F. <[email protected]>
>>>>
>>>> Thanks!
>>>>>
>>>>> I've already done it, it helped and post graphs (today) for you to
>>>>> interpret them.
>>>>> Earlier I used Hypertable.RangeServer.**MemoryL****imit.Percentage=20
>>>>> or 30 or 40 and it didn't help me at all.
>>>>> Now I use Hypertable.RangeServer.**MemoryL****imit=1900Mb (see my
>>>>> post today with graphs)
>>>>>
>>>>> I  think, the reason is when we have no limit or high limit,
>>>>> RangeServer use ALL memory (see graph "Block Cache Max Memory"), even if 
>>>>> it
>>>>> is fill at 95% or more.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> On Thursday, August 16, 2012 4:04:31 PM UTC+3, Christoph Rupp wrote:
>>>>>
>>>>>> Hi Kenny,
>>>>>>
>>>>>> can you try reducing the memory usage of the RangeServer? By default
>>>>>> the RangeServer uses 60% of the memory.
>>>>>>
>>>>>> There are two properties:
>>>>>>
>>>>>> Hypertable.RangeServer.**MemoryL****imit or Hypertable.RangeServer.**
>>>>>> MemoryL****imit.Percentage.
>>>>>>
>>>>>> You can use either of them.
>>>>>>
>>>>>> Bye
>>>>>> Christoph
>>>>>>
>>>>>> 2012/8/16 Kenny F. <[email protected]>
>>>>>>
>>>>>> >Is it possible that there is some other process (or set of
>>>>>>> processes) on the machine at the time of the crash that is using up all
>>>>>>> virtual memory?
>>>>>>> I don't think so.
>>>>>>> First, I have 2 memory greed processes only: Hypertable and Apache.
>>>>>>> Second, I didn't change the environment. And restarting the system
>>>>>>> didn't solve the issue.
>>>>>>> Third, RangeServer didn't live more then several hours.
>>>>>>> And I don't remember, when RangeServer crashes immediately.
>>>>>>>
>>>>>>>
>>>>>>> On Thursday, August 16, 2012 3:02:39 PM UTC+3, Doug Judd wrote:
>>>>>>>
>>>>>>>> Hi Kenny,
>>>>>>>>
>>>>>>>> One other thought.  Is it possible that there is some other process
>>>>>>>> (or set of processes) on the machine at the time of the crash that is 
>>>>>>>> using
>>>>>>>> up all virtual memory?  Try running 'free' at regular intervals during 
>>>>>>>> your
>>>>>>>> test to see if the system is running out of swap space.
>>>>>>>>
>>>>>>>> - Doug
>>>>>>>>
>>>>>>>>
>>>>>>  --
>>>>> You received this message because you are subscribed to the Google
>>>>> Groups "Hypertable Development" group.
>>>>> To view this discussion on the web visit https://groups.google.com/d/*
>>>>> *ms**g/hypertable-dev/-/**CWqXo6VjsGQ**J<https://groups.google.com/d/msg/hypertable-dev/-/CWqXo6VjsGQJ>
>>>>> .
>>>>> To post to this group, send email to hyperta...@googlegroups.**com.
>>>>> To unsubscribe from this group, send email to hypertable-de...@**
>>>>> googlegroups.**com.
>>>>>
>>>>> For more options, visit this group at http://groups.google.com/**group
>>>>> **/hypertable-dev?hl=en<http://groups.google.com/group/hypertable-dev?hl=en>
>>>>> .
>>>>>
>>>>
>>>>  --
>>> You received this message because you are subscribed to the Google
>>> Groups "Hypertable Development" group.
>>> To view this discussion on the web visit https://groups.google.com/d/**
>>> msg/hypertable-dev/-/**WDrfBDUZHZMJ<https://groups.google.com/d/msg/hypertable-dev/-/WDrfBDUZHZMJ>
>>> .
>>>
>>> To post to this group, send email to hyperta...@googlegroups.**com.
>>> To unsubscribe from this group, send email to hypertable-de...@**
>>> googlegroups.com.
>>> For more options, visit this group at http://groups.google.com/**
>>> group/hypertable-dev?hl=en<http://groups.google.com/group/hypertable-dev?hl=en>
>>> .
>>>
>>
>>
>>
>> --
>> Doug Judd
>> CEO, Hypertable Inc.
>>
>>   --
> You received this message because you are subscribed to the Google Groups
> "Hypertable Development" group.
> To view this discussion on the web visit
> https://groups.google.com/d/msg/hypertable-dev/-/5aqMdL8wbTIJ.
>
> To post to this group, send email to [email protected].
> To unsubscribe from this group, send email to
> [email protected].
> For more options, visit this group at
> http://groups.google.com/group/hypertable-dev?hl=en.
>



-- 
Doug Judd
CEO, Hypertable Inc.

-- 
You received this message because you are subscribed to the Google Groups 
"Hypertable Development" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/hypertable-dev?hl=en.

Reply via email to