hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Vladimir Rodionov <vrodio...@carrieriq.com>
Subject RE: Default value for Periodic Flusher
Date Tue, 10 Dec 2013 22:40:08 GMT
I do agree that flush interval must be configurable (I think its configurable).

> I've upgraded to 0.94.11. Here is my "worst-case scenario" :
> - let say each regionserver has 3 GB memstore
> - let say compaction max filesize is ~200 GB, min. 2 files, max 10 files.
> - let say memstore is growing "slowly" (1 GB / hour per RS)
> Then, automatically flushing every hour will lead into 1 GB storefiles,
> being compacted into storefiles of 2 GB, 3 GB, 4.... up to 200 GB.

Nope. The reality is even worse than one could imagine.

Memstore 'size' is estimated Java heap usage which includes 'object' overheads (mostly KeyValue)
For small rows, the ratio of serialized memstore (store file) and estimated heap size is close
to 5x (store file is smaller of course).
If you enable compression (2-3x) -> your store file will be 10-15 times smaller than your
Memstore (not 1 GB,  but 70-100MB)

Best regards,
Vladimir Rodionov
Principal Platform Engineer
Carrier IQ, www.carrieriq.com
e-mail: vrodionov@carrieriq.com

________________________________________
From: Adrien Mogenet [adrien.mogenet@gmail.com]
Sent: Tuesday, December 10, 2013 2:16 PM
To: user
Subject: Re: Default value for Periodic Flusher

Hi guys,

I've upgraded to 0.94.11. Here is my "worst-case scenario" :

- let say each regionserver has 3 GB memstore
- let say compaction max filesize is ~200 GB, min. 2 files, max 10 files.
- let say memstore is growing "slowly" (1 GB / hour per RS)

Then, automatically flushing every hour will lead into 1 GB storefiles,
being compacted into storefiles of 2 GB, 3 GB, 4.... up to 200 GB.
Sometimes, my write-load becomes very low, and periodic flusher will flush
perhaps 1 MB of data, it will trigger a minor compaction of hundreds
gigabytes + 1 MB; it seems to be lots of IO just to merge 1 MB of data.

Previously (ie. lack of periodic flusher) memstore was creating 3 GB
storefiles, and thus creating (after minor compactions) 3 GB, 6 GB, 9 GB...
up to 200 GB storefiles. And if memstore is growing slowly, it won't
generate small storefiles on HDFS. If think it looks like a more reasonable
IO-load, doesn't it?

I deeply agree with Periodic Flusher relevance, but I don't think it's
suitable for everyone. Do you share my opinion wrt. my workload?


On Sun, Dec 8, 2013 at 10:36 PM, Ted Yu <yuzhihong@gmail.com> wrote:


Confidentiality Notice:  The information contained in this message, including any attachments
hereto, may be confidential and is intended to be read only by the individual or entity to
whom this message is addressed. If the reader of this message is not the intended recipient
or an agent or designee of the intended recipient, please note that any review, use, disclosure
or distribution of this message or its attachments, in any form, is strictly prohibited. 
If you have received this message in error, please immediately notify the sender and/or Notifications@carrieriq.com
and delete or destroy any copy of this message and its attachments.

Mime
View raw message