Re: Counting Provile Visits: Keys vs. Hashtables

[Josiah Carlson]

On Tue, Oct 30, 2012 at 1:19 AM, Dominik Kukacka
<dominik...@gmail.com> wrote:
> Hey,
>
> should I count Profile Visits in Keys or Hashtables?
>
> profile_visits:{{ID}}:YYYMMDDHH
> profile_visits:{{ID}}:YYYMMDD
> profile_visits:{{ID}}:YYYMM
> profile_visits:{{ID}}:YYY
> profile_visits:{{ID}}:total
>
> or.
>
> profile:visits:{{ID}}
> YYYYMMDDHH
> YYYYMMDD
> YYYYMM
> YYYY
> total
>
> Which is more size efficient?
> Which scales better?
> Which is more logical?

Do you ever want to time out your data? If so, you may want to
structure your data such that you can have the data clean itself up
with EXPIRE. Once you have answered that question, then you may want
to consider that short hashes (up to 512 items by default) can use
less space than the equivalent higher-level key. So maybe batch
together a single day or week's worth of hourly data (depending on
your expiration policy), one month's or one year's worth of daily data
(depending on your expiration policy), and maybe keep all of your
monthly/yearly data in a single hash for a given profile.

Regards,
- Josiah

[Salvatore Sanfilippo]

On Thu, Nov 1, 2012 at 4:28 PM, Josiah Carlson <josiah....@gmail.com> wrote:

>> Which is more size efficient?
>> Which scales better?
>> Which is more logical?
>
> Do you ever want to time out your data? If so, you may want to
> structure your data such that you can have the data clean itself up
> with EXPIRE. Once you have answered that question, then you may want
> to consider that short hashes (up to 512 items by default) can use
> less space than the equivalent higher-level key. So maybe batch
> together a single day or week's worth of hourly data (depending on
> your expiration policy), one month's or one year's worth of daily data
> (depending on your expiration policy), and maybe keep all of your
> monthly/yearly data in a single hash for a given profile.

Agreed, so I would go for:

>> profile:visits:{{ID}}
>> YYYYMMDDHH
>> YYYYMMDD
>> YYYYMM
>> YYYY
>> total

YYYYMMDDHH -> hash key named YYYYMMDD, filed HH
YYYYMMDD -> hash with key named YYYYMM, field DD
YYYYMM -> hash with key named YYYY, field MM
YYYY -> hash with key named YYYY (so you reuse the previous hash),
field named "year_total"
total -> put this counter inside the user-specific hash directly to
avoid wasting a singel key for it.

In this way all the hashes should have less tha 512 fields, but you
may want to double check as I'm sleep deprived right now :-)

Cheers,
Salvatore

--
Salvatore 'antirez' Sanfilippo
open source developer - VMware
http://invece.org

Beauty is more important in computing than anywhere else in technology
because software is so complicated. Beauty is the ultimate defence
against complexity.
— David Gelernter