Nos últimos 30 anos temos vivido a hegemonia dos bancos de dados relacionais, a grande bala de prata da TI. O armazenamento de dados se tornou tão comoditizado, que nem mesmo nos questionamos se o modelo relacional é adequado as nossas necessidades. Mas será que o armazenamento de dados se resume ao modelo relacional? Será que as técnicas tradicionais de normalização ou ferramentas de produtividade como ORM são realmente adequadas? Será que você está tratando seus dados com a devida atenção?
Nesta palestra iremos responder estas e outras perguntas sobre tratamento e armazenamento de dados. Vamos colocar o "dedo na ferida" e apresentar uma nova escola de pensamento e algumas ferramentas que suportam esta nova realidade.
Tags: Anti-patterns, Arquitetura, Database, noSQL, newSQL
18. modelos
• Hierarchical (IMS): late 1960’s and 1970’s
• Directed graph (CODASYL): 1970’s
• Relational: 1970’s and early 1980’s
• Entity-Relationship: 1970’s
• Extended Relational: 1980’s
• Semantic: late 1970’s and 1980’s
• Object-oriented: late 1980’s and early 1990’s
• Object-relational: late 1980’s and early 1990’s
• Semi-structured (XML): late 1990’s to late 2000’s
• The next big thing: ???
ref: What Goes Around Comes Around por Michael Stonebraker e Joey Hellerstein
40. Anti Patterns
• Evolution from SQL Anti Patterns (NoSQL:br May 2010)
• More than just RDBMS
• Large volumes of data
• Distribution
• Architecture
• Research on other tools
• Message Queues, DHT, Job Schedulers, NoSQL
• Indexing, Map/Reduce
• New revision since QConSP 2010: included Hierarchical
Sharding, Embedded lists and Distributed Global Locking
41. RDBMS Anti Patterns
Not all things fit on a relational database, single ou distributed
• The eternal table-as-a-tree
• Dynamic table creation
• Table as cache, queue, log file
• Stoned Procedures
• Row Alignment
• Extreme JOINs
• Your scheme must be printed in an A3 sheet.
• Your ORM issue full queries for Dataset iterations
• Hierarchical Sharding
• Embedded lists
• Distributed global locking
• Throttle Control
42. The eternal tree
Problem: Most threaded discussion example uses something
like a table which contains all threads and answers, relating to
each other by an id. Usually the developer will come up with
his own binary-tree version to manage this mess.
id - parent_id -author - text
1 - 0 - gleicon - hello world
2 - 1 - elvis - shout !
Alternative: Document storage:
{ thread_id:1, title: 'the meeting', author: 'gleicon', replies:[
{
'author': elvis, text:'shout', replies:[{...}]
}
]
}
43. Dynamic table creation
Problem: To avoid huge tables, one must come with a
"dynamic schema". For example, lets think about a document
management company, which is adding new facilities over the
country. For each storage facility, a new table is created:
item_id - row - column - stuff
1 - 10 - 20 - cat food
2 - 12 - 32 - trout
Now you have to come up with "dynamic queries", which will
probably query a "central storage" table and issue a huge join
to check if you have enough cat food over the country.
Alternatives:
- Document storage, modeling a facility as a document
- Key/Value, modeling each facility as a SET
44. Table as cache
Problem: Complex queries demand that a result be stored in a
separated table, so it can be queried quickly. Worst than views
Alternatives:
- Really ?
- Memcached
- Redis + AOF + EXPIRE
- De-normalization
45. Table as queue
Problem: A table which holds messages to be completed.
Worse, they must be ordered by
time of creation.
Corolary: Job Scheduler table
Alternatives:
- RestMQ, Resque
- Any other message broker
- Redis (LISTS - LPUSH + RPOP)
- Use the right tool
46. Table as log file
Problem: A table in which data gets written as a log file. From
time to time it needs to be purged. Truncating this table once
a day usually is the first task assigned to new DBAs.
Alternative:
- MongoDB capped collection
- Redis, and RRD pattern
- RIAK
47. Stoned procedures
Problem: Stored procedures hold most of your applications
logic. Also, some triggers are used to - well - trigger important
data events.
SP and triggers has the magic property of vanishing of our
memories and being impossible to keep versioned.
Alternative:
- Now be careful so you dont use map/reduce as modern
stoned procedures. Unfit for real time search/processing
- Use your preferred language for business stuff, and let event
handling to pub/sub or message queues.
48. Row Alignment
Problem: Extra rows are created but not used, just in case.
Usually they are named as a1, a2, a3, a4 and called padding.
There's good will behind that, specially when version 1 of the
software needed an extra column in a 150M lines database
and it took 2 days to run an ALTER TABLE. But that's no
excuse.
Alternative:
- Quit being cheap. Quit feeling 'hacker' about padding
- Document based databases as MongoDB and CouchDB, has
no schema. New atributes are local to the document and can
be added easily.
49. Extreme JOINs
Problem: Business stuff modeled as tables. Table inheritance
(Product -> SubProduct_A). To find the complete data for a
user plan, one must issue gigantic queries with lots of JOINs.
Alternative:
- Document storage, as MongoDB
might help having important
information together.
- De-normalization
- Serialized objects
50. Your scheme fits in an A3 sheet
Problem: Huge data schemes are difficult to manage. Extreme
specialization creates tables which converges to key/value
model. The normal form get priority over common sense.
Product_A Product_B
id - desc id - desc
Alternatives:
- De-normalization
- Another scheme ?
- Document store for flattening model
- Key/Value
- See 'Extreme JOINs'
51. Your ORM ...
Problem: Your ORM issue full queries for dataset iterations,
your ORM maps and creates tables which mimics your
classes, even the inheritance, and the performance is bad
because the queries are huge, etc, etc
Alternative:
- Apart from denormalization and good old common sense,
ORMs are trying to bridge two things with distinct impedance.
- There is nothing to relational models which maps cleanly to
classes and objects. Not even the basic unit which is the
domain(set) of each column. Black Magic ?
52. Hierarchical Sharding
Problem: Genius at work. Distinct databases inside a RDBMS
ranging from A to Z, each database has tables for users
starting with the proper letter. Each table has user data.
Fictional example: e-mail accounts management
> show databases;
abcdefghijklmnopqrstuwxz
> use a
> show tables;
...alberto alice alma ... (and a lot more)
There is no way to query anything in common for all users with
out application side processing. In this particular case this
sharding was uncalled for as relational databases have all
tools to deal with this particular case of 'different clients and
data'
53. Embedded Lists
Problem: As data complexity grows, one thinks that it's proper
for the application to handle different data structures
embedded in a single cell or row. The popular 'Let's use
commas to separe it'. You may find distinct separators as |, -, []
and so on.
> select group_field from that_email_sharded_database.user
"a@email1.net, b@email1.net,c@email2.net"
> select flags from stupid_email_admin where id = 2
1|0|1|1|0|
Either learn to model your data, or resort to model keys on K/V
stores. Or any other way to show up flags, as you are not
programming in C over a RDBMS hopefully.
54. Distributed Global Locking
Problem: Trying to emulate JAVA's synchronize in a distributed
manner. As there is no primitive architectura block to do that,
sounds like the proper place to do that would be a RDBMS.
May starts with a reference counter in a table and end up with
this:
> select COALESCE(GET_LOCK('my_lock',0 ),0 )
Plain and simple, you might find it embedded in a magic class
called DistributedSynchronize or ClusterSemaphore. Locks,
transactions and reference counters (which may act as soft
locks) doesn't belongs to the database. While its they use is
questionable even in code, the matter of fact is that you are
doing it wrong, if you are doing like that.
55. Throttle Control
Problem: To control and track access to a given resource, a
sequence of statements is issued, varying from a
update...select to a transaction block using a stored procedure:
> select count(access) from throttle_ctl where ip_addr like ...
> update .... or begin ... commit
Apart from having IP addresses stored as string, each request
would have to check on this block. It gets worse if throttle
control is mixed with a table-based access.
Using memcached (or any other k/v) as the data is ephemeral
would work as (after creating the entry and setting expire
time):
if (add 'IPADDR:YYYY:MM:DD:HH', 1) < your_limit: do_stuff()
77. Meaningful data
• Email traffic accounting ~4500 msgs/sec in, ~2000 msgs
out
• 10 Gb SPAM/NOT SPAM tokenized data per day
• 300 Gb logfile data/day
• 80 Billion DNS queries per day
• ~1k req/sec tcptable queries.
• 0.8 Pb of data migrated over mixed quality network.
Planned 3mo, executed 6mo, online, on production.
• traffic from 400mb/s to 3.1Gb/s
78. Stuff to think about
Think if the data you use aren't de-normalized somewhere
(cached)
Most of the anti-patterns signals that there are architectural
issues instead of only database issues.
Call it NoSQL, non relational, or any other name, but assemble
a toolbox to deal with different data types.
Are you dependent on cache ? Does your application fails
when there is no warm cache ? Does it just slows down ?
Think about the way to put and to get back your data from the
database (be it SQL or NoSQL). Migrations are painful.
79. Stuff to think about
- Without operational requirements, it's a personal choice
about data organization
- Without time pressure, any migration is easy, regardless data
size.
- Out of production environment, events flow in a different time
flow
- 'Normal Accidents' (Charles Perrow) - how resilient is your
operation ? Are you ready to tackle your incidents ?
- Everything breaks under scale (Benjamin Black)
- "Picking up pennies in front of a steamroller" (Nassin N.
Taleb)