Streaming SQL (at FlinkForward, Berlin, 2016/09/12)

Streaming SQL
Julian Hyde
FlinkForward
Berlin
2016/09/12

@julianhyde
SQL
Query planning
Query federation
OLAP
Streaming
Hadoop
Apache member
VP Apache Calcite
PMC Apache Arrow, Drill, Kylin
Thanks:
● Milinda Pathirage & Yi Pan (Apache Samza)
● Haohui Mai (Apache Storm)
● Fabian Hueske & Stephan Ewen (Apache Flink)

Data center
Streaming data sources
Sources:
● Devices / sensors
● Web servers
● (Micro-)services
● Databases (CDC)
● Synthetic streams
● Logging / tracing
Transports:
● Kafka
● Nifi
IoT
Devices
Services DatabaseWeb
server

How much is your data worth?
Recent data is more valuable
➢ ...if you act on it in time
Data moves from expensive
memory to cheaper disk as it cools
Old + new data is more valuable
still
➢ ...if we have a means to
combine them Time
Value of
data
($/GB)
Now1 hour
ago
1 day
ago
1 week
ago
1 year
ago
Hot data
Read often
Likely to be modified
High value
In memory
Cold data
Read rarely
Unlikely to be modified
Low value
On disk

Why query streams?
Stream - Database Duality:
● “Your database is just a cache of my stream”
● “Your stream is just change-capture of my database”
“Data is the new oil”
● Treating events/messages as data allows you to extract and refine them
Declarative approach to streaming applications

Why SQL? ● API to your database
● Ask for what you want,
system decides how to get it
● Query planner (optimizer)
converts logical queries to
physical plans
● Mathematically sound
language (relational algebra)
● For all data, not just data in a
database
● Opportunity for novel data
organizations & algorithms
● Standard
https://www.flickr.com/photos/pere/523019984/ (CC BY-NC-SA 2.0)
➢ API to your database
➢ Ask for what you want,
system decides how to get it
➢ Query planner (optimizer)
converts logical queries to
physical plans
➢ Mathematically sound
language (relational algebra)
➢ For all data, not just “flat”
data in a database
➢ Opportunity for novel data
organizations & algorithms
➢ Standard
Why SQL?

Data workloads
● Batch
● Transaction processing
● Single-record lookup
● Search
● Interactive / OLAP
● Exploration / profiling
● Continuous execution generating alerts (CEP)
● Continuous load
A variety of workloads, requiring specialized engines, but to the user it’s all “just
data”.

Building a streaming SQL standard via
consensus
Please! No more “SQL-like” languages!
Key technologies are open source (many are Apache projects)
Calcite is providing leadership: developing example queries, TCK
(Optional) Use Calcite’s framework to build a streaming SQL parser/planner for
your engine
Several projects are working with us: Flink, Samza, Storm, Apex. (Also
non-streaming SQL in Cassandra, Drill, Druid, Elasticsearch, Flink, Hive, Kylin,
Phoenix.)

Simple queries
select *
from Products
where unitPrice < 20
select stream *
from Orders
where units > 1000
➢ Traditional (non-streaming)
➢ Products is a table
➢ Retrieves records from -∞ to now
➢ Streaming
➢ Orders is a stream
➢ Retrieves records from now to +∞
➢ Query never terminates

Stream-table duality
select *
from Orders
where units > 1000
➢ Yes, you can use a stream as
a table
➢ And you can use a table as a
stream
➢ Actually, Orders is both
➢ Use the stream keyword
➢ Where to actually find the
data? That’s up to the system
select stream *
from Orders
where units > 1000

Combining past and future
select stream *
from Orders as o
where units > (
select avg(units)
from Orders as h
where h.productId = o.productId
and h.rowtime > o.rowtime - interval ‘1’ year)
➢ Orders is used as both stream and table
➢ System determines where to find the records
➢ Query is invalid if records are not available

Semantics of streaming queries
The replay principle:
A streaming query produces the same result as the corresponding
non-streaming query would if given the same data in a table.
Output must not rely on implicit information (arrival order, arrival time,
processing time, or watermarks/punctuations)
(Some triggering schemes allow records to be emitted early and re-stated if
incorrect.)

Making progress
It’s not enough to get the right result. We
need to give the right result at the right
time.
Ways to make progress without
compromising safety:
➢ Monotonic columns (e.g. rowtime)
and expressions (e.g.
floor(rowtime to hour))
➢ Punctuations (aka watermarks)
➢ Or a combination of both
select stream productId,
count(*) as c
from Orders
group by productId;
ERROR: Streaming aggregation
requires at least one
monotonic expression in
GROUP BY clause

8
75
4
10:00 10:15 10:30 11:00 11:15
Arrival
time
1
2
3 5
6
Event
time 8
10:00 10:15 10:30 11:00 11:15
Arrival
time
1
2
3
6
Event
time
4 Drop out-of-sequence
records
Emit 10:00-11:00 window
when first record after
11:00 arrives
W 11:00
Emit 10:00-11:00
window when
11:00 watermark
arrives
W 11:00’
7
New
watermark.
Re-state
10:00-11:00
window
Policies for emitting results
Monotonic column Watermark

Controlling when data is emitted
Early emission is the defining
characteristic of a streaming query.
The emit clause is a SQL extension
inspired by Apache Beam’s “trigger”
notion. (Still experimental… and
evolving.)
A relational (non-streaming) query is
just a query with the most conservative
possible emission strategy.
select stream productId,
count(*) as c
from Orders
group by productId,
floor(rowtime to hour)
emit at watermark,
early interval ‘2’ minute,
late limit 1;
select *
from Orders
emit when complete;

Aggregation and windows
on streams
GROUP BY aggregates multiple rows into sub-totals
➢ In regular GROUP BY each row contributes to
exactly one sub-total
➢ In multi-GROUP BY (e.g. HOP, GROUPING
SETS) a row can contribute to more than one
sub-total
Window functions (OVER) leave the number of rows
unchanged, but compute extra expressions for
each row (based on neighboring rows)
Multi
GROUP BY
Window
functions
GROUP BY

GROUP BY select stream productId,
floor(rowtime to hour) as rowtime,
sum(units) as u,
count(*) as c
from Orders
group by productId,
floor(rowtime to hour)
rowtime productId units
09:12 100 5
09:25 130 10
09:59 100 3
10:00 100 19
11:05 130 20
rowtime productId u c
09:00 100 8 2
09:00 130 10 1
10:00 100 19 1
not emitted yet; waiting
for a row >= 12:00

Window types
Tumbling
window
“Every T seconds, emit the total for T seconds”
Hopping
window
“Every T seconds, emit the total for T2 seconds”
Session
window
“Emit groups of records that are separated by gaps of no
more than T seconds”
Sliding
window
“Every record, emit the total for the surrounding T
seconds”
“Every record, emit the total for the surrounding R records”

Tumbling, hopping & session windows in SQL
Tumbling window
Hopping window
Session window
select stream … from Orders
group by floor(rowtime to hour)
group by tumble(rowtime, interval ‘1’ hour)
group by hop(rowtime, interval ‘1’ hour,
interval ‘2’ hour)
group by session(rowtime, interval ‘1’ hour)

Sliding windows in SQL
select stream
sum(units) over w (partition by productId) as units1hp,
sum(units) over w as units1h,
rowtime, productId, units
from Orders
window w as (order by rowtime range interval ‘1’ hour preceding)
rowtime productId units
09:12 100 5
09:25 130 10
09:59 100 3
10:17 100 10
units1hp units1h rowtime productId units
5 5 09:12 100 5
10 15 09:25 130 10
8 18 09:59 100 3
23 13 10:17 100 10

The “pie chart” problem
➢ Task: Write a web page summarizing
orders over the last hour
➢ Problem: The Orders stream only
contains the current few records
➢ Solution: Materialize short-term history
Orders over the last hour
Beer
48%
Cheese
30%
Wine
22%
select productId, count(*)
from Orders
where rowtime > current_timestamp - interval ‘1’ hour
group by productId

Join stream to a table
Inputs are the Orders stream and the
Products table, output is a stream.
Acts as a “lookup”.
Execute by caching the table in a
hash-map (if table is not too large) and
stream order will be preserved.
What if Products table is being
modified while query executes?
select stream *
from Orders as o
join Products as p
on o.productId = p.productId

Join stream to a changing table
Execution is more difficult if the
Products table is being changed
while the query executes.
To do things properly (e.g. to get the
same results when we re-play the
data), we’d need temporal database
semantics.
(Sometimes doing things properly is
too expensive.)
select stream *
from Orders as o
join Products as p
and o.rowtime
between p.startEffectiveDate
and p.endEffectiveDate

Join stream to a stream
We can join streams if the join
condition forces them into “lock
step”, within a window (in this case,
1 hour).
Which stream to put input a hash
table? It depends on relative rates,
outer joins, and how we’d like the
output sorted.
select stream *
from Orders as o
join Shipments as s
and s.rowtime
between o.rowtime
and o.rowtime + interval ‘1’ hour

Planning queries
MySQL
Splunk
join
Key: productId
group
Key: productName
Agg: count
filter
Condition:
action = 'purchase'
sort
Key: c desc
scan
scan
Table: products
select p.productName, count(*) as c
from splunk.splunk as s
join mysql.products as p
on s.productId = p.productId
where s.action = 'purchase'
group by p.productName
order by c desc
Table: splunk

Optimized query
MySQL
Splunk
join
Key: productId
group
Key: productName
Agg: count
filter
Condition:
action = 'purchase'
sort
Key: c desc
scan
scan
Table: splunk
Table: products
select p.productName, count(*) as c
from splunk.splunk as s
join mysql.products as p
on s.productId = p.productId
where s.action = 'purchase'
group by p.productName
order by c desc

Apache Calcite
Apache top-level project since October, 2015
Query planning framework
➢ Relational algebra, rewrite rules
➢ Cost model & statistics
➢ Federation via adapters
➢ Extensible
Packaging
➢ Library
➢ Optional SQL parser, JDBC server
➢ Community-authored rules, adapters
Embedded Adapters Streaming
Apache Drill
Apache Hive
Apache Kylin
Apache Phoenix*
Cascading
Lingual
Apache
Cassandra
Apache Spark
CSV
Druid*
Elasticsearch*
In-memory
JDBC
JSON
MongoDB
Splunk
Web tables
Apache Flink*
Apache Samza
Apache Storm
* Under development

Join the community!
Calcite and Flink are projects of the Apache
Software Foundation
The Apache Way: meritocracy, openness,
consensus, community
We welcome new contributors!

Architecture
Conventional database Calcite

Relational algebra (plus streaming)
Core operators:
➢ Scan
➢ Filter
➢ Project
➢ Join
➢ Sort
➢ Aggregate
➢ Union
➢ Values
Streaming operators:
➢ Delta (converts relation to
stream)
➢ Chi (converts stream to
relation)
In SQL, the STREAM keyword
signifies Delta

Streaming algebra
➢ Filter
➢ Route
➢ Partition
➢ Round-robin
➢ Queue
➢ Aggregate
➢ Merge
➢ Store
➢ Replay
➢ Sort
➢ Lookup

Optimizing streaming queries
The usual relational transformations still apply: push filters and projects towards
sources, eliminate empty inputs, etc.
The transformations for delta are mostly simple:
➢ Delta(Filter(r, predicate)) → Filter(Delta(r), predicate)
➢ Delta(Project(r, e0, ...)) → Project(Delta(r), e0, …)
➢ Delta(Union(r0, r1), ALL) → Union(Delta(r0), Delta(r1))
But not always:
➢ Delta(Join(r0, r1, predicate)) → Union(Join(r0, Delta(r1)), Join(Delta(r0), r1)
➢ Delta(Scan(aTable)) → Empty

Other operations
Other relational operations make sense on streams (usually only if there is an
implicit time bound).
Examples:
● order by - E.g. Each hour emit the top 10 selling products
● union - E.g. Merge streams of orders and shipments
● insert, update, delete - E.g. Continuously insert into an external table
● exists, in sub-queries - E.g. Show me shipments of products for which
there has been no order in the last hour
● view - Expanded when query is parsed; zero runtime cost

Summary
Features of streaming SQL:
● Standard SQL over streams and relations
● Relational queries on streams, and vice versa
● Materialized views and standing queries
Benefits:
● Brings streaming data to DB tools and traditional users
● Brings historic data to message-oriented applications
● Lets the system optimize quality of service (QoS) and data location

Thank you!
@julianhyde
@ApacheCalcite
http://calcite.apache.org
http://calcite.apache.org/docs/stream.html
References
● Hyde, Julian. "Data in flight." Communications of the ACM 53.1
(2010): 48-52. [pdf]
● Akidau, Tyler, et al. "The dataflow model: a practical approach to
balancing correctness, latency, and cost in massive-scale,
unbounded, out-of-order data processing." Proceedings of the
VLDB Endowment 8.12 (2015): 1792-1803. [pdf]
● Arasu, Arvind, Shivnath Babu, and Jennifer Widom. "The CQL
continuous query language: semantic foundations and query
execution." The VLDB Journal—The International Journal on Very
Large Data Bases 15.2 (2006): 121-142. [pdf]

Streaming SQL (at FlinkForward, Berlin, 2016/09/12)

More Related Content

What's hot (20)

Viewers also liked (13)

Similar to Streaming SQL (at FlinkForward, Berlin, 2016/09/12) (20)

More from Julian Hyde (16)

Recently uploaded (20)

Streaming SQL (at FlinkForward, Berlin, 2016/09/12)