Conquering the Lambda architecture in LinkedIn metrics
platform with Apache Calcite and Apache Samza
​Khai Tran
​Staff Software Engineer
Agenda
● Overview of LinkedIn metrics platform
● Moving from offline to nearline
● Under the hood of the nearline architecture
● Nearline production usecase
● Conclusion
Overview of LinkedIn metrics platform
Metrics @ LinkedIn
● Metrics = Measurements over tracking data
● Crucial for decision making:
○ Experimentation - test everything
○ Reporting - monitor and alert
○ In production, site-facing applications
We provide:
● A trusted repository of metrics
● A self-serve platform for
sustainable lifecycle of metrics
In
production
Experimentation
Reporting
Primary Data
Unified
Metrics
Platform
LinkedIn unified metrics platform (UMP)
Growth of UMP Metrics
2016 20172015
6800
4680
1100
Current: 8000+ metrics
# code
LOAD …
# data
# transformation
# code
STORE …
# config
Metrics:
- A = SUM(A’)
- B = Unique(id)
Downstream:
- XLNT
- Raptor
UMP
User Code
Platform
Generated
Code
To
App
To
App
DefineDeclare
Onboard
Data
Metadata
Onboarding process
User
Moving from offline to nearline
Offline computation flows
Hourly job latency: 3-6 hours -> want realtime/nearline
......
Metric union
User code
User code
Cubing, Rollup
Dimension
augmentation
HDFS tables
Dali views
Pinot,
Presto
Azkaban execution
Espresso,
Oracle,
MySQL
...
What we want for nearline flows
Metric unionUser code
User code
Samza job
Dimension
augmentation
Pinot
Latency is not the only requirement
Easy to onboard ● Minimum effort to convert existing offline into nearline
● Easy to write user code for new nearline flows
Easy to maintain ● Just one version of user code - single source of truth
● Run as a service
Latency ● ~5 - 30 mins
Samza jobs
Putting things together
Pinot
Batch
jobs
UMP realtime platform
UMP offline platform
HDFS
Raptor
Lambda architecture with a single codebase
code configMetrics
definition
Current support
User code in Pig ● LOAD, STORE
● FILTER, SAMPLE, SPLIT, UNION
● Simple FOREACH
● JOIN - all semantics
● GROUP/COGROUP, DISTINCT
● Record/Array FLATTEN
● Java UDFs, Python UDFs
● Pig Nested FOREACH and sort/limit (in Windows)
● Hive
Not yet
Under the hood of the nearline architecture
Pig to Samza through SQL processing
Open source framework for building dynamic
data management systems. Including:
➢ SQL Parser
➢ Relational algebra APIs
➢ Query planning engine
We built UMP nearline with:
➢ Pig’s Grunt parser
➢ Calcite relational algebra
➢ Calcite query planning engine
Architecture
...
Metric union
User code
User code
Dimension
augmentation
Calcite relational
algebra as an IR
convert generate
Samza code
optimize
Samza
physical plan
Samza
configuration
Pig to Calcite Calcite to Samza
Pig to Calcite
# code
LOAD …
LOAD ...
COGROUP ...
STORE …
GruntParser
CO-
GROUP
LOAD LOAD
PigRelConverter
FULL
OUTERJ
OIN
AGGRE
GATE
AGGRE
GATE
TABLE
SCAN
TABLE
SCAN
PRO-
JECT
User scripts Pig Logical Plan Calcite relational algebra
Example
Example
Example
INNER
JOIN
FILTER FILTER
PROJECT PROJECT
PROJECT
TABLE
SCAN
TABLE
SCAN
Calcite logical plan
Planning/Optimization
➢ Calcite logical plans:
○ Relational algebra: What to do
➢ Samza physical plans:
○ Samza physical node: How to do it
➢ Calcite Samza planner:
○ Calcite logical plan -> optimized Samza physical plan
Example
Stream-
Stream Self
Join
Samza
Project
Samza
Project
Samza
Filter
Samza
Filter
Samza
Project
Input
Stream
INNER
JOIN
FILTER FILTER
PROJECT PROJECT
PROJECT
TABLE
SCAN
TABLE
SCAN
Calcite Samza
planner
Calcite logical plan Samza physical plan
Code-gen
From Samza physical plans:
➢ Generate Samza code for constructing the stream graph using Samza Fluent APIs .
Mapping:
➢ Samza physical nodes -> corresponding stream APIs:
○ Samza project -> stream.map()
○ Samza filter -> stream.filter()
○ ...
➢ Relational expressions -> lamba functions:
○ Filter expressions -> filter() functions
○ Project expressions -> map() functions
○ ...
Example
Schema and UDF declarations
Operator mapping
Filter functions
Map functions
Produce to Kafka
...
Config-gen
Stream
Stream
Join
Samza
Project
Samza
Project
Samza
Filter
Samza
Filter
Samza
Project
Input
Stream
# dataset.conf
app-src
app-def
Nearline production use case - Storylines
Top stories picked
up by editors
Feedback to editor - powered by UMP realtime
Conclusion
Samza jobs
From improved Lambda architecture...
Pinot
Batch
jobs
UMP realtime platform
UMP offline platform
HDFS
Raptor
Lambda architecture with a single codebase
code configMetrics
definition
… to our bigger picture
Pig Latin
Calcite
relational
algebra
HiveQL
SparkSQL/
RDD
Presto SQL
Portable
UDFs
AORA (Author Once, Run Anywhere) architecture
Conquering the Lambda architecture in LinkedIn metrics platform with Apache Calcite and Apache Samza

Conquering the Lambda architecture in LinkedIn metrics platform with Apache Calcite and Apache Samza

  • 1.
    Conquering the Lambdaarchitecture in LinkedIn metrics platform with Apache Calcite and Apache Samza ​Khai Tran ​Staff Software Engineer
  • 2.
    Agenda ● Overview ofLinkedIn metrics platform ● Moving from offline to nearline ● Under the hood of the nearline architecture ● Nearline production usecase ● Conclusion
  • 3.
    Overview of LinkedInmetrics platform
  • 4.
    Metrics @ LinkedIn ●Metrics = Measurements over tracking data ● Crucial for decision making: ○ Experimentation - test everything ○ Reporting - monitor and alert ○ In production, site-facing applications
  • 5.
    We provide: ● Atrusted repository of metrics ● A self-serve platform for sustainable lifecycle of metrics In production Experimentation Reporting Primary Data Unified Metrics Platform LinkedIn unified metrics platform (UMP)
  • 6.
    Growth of UMPMetrics 2016 20172015 6800 4680 1100 Current: 8000+ metrics
  • 7.
    # code LOAD … #data # transformation # code STORE … # config Metrics: - A = SUM(A’) - B = Unique(id) Downstream: - XLNT - Raptor UMP User Code Platform Generated Code To App To App DefineDeclare Onboard Data Metadata Onboarding process User
  • 8.
  • 9.
    Offline computation flows Hourlyjob latency: 3-6 hours -> want realtime/nearline ...... Metric union User code User code Cubing, Rollup Dimension augmentation HDFS tables Dali views Pinot, Presto Azkaban execution Espresso, Oracle, MySQL
  • 10.
    ... What we wantfor nearline flows Metric unionUser code User code Samza job Dimension augmentation Pinot
  • 11.
    Latency is notthe only requirement Easy to onboard ● Minimum effort to convert existing offline into nearline ● Easy to write user code for new nearline flows Easy to maintain ● Just one version of user code - single source of truth ● Run as a service Latency ● ~5 - 30 mins
  • 12.
    Samza jobs Putting thingstogether Pinot Batch jobs UMP realtime platform UMP offline platform HDFS Raptor Lambda architecture with a single codebase code configMetrics definition
  • 13.
    Current support User codein Pig ● LOAD, STORE ● FILTER, SAMPLE, SPLIT, UNION ● Simple FOREACH ● JOIN - all semantics ● GROUP/COGROUP, DISTINCT ● Record/Array FLATTEN ● Java UDFs, Python UDFs ● Pig Nested FOREACH and sort/limit (in Windows) ● Hive Not yet
  • 14.
    Under the hoodof the nearline architecture
  • 15.
    Pig to Samzathrough SQL processing Open source framework for building dynamic data management systems. Including: ➢ SQL Parser ➢ Relational algebra APIs ➢ Query planning engine We built UMP nearline with: ➢ Pig’s Grunt parser ➢ Calcite relational algebra ➢ Calcite query planning engine
  • 16.
    Architecture ... Metric union User code Usercode Dimension augmentation Calcite relational algebra as an IR convert generate Samza code optimize Samza physical plan Samza configuration Pig to Calcite Calcite to Samza
  • 17.
    Pig to Calcite #code LOAD … LOAD ... COGROUP ... STORE … GruntParser CO- GROUP LOAD LOAD PigRelConverter FULL OUTERJ OIN AGGRE GATE AGGRE GATE TABLE SCAN TABLE SCAN PRO- JECT User scripts Pig Logical Plan Calcite relational algebra
  • 18.
  • 19.
  • 20.
  • 21.
    Planning/Optimization ➢ Calcite logicalplans: ○ Relational algebra: What to do ➢ Samza physical plans: ○ Samza physical node: How to do it ➢ Calcite Samza planner: ○ Calcite logical plan -> optimized Samza physical plan
  • 22.
    Example Stream- Stream Self Join Samza Project Samza Project Samza Filter Samza Filter Samza Project Input Stream INNER JOIN FILTER FILTER PROJECTPROJECT PROJECT TABLE SCAN TABLE SCAN Calcite Samza planner Calcite logical plan Samza physical plan
  • 23.
    Code-gen From Samza physicalplans: ➢ Generate Samza code for constructing the stream graph using Samza Fluent APIs . Mapping: ➢ Samza physical nodes -> corresponding stream APIs: ○ Samza project -> stream.map() ○ Samza filter -> stream.filter() ○ ... ➢ Relational expressions -> lamba functions: ○ Filter expressions -> filter() functions ○ Project expressions -> map() functions ○ ...
  • 24.
    Example Schema and UDFdeclarations Operator mapping Filter functions Map functions Produce to Kafka ...
  • 25.
  • 26.
    Nearline production usecase - Storylines
  • 27.
  • 28.
    Feedback to editor- powered by UMP realtime
  • 29.
  • 30.
    Samza jobs From improvedLambda architecture... Pinot Batch jobs UMP realtime platform UMP offline platform HDFS Raptor Lambda architecture with a single codebase code configMetrics definition
  • 31.
    … to ourbigger picture Pig Latin Calcite relational algebra HiveQL SparkSQL/ RDD Presto SQL Portable UDFs AORA (Author Once, Run Anywhere) architecture