Part #1 Part #2 Part #3. Background Engineering Oracle Rant

Transcription

1 @andy_pavlo

2 Part #1 Part #2 Part #3 Background Engineering Oracle Rant

3 AUTONOMOUS DBMSs 3 SELF-ADAPTIVE DATABASES s Self-Adaptive Databases Index Selection Partitioning / Sharding Data Placement

4 AUTONOMOUS DBMSs 3 SELF-ADAPTIVE DATABASES Admin SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' s Self-Adaptive Databases

5 AUTONOMOUS DBMSs 3 SELF-ADAPTIVE DATABASES SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' Admin Tuning Algorithm A.ID A.VAL B.ID B.NAME s Self-Adaptive Databases

6 AUTONOMOUS DBMSs 3 SELF-ADAPTIVE DATABASES SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' Admin Tuning Algorithm A.ID A.VAL B.ID B.NAME s Self-Adaptive Databases

7 AUTONOMOUS DBMSs 3 SELF-ADAPTIVE DATABASES SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' Admin Tuning Algorithm A.ID A.VAL B.ID B.NAME s Self-Adaptive Databases

8 AUTONOMOUS DBMSs 3 SELF-ADAPTIVE DATABASES SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' Admin Tuning Algorithm A.ID A.VAL B.ID B.NAME s Self-Adaptive Databases

9 AUTONOMOUS DBMSs 3 SELF-ADAPTIVE DATABASES SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' Admin Tuning Algorithm A.ID A.VAL B.ID B.NAME s Self-Adaptive Databases Index Selection Partitioning / Sharding Data Placement

10 AUTONOMOUS DBMSs 4 SELF-TUNING DATABASES SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' Admin Tuning Algorithm A.ID A.VAL B.ID B.NAME s Self-Tuning Databases Index Selection Partitioning / Sharding Data Placement

11 AUTONOMOUS DBMSs 4 SELF-TUNING DATABASES SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' s Self-Tuning Databases Admin Tuning Algorithm A.ID A.VAL B.ID B.NAME AutoAdmin Optimizer Cost Model

12 AUTONOMOUS DBMSs 4 SELF-TUNING DATABASES SELECT * FROM A JOIN B ON A.ID = B.ID WHERE A.VAL > 123 AND B.NAME LIKE 'XY%' s Self-Tuning Databases Admin Tuning Algorithm A.ID A.VAL B.ID B.NAME AutoAdmin Optimizer Cost Model

13 AUTONOMOUS DBMSs 4 SELF-TUNING DATABASES Number of Knobs s Self-Tuning Databases Knob Configuration

14 AUTONOMOUS DBMSs 5 CLOUD MANAGED DATABASES 2010s Cloud Databases

15 AUTONOMOUS DBMSs 5 CLOUD MANAGED DATABASES 2010s Cloud Databases

16 AUTONOMOUS DBMSs 5 CLOUD MANAGED DATABASES Initial Placement Tenant Migration 2010s Cloud Databases

17 Why is this previous work insufficient?

18 AUTONOMOUS DBMSs 7 A BRIEF HISTORY Problem #1 Human Judgements Problem #2 Reactionary Measures

19 What is different this time?

20 AUTONOMOUS DATABASES WHY NOW? Better hardware. Better machine learning tools. Better appreciation for data. We seek to complete the circle in autonomous databases.

21 CARNEGIE MELLON UNIVERSITY 10 RESEARCH PROJECTS OtterTune Existing Systems Peloton New System

22 OtterTune ottertune.cs.cmu.edu Database Tuning-as-a-Service Automatically generate DBMS knob configurations. Reuse data from previous tuning sessions. Supported Systems

23 OTTERTUNE 12 AUTOMATIC DBMS TUNING SERVICE CONTROLLER COLLECTOR TARGET DATABASE

24 OTTERTUNE 12 AUTOMATIC DBMS TUNING SERVICE CONTROLLER COLLECTOR TUNING MANAGER Internal Repository Configuration Recommender Metric Analyzer TARGET DATABASE Knob Analyzer

25 OTTERTUNE 12 AUTOMATIC DBMS TUNING SERVICE CONTROLLER COLLECTOR TUNING MANAGER Internal Repository Configuration Recommender Metric Analyzer TARGET DATABASE Knob Analyzer

26 OTTERTUNE 12 AUTOMATIC DBMS TUNING SERVICE CONTROLLER COLLECTOR TUNING MANAGER Internal Repository Configuration Recommender Metric Analyzer TARGET DATABASE Knob Analyzer

27 OTTERTUNE 12 AUTOMATIC DBMS TUNING SERVICE CONTROLLER COLLECTOR TUNING MANAGER Internal Repository Configuration Recommender Metric Analyzer TARGET DATABASE Knob Analyzer

28 OTTERTUNE 12 AUTOMATIC DBMS TUNING SERVICE CONTROLLER COLLECTOR TUNING MANAGER Internal Repository Configuration Recommender Metric Analyzer TARGET DATABASE Knob Analyzer

29 OTTERTUNE 12 AUTOMATIC DBMS TUNING SERVICE CONTROLLER COLLECTOR TUNING MANAGER Internal Repository Configuration Recommender Metric Analyzer TARGET DATABASE Knob Analyzer

30 OTTERTUNE 12 AUTOMATIC DBMS TUNING SERVICE CONTROLLER COLLECTOR TUNING MANAGER Internal Repository TARGET DATABASE INSTALL AGENT Configuration Recommender Metric Analyzer Knob Analyzer

31 OTTERTUNE 13 DEMO Demonstration Postgres v9.3 TPC-C Benchmark

32 OTTERTUNE TPC-C TUNING Default Scripts RDS DBA Throughput (txn/sec) OtterTune AUTOMATIC DATABASE MANAGEMENT SYSTEM TUNING THROUGH LARGE-SCALE MACHINE LEARNING SIGMOD 2017

33 Peloton pelotondb.io Self-Driving Database System In-memory DBMS with integrated ML/RL framework. Designed for autonomous operations.

34 PELOTON 16 THE SELF-DRIVING DBMS WORKLOAD HISTORY TARGET DATABASE

35 PELOTON 16 THE SELF-DRIVING DBMS WORKLOAD HISTORY TARGET DATABASE FORECAST MODELS

36 PELOTON 16 THE SELF-DRIVING DBMS "THE BRAIN" WORKLOAD HISTORY Search Tree ACTION CATALOG TARGET DATABASE FORECAST MODELS

37 PELOTON 16 THE SELF-DRIVING DBMS "THE BRAIN" WORKLOAD HISTORY Search Tree TARGET DATABASE FORECAST MODELS ACTION CATALOG ACTION SEQUENCE

38 PELOTON 16 THE SELF-DRIVING DBMS "THE BRAIN" WORKLOAD HISTORY Search Tree TARGET DATABASE FORECAST MODELS ACTION CATALOG ACTION SEQUENCE

39 PELOTON 16 THE SELF-DRIVING DBMS "THE BRAIN" WORKLOAD HISTORY Search Tree TARGET DATABASE FORECAST MODELS ACTION CATALOG ACTION SEQUENCE

40 PELOTON 16 THE SELF-DRIVING DBMS "THE BRAIN" WORKLOAD HISTORY Search Tree ACTION CATALOG??? TARGET DATABASE FORECAST MODELS ACTION SEQUENCE

41 Queries Per Hour PELOTON BUS TRACKING APP WITH ONE-HOUR HORIZON Actual Ensemble (LR+RNN) Predicted Jan 11-Jan 13-Jan 15-Jan 17-Jan QUERY-BASED WORKLOAD FORECASTING FOR SELF-DRIVING DATABASE MANAGEMENT SYSTEM SIGMOD 2018

42 Queries Per Hour Millions PELOTON ADMISSIONS APP WITH THREE-DAY HORIZON Actual Predicted Ensemble (LR+RNN) Nov 30-Nov 4-Dec 8-Dec 12-Dec 16-Dec

43 Queries Per Hour Millions PELOTON ADMISSIONS APP WITH THREE-DAY HORIZON Actual Predicted Ensemble (LR+RNN) Nov 30-Nov 4-Dec 8-Dec 12-Dec 16-Dec

44 Queries Per Hour Millions Millions PELOTON ADMISSIONS APP WITH THREE-DAY HORIZON Actual Predicted Ensemble (LR+RNN) Nov 30-Nov 4-Dec 8-Dec 12-Dec 16-Dec Hybrid (LR+RNN+KR) Nov 30-Nov 4-Dec 8-Dec 12-Dec 16-Dec

45 OTTERTUNE 19 DEMO Let's on check the demo

46 Design Considerations for Autonomous Operation

47 AUTONMOUS DBMS 21 DESIGN CONSIDERATIONS Configuration Knobs Internal Metrics Action Engineering

48 CONFIGURATION KNOBS UNTUNABLE KNOBS 22 Anything that requires a human value judgement should be marked as off-limits to autonomous components. File Paths Network Addresses Durability / Isolation Levels

49 CONFIGURATION KNOBS HOW TO CHANGE 23 The autonomous components need hints about how to change a knob Min/max ranges. Separate knobs to enable/disable a feature. Non-uniform deltas.

50 CONFIGURATION KNOBS HOW TO CHANGE 23 The autonomous components need hints about how to change a knob Min/max ranges. Separate knobs to enable/disable a feature. Non-uniform deltas. 1 KB 1 MB 1 GB 1 TB +10 KB +10 MB +10 GB

51 CONFIGURATION KNOBS HOW TO CHANGE 23 The autonomous components need hints about how to change a knob Min/max ranges. Separate knobs to enable/disable a feature. Non-uniform deltas.

52 CONFIGURATION KNOBS HARDWARE RESOURCES 24 Indicate which knobs are constrained by hardware resources. The sum of all buffers cannot exceed the total amount of available memory. The problem is that sometimes it makes sense to overprovision.

53 INTERNAL METRICS HARDWARE INFORMATION 25 Expose DBMS's hardware capabilities: CPU, Memory, Disk, Network Configuration Recommender

54 INTERNAL METRICS HARDWARE INFORMATION 25 Expose DBMS's hardware capabilities: CPU, Memory, Disk, Network Otherwise you have to come up with clever ways to approximate this Microbenchmark Threads

55 Factor 2 INTERNAL METRICS 26 HARDWARE MICROBENCHMARKS 2 vcpus 4 vcpus 8 vcpus 16 vcpus 32 vcpus Factor Analysis c3.large c3.8xlarge h1.8xlarge i3.large m3.lar r3.large i3.4xlarge h1.4xlarge d2.4xlar c3.4xlarge i2.4xlarge r3.4xlarge i3.xlarge d2.xlar c3.xlarge r3.xlarge i2.xlarge m3.xlar i3.2xlarge h1.2xlarge d2.2xlarge c3.2xlarge r3.2xlarge i2.2xlarge m3.2xlar Factor 1

56 INTERNAL METRICS SUB-COMPONENTS 27 If the DBMS has sub-components that are tunable, then it must expose separate metrics for those components. Bad Example:

57 INTERNAL METRICS 28 SUB-COMPONENTS RocksDB Column Family Knobs Column Family Metrics Missing: Reads Writes

58 INTERNAL METRICS 28 SUB-COMPONENTS RocksDB Column Family Knobs Global Metrics Aggregated Metrics

59 ACTION ENGINEERING NO SHUTDOWN 29 No action should ever require the DBMS to restart in order for it to take affect. The commercial systems are much better than this than the open-source systems.

60 ACTION ENGINEERING NOTIFICATIONS 30 Provide a notification callback to indicate when an action starts and when it completes. Harder for changes that can be used before the action completes.

61 ACTION ENGINEERING RESOURCE USAGE 31 Support executing the same action with different resource usage levels.

62 ACTION ENGINEERING REPLICA EXPLORATION 32 Allow replica configurations to diverge from each other. Master Replicas

63 ACTION ENGINEERING REPLICA EXPLORATION 32 Allow replica configurations to diverge from each other. Master Replicas

64 ACTION ENGINEERING REPLICA EXPLORATION 32 Allow replica configurations to diverge from each other. Master Replicas

65 ACTION ENGINEERING REPLICA EXPLORATION 32 Allow replica configurations to diverge from each other. Master Replicas

66 ACTION ENGINEERING REPLICA EXPLORATION 32 Allow replica configurations to diverge from each other. Master Replicas

67 What About Oracle's Self-Driving DBMS?

68 ORACLE SELF-DRIVING DBMS 34 September 2017 January 2017

69 ORACLE SELF-DRIVING DBMS 34 Automatic Indexing Automatic Recovery Automatic Scaling Automatic Query Tuning September 2017

70 ORACLE SELF-DRIVING DBMS 34 Automatic Indexing Automatic Recovery Automatic Scaling Automatic Query Tuning Problem #2 Reactionary Measures September 2017

71 ORACLE SELF-DRIVING DBMS 34 Automatic Indexing Automatic Recovery Automatic Scaling Automatic Query Tuning Problem #2 Reactionary Measures September 2017

72 ORACLE SELF-DRIVING DBMS 34 Automatic Indexing Automatic Recovery Automatic Scaling Automatic Query Tuning Problem #2 Reactionary Measures September 2017

73 CONCLUSION MAIN TAKEAWAYS 35 True autonomous DBMSs are achievable in the next decade. You should think about how each new feature can be controlled by a machine.

74 OTTERTUNE 36 DEMO Demo Results

75

76