Ce diaporama a bien été signalé.
Nous utilisons votre profil LinkedIn et vos données d’activité pour vous proposer des publicités personnalisées et pertinentes. Vous pouvez changer vos préférences de publicités à tout moment.

Support Presto as a feature of SaaS

964 vues

Publié le

Treasure Data provides Presto as a feature of SaaS to the customers. While customers are able to write queries flexibly without being aware of Presto, this lack of awareness can lead to a variety of issues. We will introduce how we support users to use Presto from the perspective of our support team.

Treasure DataではPrestoをSaaSの一機能として顧客に提供をしています。顧客がPrestoを意識せずに柔軟にクエリを書ける一方で、意識しないがために様々な課題も発生します。TDではどうやってPrestoを利用するユーザサポートをしているかをサポートチームの観点から紹介します。

Publié dans : Ingénierie
  • Soyez le premier à commenter

Support Presto as a feature of SaaS

  1. 1. Support Presto as a feature of SaaS Presto Conference Tokyo 2020 November 20th, 2020 Satoru Kamikaseda Staff Technical Support Engineer, Treasure Data
  2. 2. © 2020 Treasure Data 上加世田 暁(Kamikaseda Satoru) Background ● Rakuten - Database Administrator (2009/04 ~ ) ● Treasure Data - Technical Support Engineer (2016/04 ~ ) Etc… ● Junior Football club ● Foot Golf
  3. 3. © 2020 Treasure Data Topics in this Presentation ● About Treasure Data & Support team ● Customer Inquiries ● How to support ● Frequently struggle points ● Proactive approaches ● Future ambitions
  4. 4. © 2020 Treasure Data About Treasure Data & Support team
  5. 5. © 2020 Treasure Data About Treasure Data
  6. 6. © 2020 Treasure Data About Treasure Data
  7. 7. © 2020 Treasure Data About Support Team ● Head count ○ Manager 1 ○ Japan 7 ○ USA 2 ○ Canada 1 ○ UK 1 ○ Uganda 1 ● Many components ● Focussing on Presto in this session
  8. 8. © 2020 Treasure Data Customer Inquiries
  9. 9. © 2020 Treasure Data Customer Inquiries ● Total num of inquiries ○ Around 650 / Month ○ 170 / Week
  10. 10. © 2020 Treasure Data Customer Inquiries - Percentage 2020 ● by inquiry category ○ Data Processing 26.64% ■ Presto ■ Hive ■ General SQL ■ Etc.. ○ Workflow ○ Export ○ Import ○ Etc...
  11. 11. © 2020 Treasure Data ● Ratio of query engine Customer Inquiries - Percentage 2020 ● Ratio of inquiry
  12. 12. © 2020 Treasure Data Customer Inquiries - Types 2020 ● Job Investigation - 38.46% ○ The reason of Job Failure, Result, etc... ● SQL Help - 36.11% ○ Explain SQL Syntax, Functions, Advices… ● Notification - 11.32% ○ Proactive Support ■ Incident/Job failure notification, Query tune advice, Etc…. ● Performance Issue - 11.11% ○ Query execution duration issue
  13. 13. © 2020 Treasure Data ● Cases that are difficult to resolve with support alone ○ Cases the cause cannot be identified ○ An error that's first time ○ Buggy behavior ● Aiming for 15% or less ● Roughly achieve around 8% Customer Inquiries - Escalation Rate
  14. 14. © 2020 Treasure Data How to support
  15. 15. © 2020 Treasure Data ● Accurate catch-up of the situation ● Check the actual things ● Deep investigation ● Sorting out the situation ● Answer/Report it How to support
  16. 16. © 2020 Treasure Data How to support - First of all ● Accurate catch-up of the situation ○ Free format inquiry form ○ Communication is quite important What’s happening!? The query results are wrong! Job is slow! What’s SQL? How to write?
  17. 17. © 2020 Treasure Data ● Check the actual things (sql, log, etc....) How to support - Fact check
  18. 18. © 2020 Treasure Data ● Check the actual things (sql, log, etc....) How to support - Fact check
  19. 19. © 2020 Treasure Data ● Check cluster status (DATADOG) ○ Memory, Internal Metrics (Driver, Splits, Tasks), Coordinator, Worker, Storage, etc…. How to support - Perspective
  20. 20. © 2020 Treasure Data ● Processing Cost Comparison (Splunk) ○ Elapsed, Splits, Total Bytes/Rows, Peak Memory, etc... How to support - In-depth analyses(1)
  21. 21. © 2020 Treasure Data ● Job Timeline (Splunk) ○ Job Concurrency, Memory Limitation How to support - In-depth analyses(2)
  22. 22. © 2020 Treasure Data ● Job Timeline (Splunk) ○ Job Concurrency, Memory Limitation How to support - In-depth analyses(2)
  23. 23. © 2020 Treasure Data ● Investigate as a Workflow (Splunk) ○ A single query has a small delay, but when they accumulate, it becomes a big delay. How to support - Multifaceted approach
  24. 24. © 2020 Treasure Data ● Sorting out the situation or escalate to engineering team ● Answer/Report it ○ Make a concise and understandable report How to support
  25. 25. © 2020 Treasure Data Frequently struggle points
  26. 26. © 2020 Treasure Data Frequently struggle points ● Syntax error ● Memory exceeded ○ Join order ○ Efficient use of partitions ○ Optimal Filtering ● Inefficient query ○ Multiple scans to the same table(s) ○ Improper use of CTE (Common Table Expression, WITH Statement)
  27. 27. © 2020 Treasure Data Proactive approaches
  28. 28. © 2020 Treasure Data ● Find high cost queries ○ Memory ○ Splits ○ Frequency ○ Errors ○ Others Proactive approaches
  29. 29. © 2020 Treasure Data Proactive approaches ● How get things done ○ Make a benefit for the customer ■ If no benefit (motivation), nobody will get action ○ Concrete advices ■ Solutions, not just problems, are essential ○ Best communication method ■ From Support? Customer Success? ■ By mail? Slack? Call? Meeting?
  30. 30. © 2020 Treasure Data Future ambitions
  31. 31. © 2020 Treasure Data ● Resource analysis automation ○ Automatic analysis and reporting of various factors ● Query tuning systemization ○ Detect inefficient queries and suggest specific tuning points to executors ● Performance validness monitoring ○ “Performance” is an indeterminate measure ○ However, want to embody it from the log and detect performance problems Future ambitions
  32. 32. © 2020 Treasure Data Thank You!

×