Peter Zaitsev "18 ways to fix MySQL bottlenecks"
Download as PPTX, PDF0 likes217 views
The document discusses strategies for managing traffic spikes in MySQL, emphasizing immediate actions such as scaling cloud instances, deploying additional MySQL slaves, and optimizing queries. It highlights best practices for resource management, including checking for resource bottlenecks, adding or optimizing indexes, and configuring MySQL settings. Overall, it offers practical solutions to enhance system performance without overhauling infrastructure.
Peter Zaitsev "18 ways to fix MySQL bottlenecks"
- 1. © 2020 Percona 1
- 2. Focus on the Traffic Spike Different causes of slowdown have different approaches, we focus on Traffic Spike in this Presentation © 2020 Percona 3
- 3. Something you can do NOW Focus on Changes which you can implement in Hours or Days, not weeks or Months © 2020 Percona 4
- 5. #1 Scale your Cloud Instance Size © 2020 Percona6 https://aws.amazon.com/ec2/instance-types/
- 6. Scaling Instance Size when Money are not concern © 2020 Percona7
- 7. Best Practice © 2020 Percona 8 Scaling by the Credit Card Use As Temporary Solution While you find more cost effective options Or until Traffic Spike Passes
- 8. #2 Deploy More MySQL Slaves © 2020 Percona9
- 9. Read Intensive Applications © 2020 Percona 10 Works best for Read Intensive Applications Still needs massive amount of data copies (you pay for) Caching not very efficient due to many copies in memory Changing Application to get advantage of Replicas may be hard
- 10. #3 Deploy ProxySQL © 2020 Percona 11
- 11. ProxySQL Considerations © 2020 Percona 12 Great for Connection Pooling Can assist with Read-Write Splitting Supports TTL (Time to Live) based Query Caching Helps to improve High Availability with Failover Handling
- 12. Why NOT ProxySQL © 2020 Percona 13 ProxySQL is NOT 100% MySQL Compatible Some application Adjustments may need be required ThreadPool in Percona Server for MySQL, MariaDB Connection Pool in your Application Server (Java etc)
- 13. #4 Disable Heavy Load Application Features © 2020 Percona 14 Great Application Design Practice – allow disabling features easier What Are high load and low business value features ? Can be applied to specific users too
- 14. #5 Check for Resource Bottlenecks © 2020 Percona15 https://per.co.na/PMM
- 15. Primary Raw Resources of a System © 2020 Percona16
- 16. #6 Get More CPU Cores, or Faster Cores © 2020 Percona17
- 17. Look at Run Queue Latency © 2020 Percona18 Tool from BCC Tools Collection https://github.com/iovisor/bcc
- 18. Beware of Stealing © 2020 Percona19
- 20. #7 Get More Memory © 2020 Percona21
- 21. Do not Run out of Virtual Memory © 2020 Percona 22
- 22. Beware Serious Swapping © 2020 Percona 23
- 23. True Memory Usage by Process © 2020 Percona24 Look at Resident + Swap
- 24. #8 Move to Faster Storage © 2020 Percona 25 Memory is faster than fastest storage Fast Storage for Write Performance Fast Storage for Read Performance, when data set is so large fitting in memory is impractical
- 25. Amount and Type of IO Activity © 2020 Percona 26
- 26. How Latency Compares to the Historical © 2020 Percona 27
- 27. “Fsync” Is Special kind of IO © 2020 Percona 28
- 28. Biolatency: Block Device Latency © 2020 Percona29 Tool from BCC Tools Collection https://github.com/iovisor/bcc
- 29. #9 Check your Network © 2020 Percona 30 Bandwidth Latency Packet Loss
- 30. How Much Network Traffic do you Have ? © 2020 Percona 31
- 32. Retransmits – The silent Killer © 2020 Percona 33
- 33. Where Retransmits come from © 2020 Percona34 Tool from BCC Tools Collection https://github.com/iovisor/bcc
- 34. #10 Locate and Optimize Queries which cause the load © 2020 Percona35
- 35. Query Execution Details © 2020 Percona36
- 36. Check Explain © 2020 Percona 37
- 37. #11 Add Missing Indexes © 2020 Percona 38 Missing Index can cause query to run 1000s times slower than it should Most Applications Have Some Missing Indexes Index Purpose: Lookup Data, Sorting, Serving The Data
- 38. #12 Drop Unneeded Indexes © 2020 Percona 39 Indexes are Expensive both for Writes and for Reads You only should have indexes you need – Drop Unused Indexes Drop Duplicate Indexes Assess Redundant Indexes for Dropping
- 39. #13 Configure MySQL Server Properly MySQL Defaults are not designed for High Performance Servers or High Intensity Workloads © 2020 Percona40
- 40. Top MySQL Settings To Change max_connections=N table_definition_cache=M table_open_cache=X innodb_buffer_pool_size= 16G (70% of RAM) innodb_flush_method=O_DIRECT innodb_log_file_size=2G © 2020 Percona 41
- 41. Operating System Limits Limit of Open Files is the most important for MySQL © 2020 Percona42
- 42. #14 Purge Unneeded Data © 2020 Percona 43 Is old data needed in the same operational database ? Move to different Server Move to Archival Storage
- 43. Tool to Consider pt-archiver © 2020 Percona44
- 44. #15 Perform Database Maintenance • Database Stats getting out of sync – ANALYZE TABLE • Database Fragmentation and Bloat – OPTIMIZE TABLE © 2020 Percona45
- 45. Optimizing All Tables mysqlcheck –optimize -A © 2020 Percona46
- 46. Word of Caution Database Maintenance makes database faster but takes a lot of resources to run - not a good thing to do during peak load © 2020 Percona 47
- 47. #16 Check your Background Jobs © 2020 Percona 48 Can this Background Jobs postponed beyond Peak Times Can this job be run on the replica (or other data jobs) Am I avoiding batch job overlaps ? Is this background job overdue for optimization ? Can resources this background job uses be limited ? (throttling)
- 48. #17 Check for Data Hotspots Data Hotspots – data objects (typically row) which have read, write or locking contention © 2020 Percona49
- 49. Can cause excessive Errors due to Rollbacks © 2020 Percona 50
- 50. Can see specific queries having issues © 2020 Percona 51
- 51. #18 Check your Application Server Configuration © 2020 Percona 52 Re-Establishing Connections Unnecessary, Especially with SSL/TLS Poor Connection Pool Settings Poor Performance Settings (especially for JDBC) Use of Prepared Statements Caching Settings (if Available)
- 52. Summary © 2020 Percona 53 Facing Traffic spike with MySQL – Do not Panic There are many tools available for you Do not try applying all of them, pick most appropriate ones Average system can candle at least 3x times traffic using only changing “low hanging fruits”
- 54. Thank you, Let’s Connect! https://www.linkedin.com/in/peterzaitsev/ https://twitter.com/PeterZaitsev