PGConf APAC 2018 - Monitoring PostgreSQL at Scale
•
2 j'aime•353 vues
Speaker: Lukas Fittl Your PostgreSQL database is one of the most important pieces of your architecture - yet the level of introspection available in Postgres is often hard to work with. Its easy to get very detailed information, but what should you really watch out for, send reports on and alert on? In this talk we'll discuss how query performance statistics can be made accessible to application developers, critical entries one should monitor in the PostgreSQL log files, how to collect EXPLAIN plans at scale, how to watch over autovacuum and VACUUM operations, and how to flag issues based on schema statistics. We'll also talk a bit about monitoring multi-server setups, first going into high availability and read standbys, logical replication, and then reviewing how monitoring looks like for sharded databases like Citus. The talk will primarily describe free/open-source tools and statistics views readily available from within Postgres.
Recommandé
Recommandé
Contenu connexe
Tendances
Tendances (20)
Similaire à PGConf APAC 2018 - Monitoring PostgreSQL at Scale
Similaire à PGConf APAC 2018 - Monitoring PostgreSQL at Scale (20)
Plus de PGConf APAC
Plus de PGConf APAC (20)
Dernier
Dernier (20)
PGConf APAC 2018 - Monitoring PostgreSQL at Scale
- 4. @LukasFittl For the purposes of this talk, Monitoring= Monitoring+Observability
- 6. @LukasFittl “We had an outage yesterday at 10am - whathappened?”
- 8. @LukasFittl DIYMonitoringHack: Save pg_stat_activity and pg_stat_database every 10 seconds into a separate monitoring database
- 9. @LukasFittl
- 10. @LukasFittl
- 12. @LukasFittl AppServers PostgreSQL DataDirectory Storage WAL Storage CPU Memory Connection Pooler
- 13. @LukasFittl AppServers PostgreSQL DataDirectory Storage WAL Storage CPU Memory Connection Pooler
- 22. @LukasFittl System Statistics (CPU Util %, I/O Util %, etc)
- 24. @LukasFittl OperationalMetrics Can Be Understood Better Through A Focus on Workload
- 26. @LukasFittl AgeOfOldestTransaction SELECT MAX(now() - xact_start) FROM pg_stat_activity WHERE state <> ‘idle’
- 27. @LukasFittl SELECT sum(heap_blks_hit) / nullif(sum(heap_blks_hit + heap_blks_read),0) FROM pg_statio_user_tables TableCacheHitRate Target: >= 99%
- 28. @LukasFittl SELECT relname, n_live_tup, seq_scan + idx_scan, 100 * idx_scan / (seq_scan + idx_scan) FROM pg_stat_user_tables ORDER BY n_live_tup DESC IndexHitRate Target: >= 95% on large, active tables
- 33. @LukasFittl New code deployed New customers App config change
- 34. @LukasFittl Ability to Drill Down From “HighCPUUtilization” To Specific Set of Queries
- 36. @LukasFittl Did the schema change recently?
- 37. @LukasFittl Did the queryplan change recently?
- 40. @LukasFittl shared_buffers max_connections (maintenance_)work_mem autovacuum settings checkpointer settings planner settings
- 42. @LukasFittl On Cloud Providers, Closely Watch Your I/OMetrics
- 43. @LukasFittl
- 46. @LukasFittl userid | 10 dbid | 1397527 query | SELECT * FROM x WHERE y = $1 calls | 5 total_time | 15.249 rows | 0 shared_blks_hit | 451 shared_blks_read | 41 shared_blks_dirtied | 26 shared_blks_written | 0 local_blks_hit | 0 local_blks_read | 0 local_blks_dirtied | 0 local_blks_written | 0 temp_blks_read | 0 temp_blks_written | 0 blk_read_time | 0 blk_write_time | 0 pg_stat_statements
- 48. @LukasFittl SELECT * FROM pg_stat_statements LIMIT 1; userid | 10 dbid | 17025 queryid | 1720234670 query | SELECT * FROM x WHERE y = $1 calls | 14 total_time | 0.151 rows | 28 shared_blks_hit | 14 shared_blks_read | 0 shared_blks_dirtied | 0 shared_blks_written | 0 local_blks_hit | 0 local_blks_read | 0 local_blks_dirtied | 0 local_blks_written | 0 temp_blks_read | 0 temp_blks_written | 0 blk_read_time | 0 blk_write_time | 0
- 49. @LukasFittl queryid | 1720234670 query | SELECT * FROM x WHERE y = ? calls | 5 total_time | 15.249 Query+No.ofCalls+AvgTime
- 50. @LukasFittl shared_blks_hit | 2447215 shared_blks_read | 55335 Avg.SharedBufferHitRate:97% hit_rate = shared_blks_hit / (shared_blks_hit + shared_blks_read)
- 51. @LukasFittl blk_read_time | 14.594 blk_write_time | 465.661 Timespentreading/writingtodisk track_io_timing = on
- 52. pg_qtop Simple top-like tool that shows pg_stat_statements data https://github.com/lfittl/pg_qtop
- 53. AVG | QUERY -------------------------------------------------------------------------------- 10.7ms | SELECT oid, typname, typelem, typdelim, typinput FROM pg_type 3.0ms | SET time zone 'UTC' 0.4ms | SELECT a.attname, format_type(a.atttypid, a.atttypmod), pg_get_expr(d.adbin, d.adrelid), a.attnotnull, a.atttypid, a.atttypmod FROM pg_attribute a LEFT JOIN pg_attrdef d ON a.attrelid = d.adrelid AND a.attnum = d.adnum WHERE a.attrelid = ?::regclass AND a.attnum > ? AND NOT a.attisdropped ORDER BY a.attnum 0.2ms | SELECT pg_stat_statements_reset() 0.1ms | SELECT query, calls, total_time FROM pg_stat_statements 0.1ms | SELECT attr.attname FROM pg_attribute attr INNER JOIN pg_constraint cons ON attr.attrelid = cons.conrelid AND attr.attnum = cons.conkey[?] WHERE cons.contype = ? AND cons.conrelid = ?: :regclass 0.0ms | SELECT COUNT(*) FROM pg_class c LEFT JOIN pg_namespace n ON n.oid = c.relnamespace WHERE c.relkind in (?,?) AND c.relname = ? AND n.nspname = ANY (current_schemas(?)) 0.0ms | SELECT * FROM posts JOIN users ON (posts.author_id = users.id) WHERE users.login = ?; 0.0ms | SET client_min_messages TO 'panic' 0.0ms | set client_encoding to 'UTF8' 0.0ms | SHOW client_min_messages 0.0ms | SELECT * FROM ad_reels WHERE id = ?; 0.0ms | SELECT * FROM posts WHERE guid = ?; 0.0ms | SELECT ? 0.0ms | SET client_min_messages TO 'warning' 0.0ms | SET standard_conforming_strings = on 0.0ms | SELECT "posts".* FROM "posts" ORDER BY "posts"."id" DESC LIMIT ? 0.0ms | SHOW TIME ZONE pg_qtop -d testdb
- 54. AVG | QUERY -------------------------------------------------------------------------------- 0.0ms | SELECT * FROM posts JOIN users ON (posts.author_id = users.id) WHERE users.login = ?; 0.0ms | SELECT * FROM posts WHERE guid = ?; 0.0ms | SELECT "posts".* FROM "posts" ORDER BY "posts"."id" DESC LIMIT ? pg_qtop -d testdb -t posts
- 55. AVG | CALLS | HIT RATE | QUERY -------------------------------------------------------------------------------- 0.1ms | 1 | 100.0 | SELECT * FROM users; 0.1ms | 1 | - | SELECT * FROM databases; 0.0ms | 1 | - | SELECT * FROM invoices; 0.0ms | 1 | - | SELECT * FROM query_snapshots; pg_qtop -d testdb -s select
- 56. pganalyze.com
- 58. @LukasFittl
- 59. @LukasFittl log_min_duration_statement gives you queries withparameters
- 60. @LukasFittl
- 61. @LukasFittl
- 62. @LukasFittl auto_explain logs the query plan for specific slow queries
- 63. @LukasFittl
- 64. @LukasFittl Scaling # of Servers Scaling # of Applications Scaling # of Developers
- 65. @LukasFittl Scaling # of Servers Scaling # of Applications Scaling # of Developers
- 66. @LukasFittl Read Replicas: Monitor ReplicaLag & CPUUtilization
- 67. @LukasFittl Sharding: Monitor Everything* * Each Data/Worker Node Should Be Treated Like Single-Node Postgres
- 68. @LukasFittl Scaling # of Servers Scaling # of Applications Scaling # of Developers
- 71. @LukasFittl LogFiles Add %a to your log_line_prefix
- 72. @LukasFittl Annotate Queries With Their QueryOrigin
- 73. @LukasFittl
- 74. @LukasFittl
- 75. @LukasFittl
- 77. @LukasFittl Make sure all apps are monitored the same way through Checklists
- 78. @LukasFittl Scaling # of Servers Scaling # of Applications Scaling # of Developers
- 79. @LukasFittl Make Tooling Accessible To AppDevelopers,NotJustDBAs
- 80. @LukasFittl It Only Takes 1Developer To …IntroduceASlowQueryWithoutAnIndex …WriteaMigrationThatTakesALongExclusiveLock …DropATable
- 81. @LukasFittl Code Review Needs To Include SchemaChangeReview & MissingIndexReview
- 82. @LukasFittl SETstatement_timeout=30s When Your (Web) App Connects To Protect Against Unexpected Slow Queries
- 85. @LukasFittl LevelsofNotification: - Page (Wake Up On-Call) - Instant Email Notification - Daily/Weekly Report Email
- 89. @LukasFittl FocusonQueryPerformance Distinguish user-visible slowness from secondary systems (e.g. background workers)
- 95. @LukasFittl Monitor YourPostgres: pganalyze.com Shard Your Postgres: citusdata.com Thanks!
- 96. @LukasFittl
- 98. @LukasFittl LongHeldLocksinTransactions Rails Counter Cache & Timestamps BEGIN SELECT 1 AS one FROM "post_votes" WHERE (…) LIMIT 1 SELECT "posts".* FROM "posts" WHERE "posts"."id" = $1 LIMIT 1 INSERT INTO "notifications" (…) VALUES (…) RETURNING "id" SELECT "users".* FROM "users" WHERE "users"."id" = $1 LIMIT 1 UPDATE "users" SET "updated_at" = ? WHERE "users"."id" = ? INSERT INTO "post_votes" (…) VALUES (…) RETURNING "id" UPDATE "posts" SET "votes" = COALESCE("votes", 0) + 1 WHERE "posts"."id" = ? UPDATE "posts" SET "credible_post_votes_count" = … WHERE "posts"."id" = ? UPDATE "users" SET "updated_at" = ? WHERE "users"."id" = ? UPDATE "posts" SET "updated_at" = ? WHERE "posts"."id" = ? COMMIT
- 99. @LukasFittl log_lock_waits=on deadlock_timeout = 1000 ms process 20679 still waiting for ExclusiveLock on tuple (566,1) of relation 16421 after 1000.115 ms process 20678 still waiting for ExclusiveLock on tuple (566,1) of relation 16421 after 1000.126 ms process 15533 still waiting for ExclusiveLock on tuple (566,1) of relation 16421 1000.129 ms process 20663 still waiting for ExclusiveLock on tuple (566,1) of relation 16421 1000.100 ms process 15537 still waiting for ExclusiveLock on tuple (566,1) of relation 16421 1000.130 ms process 15536 still waiting for ExclusiveLock on tuple (566,1) of relation 16421 1000.222 ms process 20734 still waiting for ExclusiveLock on tuple (566,1) of relation 16421 1000.130 ms process 15538 still waiting for ExclusiveLock on tuple (566,1) of relation 16421 1000.136 ms process 15758 still waiting for ShareLock on transaction 250175899 after 1000.073 ms
- 102. @LukasFittl RED: Long Queries Pid Duration Query ----- --------------- ----------------------------------------------------------------------------------------- 31868 00:10:52.165883 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 28835 01:06:16.505554 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 28836 01:06:16.486394 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 28837 01:06:16.41853 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 28838 01:06:16.380474 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 31869 00:10:26.068636 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 31870 00:08:47.790245 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 23781 01:12:14.603475 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 31862 00:10:47.98641 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 31863 00:11:05.921372 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 27648 01:17:38.773909 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 27098 01:28:51.216489 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 27106 01:27:18.455351 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 31880 00:10:52.877779 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 31881 00:10:52.168877 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 RED: Idle in Transaction 23729 01:37:12.566497 UPDATE "posts" SET "link_unique_visits" = COALESCE("link_unique_visits", 0) + 1 WHERE "posts"."id" = $1 RED: Blocking Queries 23760 UPDATE "posts" SET "link_unique_visits" = COALESCE("link_unique_visits", 0) + 1 WHERE "posts"."id" = $1 01:37:12.646868 23729 UPDATE "posts" SET "link_visits" = COALESCE("link_visits", 0) + 1 WHERE "posts"."id" = $1 01:31:59.088208 Dead Background Worker caused silent queue back-up for 1hr+