executor, util: improve concurrency of statement summary #14490

djshow832 · 2020-01-16T03:17:45Z

What problem does this PR solve?

Improve the performance in point get.
In point_get of wide_table_200, QPS decreases from 59k to 49k (refer to issue #14432). Because there're too many kinds of summaries to fit in the cache, creating a new summary needs to lock the cache, so much time is spent on waiting for the lock.
Also, some time is spent on generating digest.

After this PR, QPS increases back to 57k.

What is changed and how it works?

Move some calculation out of cache lock to decrease lock contention.
If it's a point_get, only generate digest once.

Check List

Tests

Unit test

Code changes

N/A

Side effects

Increased code complexity

Related changes

N/A

Release note

Improve concurrency of statement summary by decreasing lock contention.

djshow832 · 2020-01-16T04:44:29Z

/bench + sysbench + tpcc

sre-bot · 2020-01-16T06:09:47Z

Benchmark Report

Run TPC-C Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: e39b504177da2efee6a8981b31c45ac775c03612
+++ tidb: 9ceaa9b3ed8de92e66962196f99fb9ed97f5b0fa
tikv: 18a4fe02b916eb4f96873e5e92ad02ec349077a4
pd: 541e3a57cd996bc46e91def9eb5437b05b5da6b0
================================================================================
Measured tpmC (NewOrders): 6597.87 ± 0.09% (std=4.30), delta: -0.60% (p=0.866)

sre-bot · 2020-01-16T07:27:37Z

Benchmark Report

Run Sysbench Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: e39b504177da2efee6a8981b31c45ac775c03612
+++ tidb: 9ceaa9b3ed8de92e66962196f99fb9ed97f5b0fa
tikv: 18a4fe02b916eb4f96873e5e92ad02ec349077a4
pd: 541e3a57cd996bc46e91def9eb5437b05b5da6b0
================================================================================
oltp_insert:
    * QPS: 7430.87 ± 0.29% (std=15.38) delta: 0.16% (p=0.243)
    * Latency p50: 17.22 ± 0.29% (std=0.04) delta: -0.16%
    * Latency p99: 29.32 ± 2.23% (std=0.44) delta: -0.44%
            
oltp_point_select:
    * QPS: 43952.32 ± 0.06% (std=18.98) delta: 1.18% (p=0.429)
    * Latency p50: 2.91 ± 0.00% (std=0.00) delta: -0.23%
    * Latency p99: 9.91 ± 0.00% (std=0.00) delta: 0.00%
            
oltp_read_write:
    * QPS: 15891.28 ± 0.20% (std=27.82) delta: -0.07% (p=0.681)
    * Latency p50: 161.43 ± 0.20% (std=0.25) delta: 0.05%
    * Latency p99: 306.13 ± 2.72% (std=6.16) delta: 1.53%
            
oltp_update_index:
    * QPS: 4203.73 ± 0.57% (std=14.79) delta: -0.07% (p=0.412)
    * Latency p50: 30.45 ± 0.57% (std=0.11) delta: 0.10%
    * Latency p99: 52.91 ± 3.63% (std=1.51) delta: -1.16%
            
oltp_update_non_index:
    * QPS: 4722.89 ± 0.30% (std=10.55) delta: 0.02% (p=0.925)
    * Latency p50: 27.10 ± 0.30% (std=0.06) delta: -0.02%
    * Latency p99: 41.85 ± 0.00% (std=0.00) delta: -1.36%

djshow832 · 2020-01-16T08:09:45Z

/run-unit-test

zyxbest · 2020-01-16T08:52:00Z

/rebuild

zyxbest · 2020-01-16T08:52:49Z

/build

djshow832 · 2020-01-17T03:47:01Z

/bench + sysbench + tpcc

sre-bot · 2020-01-17T05:17:06Z

Benchmark Report

Run TPC-C Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: 74bc00dfa953da06d8e5823d0acbf6bf1858a680
+++ tidb: 0b5f79dc3a4ee74ce6abc90f9242c00d4324db61
tikv: c75e85a7535127b4373514f0c80fc631dd30b7b7
pd: 541e3a57cd996bc46e91def9eb5437b05b5da6b0
================================================================================
Measured tpmC (NewOrders): 6624.33 ± 1.08% (std=52.15), delta: 1.17% (p=0.086)

sre-bot · 2020-01-17T05:18:10Z

Benchmark Report

Run Sysbench Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: 74bc00dfa953da06d8e5823d0acbf6bf1858a680
+++ tidb: 0b5f79dc3a4ee74ce6abc90f9242c00d4324db61
tikv: c75e85a7535127b4373514f0c80fc631dd30b7b7
pd: 541e3a57cd996bc46e91def9eb5437b05b5da6b0
================================================================================
oltp_insert:
    * QPS: 4669.07 ± 0.36% (std=12.75) delta: -0.00% (p=0.987)
    * Latency p50: 27.41 ± 0.36% (std=0.07) delta: 0.01%
    * Latency p99: 44.60 ± 4.56% (std=1.34) delta: 0.93%
            
oltp_point_select:
    * QPS: 43083.57 ± 0.32% (std=96.11) delta: 2.10% (p=0.001)
    * Latency p50: 2.97 ± 0.22% (std=0.00) delta: -1.87%
    * Latency p99: 10.09 ± 0.00% (std=0.00) delta: -1.18%
            
oltp_read_write:
    * QPS: 14747.10 ± 0.66% (std=59.57) delta: -0.06% (p=0.819)
    * Latency p50: 173.89 ± 0.66% (std=0.71) delta: 0.07%
    * Latency p99: 235.74 ± 0.00% (std=0.00) delta: 0.00%
            
oltp_update_index:
    * QPS: 4223.51 ± 0.66% (std=16.94) delta: -0.17% (p=0.599)
    * Latency p50: 30.31 ± 0.67% (std=0.12) delta: 0.18%
    * Latency p99: 54.50 ± 1.20% (std=0.46) delta: 1.82%
            
oltp_update_non_index:
    * QPS: 4737.39 ± 0.23% (std=7.30) delta: 0.39% (p=0.314)
    * Latency p50: 27.02 ± 0.23% (std=0.04) delta: -0.39%
    * Latency p99: 42.06 ± 5.03% (std=1.26) delta: -1.79%

djshow832 · 2020-01-17T06:01:50Z

/run-all-tests

djshow832 · 2020-01-17T07:54:02Z

/run-all-tests

bb7133

LGTM

djshow832 · 2020-02-03T04:16:33Z

/run-all-tests

crazycs520 · 2020-02-03T05:24:22Z

executor/adapter.go


 	execDetail := stmtCtx.GetExecDetails()
 	copTaskInfo := stmtCtx.CopTasksDetails()
 	memMax := stmtCtx.MemTracker.MaxConsumed()
 	var userString string
 	if sessVars.User != nil {
-		userString = sessVars.User.String()
+		userString = sessVars.User.Username


Attention, User.Username doesn't contain the User.Hostname, are you sure about this change?

If User.String() is slow, how bout optimize the funtion, we can use string + string... instead of fmt.Sprintf.

I'm not sure whether host name helps when analyzing performance. According to your experience, is it helpful?

I rarely use use.host info.

util/stmtsummary/statement_summary.go

executor/adapter.go

sre-bot · 2020-02-03T10:59:27Z

/run-all-tests

sre-bot · 2020-02-03T11:01:32Z

@djshow832 merge failed.

djshow832 · 2020-02-03T15:13:02Z

/merge

sre-bot · 2020-02-03T15:15:25Z

/run-all-tests

sre-bot · 2020-02-03T15:17:30Z

@djshow832 merge failed.

sre-bot · 2020-02-04T03:47:22Z

/run-all-tests

djshow832 · 2020-02-04T06:09:55Z

/merge

sre-bot · 2020-02-04T06:12:26Z

/run-all-tests

djshow832 · 2020-02-04T07:37:19Z

/build

djshow832 · 2020-02-04T08:17:27Z

/rebuild

djshow832 · 2020-02-04T09:27:14Z

/merge

sre-bot · 2020-02-04T09:29:26Z

/run-all-tests

sre-bot · 2020-02-04T13:08:08Z

Benchmark Report

Run Sysbench Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: 2192448e501757fd771b0a255b1265b510f637e5
+++ tidb: 2b6eba64a05c429e5b867f2833c93727c330f7e6
tikv: 0f08b6beafe17825329d4a1c00a22db6b6699e6c
pd: 4686d4622ad3ebc7fc267b5c64437129cee356fd
================================================================================
oltp_insert:
    * QPS: 4714.69 ± 0.43% (std=14.24) delta: 0.05% (p=0.847)
    * Latency p50: 27.15 ± 0.43% (std=0.08) delta: -0.05%
    * Latency p99: 47.09 ± 6.21% (std=2.19) delta: 1.42%
            
oltp_point_select:
    * QPS: 42531.46 ± 0.47% (std=138.74) delta: -0.36% (p=0.292)
    * Latency p50: 3.01 ± 0.58% (std=0.01) delta: 0.42%
    * Latency p99: 10.09 ± 0.00% (std=0.00) delta: -1.75%
            
oltp_read_write:
    * QPS: 15408.32 ± 0.45% (std=45.62) delta: 0.41% (p=0.194)
    * Latency p50: 166.50 ± 0.41% (std=0.48) delta: -0.38%
    * Latency p99: 327.49 ± 2.24% (std=4.87) delta: -1.78%
            
oltp_update_index:
    * QPS: 6261.06 ± 0.62% (std=23.69) delta: -0.09% (p=0.727)
    * Latency p50: 20.44 ± 0.62% (std=0.08) delta: 0.09%
    * Latency p99: 38.25 ± 1.80% (std=0.49) delta: 0.00%
            
oltp_update_non_index:
    * QPS: 4707.55 ± 0.33% (std=10.86) delta: 0.14% (p=0.332)
    * Latency p50: 27.19 ± 0.33% (std=0.06) delta: -0.11%
    * Latency p99: 41.60 ± 1.20% (std=0.35) delta: -0.60%

)

djshow832 added the status/DNM label Jan 16, 2020

djshow832 force-pushed the stmt_summary_improve_concurrency branch from 0f332bd to 9ceaa9b Compare January 16, 2020 04:40

djshow832 changed the title ~~util: improve concurrency of statement summary~~ executor, util: improve concurrency of statement summary Jan 17, 2020

djshow832 added type/usability type/performance and removed status/DNM labels Jan 17, 2020

djshow832 requested review from crazycs520 and bb7133 January 17, 2020 04:32

djshow832 mentioned this pull request Jan 17, 2020

70% performance regression on wide_table_200 and 4% regression on TPCC workload #14432

Closed

djshow832 force-pushed the stmt_summary_improve_concurrency branch from 0b5f79d to 2e8d43f Compare January 17, 2020 07:53

bb7133 reviewed Jan 19, 2020

View reviewed changes

bb7133 added status/LGT1 Indicates that a PR has LGTM 1. type/enhancement The issue or PR belongs to an enhancement. labels Jan 19, 2020

bb7133 added this to the v3.0.10 milestone Jan 19, 2020

djshow832 force-pushed the stmt_summary_improve_concurrency branch from ac86eeb to 5899162 Compare February 3, 2020 04:07

crazycs520 reviewed Feb 3, 2020

View reviewed changes

util/stmtsummary/statement_summary.go Show resolved Hide resolved

crazycs520 reviewed Feb 3, 2020

View reviewed changes

executor/adapter.go Show resolved Hide resolved

djshow832 added 7 commits February 4, 2020 11:13

release map lock sooner

c528f25

fix select stmt summary

79cd782

skip get plan digest when it is point get

c5015b9

add test cases

b2fe7f5

add comments

5967a1f

release map lock sooner

cf4883b

skip get plan digest when it is point get

3440c84

djshow832 dismissed crazycs520’s stale review via 3440c84 February 4, 2020 03:37

djshow832 force-pushed the stmt_summary_improve_concurrency branch from f5a6a9a to 3440c84 Compare February 4, 2020 03:37

crazycs520 approved these changes Feb 4, 2020

View reviewed changes

Merge branch 'master' into stmt_summary_improve_concurrency

c058842

Merge branch 'master' into stmt_summary_improve_concurrency

2b6eba6

sre-bot merged commit 7ba002e into pingcap:master Feb 4, 2020

djshow832 added a commit to djshow832/tidb that referenced this pull request Feb 4, 2020

executor, util: improve concurrency of statement summary (pingcap#14490)

43105cb

sre-bot pushed a commit that referenced this pull request Feb 5, 2020

executor, util: improve concurrency of statement summary(#14490) (#14627

ac3f7f7

)

hsqlu pushed a commit to hsqlu/tidb that referenced this pull request Feb 8, 2020

executor, util: improve concurrency of statement summary (pingcap#14490)

e029879

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

executor, util: improve concurrency of statement summary #14490

executor, util: improve concurrency of statement summary #14490

djshow832 commented Jan 16, 2020 •

edited

Loading

djshow832 commented Jan 16, 2020

sre-bot commented Jan 16, 2020

sre-bot commented Jan 16, 2020

djshow832 commented Jan 16, 2020

zyxbest commented Jan 16, 2020

zyxbest commented Jan 16, 2020

djshow832 commented Jan 17, 2020

sre-bot commented Jan 17, 2020

sre-bot commented Jan 17, 2020

djshow832 commented Jan 17, 2020

djshow832 commented Jan 17, 2020

bb7133 left a comment

djshow832 commented Feb 3, 2020

crazycs520 Feb 3, 2020

djshow832 Feb 3, 2020

crazycs520 Feb 3, 2020

sre-bot commented Feb 3, 2020

sre-bot commented Feb 3, 2020

djshow832 commented Feb 3, 2020

sre-bot commented Feb 3, 2020

sre-bot commented Feb 3, 2020

sre-bot commented Feb 4, 2020

djshow832 commented Feb 4, 2020

sre-bot commented Feb 4, 2020

djshow832 commented Feb 4, 2020

djshow832 commented Feb 4, 2020

djshow832 commented Feb 4, 2020

sre-bot commented Feb 4, 2020

sre-bot commented Feb 4, 2020

executor, util: improve concurrency of statement summary #14490

executor, util: improve concurrency of statement summary #14490

Conversation

djshow832 commented Jan 16, 2020 • edited Loading

What problem does this PR solve?

What is changed and how it works?

Check List

djshow832 commented Jan 16, 2020

sre-bot commented Jan 16, 2020

Benchmark Report

sre-bot commented Jan 16, 2020

Benchmark Report

djshow832 commented Jan 16, 2020

zyxbest commented Jan 16, 2020

zyxbest commented Jan 16, 2020

djshow832 commented Jan 17, 2020

sre-bot commented Jan 17, 2020

Benchmark Report

sre-bot commented Jan 17, 2020

Benchmark Report

djshow832 commented Jan 17, 2020

djshow832 commented Jan 17, 2020

bb7133 left a comment

Choose a reason for hiding this comment

djshow832 commented Feb 3, 2020

crazycs520 Feb 3, 2020

Choose a reason for hiding this comment

djshow832 Feb 3, 2020

Choose a reason for hiding this comment

crazycs520 Feb 3, 2020

Choose a reason for hiding this comment

sre-bot commented Feb 3, 2020

sre-bot commented Feb 3, 2020

djshow832 commented Feb 3, 2020

sre-bot commented Feb 3, 2020

sre-bot commented Feb 3, 2020

sre-bot commented Feb 4, 2020

djshow832 commented Feb 4, 2020

sre-bot commented Feb 4, 2020

djshow832 commented Feb 4, 2020

djshow832 commented Feb 4, 2020

djshow832 commented Feb 4, 2020

sre-bot commented Feb 4, 2020

sre-bot commented Feb 4, 2020

Benchmark Report

djshow832 commented Jan 16, 2020 •

edited

Loading