dbtester

README

Distributed database benchmark tester: etcd, Zookeeper, Consul, zetcd, cetcd

It includes github.com/golang/freetype, which is based in part on the work of the FreeType Team.

Performance Analysis

Latest test results can be found at https://github.com/etcd-io/dbtester/tree/master/test-results
Exploring Performance of etcd, Zookeeper and Consul Consistent Key-value Datastores (February 17, 2017)
- https://coreos.com/blog/performance-of-etcd.html

Project

dbtester system architecture

Database Agent
- https://github.com/etcd-io/dbtester/tree/master/agent
Database Client
- https://github.com/etcd-io/dbtester/tree/master/control
System Metrics
- https://github.com/gyuho/linux-inspect
Test Data Analysis

For etcd, we recommend etcd benchmark tool.

All logs and results can be found at https://github.com/etcd-io/dbtester/tree/master/test-results or https://console.cloud.google.com/storage/browser/dbtester-results/?authuser=0&project=etcd-development.

Noticeable Warnings: Zookeeper

Snapshot, when writing 1-million entries (256-byte key, 1KB value value), with 500 concurrent clients

# snapshot warnings
cd 2017Q1-00-etcd-zookeeper-consul/02-write-1M-keys-best-throughput
grep -r -i fsync-ing\ the zookeeper-r3.4.9-java8-* | less

2017-02-10 18:55:38,997 [myid:3] - WARN  [SyncThread:3:SyncRequestProcessor@148] - Too busy to snap, skipping
2017-02-10 18:55:38,998 [myid:3] - INFO  [SyncThread:3:FileTxnLog@203] - Creating new log file: log.1000c0c51
2017-02-10 18:55:40,855 [myid:3] - INFO  [SyncThread:3:FileTxnLog@203] - Creating new log file: log.1000cd2e6
2017-02-10 18:55:40,855 [myid:3] - INFO  [Snapshot Thread:FileTxnSnapLog@240] - Snapshotting: 0x1000cd1ca to /home/gyuho/zookeeper/zookeeper.data/version-2/snapshot.1000cd1ca
2017-02-10 18:55:46,382 [myid:3] - WARN  [SyncThread:3:FileTxnLog@338] - fsync-ing the write ahead log in SyncThread:3 took 1062ms which will adversely effect operation latency. See the ZooKeeper troubleshooting guide
2017-02-10 18:55:47,471 [myid:3] - WARN  [SyncThread:3:FileTxnLog@338] - fsync-ing the write ahead log in SyncThread:3 took 1084ms which will adversely effect operation latency. See the ZooKeeper troubleshooting guide
2017-02-10 18:55:49,425 [myid:3] - WARN  [SyncThread:3:FileTxnLog@338] - fsync-ing the write ahead log in SyncThread:3 took 1142ms which will adversely effect operation latency. See the ZooKeeper troubleshooting guide
2017-02-10 18:55:51,188 [myid:3] - WARN  [SyncThread:3:FileTxnLog@338] - fsync-ing the write ahead log in SyncThread:3 took 1201ms which will adversely effect operation latency. See the ZooKeeper troubleshooting guide
2017-02-10 18:55:52,292 [myid:3] - WARN  [SyncThread:3:FileTxnLog@338] - fsync-ing the write ahead log in SyncThread:3 took 1102ms which will adversely effect operation latency. See the ZooKeeper troubleshooting guide

When writing more than 2-million entries (256-byte key, 1KB value value) with 500 concurrent clients

# leader election
cd 2017Q1-00-etcd-zookeeper-consul/04-write-too-many-keys
grep -r -i election\ took  zookeeper-r3.4.9-java8-* | less

# leader election is taking more than 10 seconds...
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:22:16,549 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Follower@61] - FOLLOWING - LEADER ELECTION TOOK - 22978
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:23:02,279 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 10210
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:23:14,498 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 203
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:23:36,303 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 9791
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:23:52,151 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 3836
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:24:13,849 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 9686
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:24:29,694 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 3573
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:24:51,392 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 8686
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:25:07,231 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 3827
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:25:28,940 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 9697
zookeeper-r3.4.9-java8-2-database.log:2017-02-10 19:25:44,772 [myid:2] - INFO  [QuorumPeer[myid=2]/0:0:0:0:0:0:0:0:2181:Leader@361] - LEADING - LEADER ELECTION TOOK - 3820

Noticeable Warnings: Consul

Snapshot, when writing 1-million entries (256-byte key, 1KB value value), with 500 concurrent clients

# snapshot warnings
cd 2017Q1-00-etcd-zookeeper-consul/02-write-1M-keys-best-throughput
grep -r -i installed\ remote consul-v0.7.4-go1.7.5-* | less

    2017/02/10 18:58:43 [INFO] snapshot: Creating new snapshot at /home/gyuho/consul.data/raft/snapshots/2-900345-1486753123478.tmp
    2017/02/10 18:58:45 [INFO] snapshot: reaping snapshot /home/gyuho/consul.data/raft/snapshots/2-849399-1486753096972
    2017/02/10 18:58:46 [INFO] raft: Copied 1223270573 bytes to local snapshot
    2017/02/10 18:58:55 [INFO] raft: Compacting logs from 868354 to 868801
    2017/02/10 18:58:56 [INFO] raft: Installed remote snapshot
    2017/02/10 18:58:57 [INFO] snapshot: Creating new snapshot at /home/gyuho/consul.data/raft/snapshots/2-911546-1486753137827.tmp
    2017/02/10 18:58:59 [INFO] consul.fsm: snapshot created in 32.255µs
    2017/02/10 18:59:01 [INFO] snapshot: reaping snapshot /home/gyuho/consul.data/raft/snapshots/2-873921-1486753116619
    2017/02/10 18:59:02 [INFO] raft: Copied 1238491373 bytes to local snapshot
    2017/02/10 18:59:11 [INFO] raft: Compacting logs from 868802 to 868801
    2017/02/10 18:59:11 [INFO] raft: Installed remote snapshot

Logs do not tell much but average latency spikes (e.g. from 70.27517 ms to 10407.900082 ms)

2017Q2-02-write-1M-network-traffic-best-throughput

2017Q2-01-write-1M-throughput-client-scaling

2017Q2-02-write-1M-latency-best-throughput

Write 1M keys, 256-byte key, 1KB value, Best Throughput (etcd 1K clients with 100 conns, Zookeeper 700, Consul 500 clients)

Google Cloud Compute Engine
4 machines of 16 vCPUs + 60 GB Memory + 300 GB SSD (1 for client)
Ubuntu 17.10 (GNU/Linux kernel 4.13.0-25-generic)
ulimit -n is 120000
etcd v3.3.0 (Go 1.9.2)
Zookeeper r3.5.3-beta
- Java 8
- javac 1.8.0_151
- Java(TM) SE Runtime Environment (build 1.8.0_151-b12)
- Java HotSpot(TM) 64-Bit Server VM (build 25.151-b12, mixed mode)
- /usr/bin/java -Djute.maxbuffer=33554432 -Xms50G -Xmx50G
Consul v1.0.2 (Go 1.9.2)

+---------------------------------------+---------------------+-----------------------------+-----------------------+
|                                       | etcd-v3.3.0-go1.9.2 | zookeeper-r3.5.3-beta-java8 | consul-v1.0.2-go1.9.2 |
+---------------------------------------+---------------------+-----------------------------+-----------------------+
|                         TOTAL-SECONDS |         28.3623 sec |                 59.2167 sec |          178.9443 sec |
|                  TOTAL-REQUEST-NUMBER |           1,000,000 |                   1,000,000 |             1,000,000 |
|                        MAX-THROUGHPUT |      37,330 req/sec |              25,124 req/sec |        15,865 req/sec |
|                        AVG-THROUGHPUT |      35,258 req/sec |              16,842 req/sec |         5,588 req/sec |
|                        MIN-THROUGHPUT |      13,505 req/sec |                  20 req/sec |             0 req/sec |
|                       FASTEST-LATENCY |           4.6073 ms |                   2.9094 ms |            11.6604 ms |
|                           AVG-LATENCY |          28.2625 ms |                  30.9499 ms |            89.4351 ms |
|                       SLOWEST-LATENCY |         117.4918 ms |                4564.6788 ms |          4616.2947 ms |
|                           Latency p10 |        13.508626 ms |                 9.068163 ms |          30.408863 ms |
|                           Latency p25 |        16.869586 ms |                 9.351597 ms |          34.224021 ms |
|                           Latency p50 |        22.167478 ms |                10.093377 ms |          39.881181 ms |
|                           Latency p75 |        34.855941 ms |                14.951189 ms |          52.644787 ms |
|                           Latency p90 |        54.613394 ms |                28.497256 ms |         118.340402 ms |
|                           Latency p95 |        59.785127 ms |                72.671788 ms |         229.129526 ms |
|                           Latency p99 |        74.139638 ms |               273.218523 ms |        1495.660763 ms |
|                         Latency p99.9 |        97.385495 ms |              2526.873285 ms |        3499.225138 ms |
|      SERVER-TOTAL-NETWORK-RX-DATA-SUM |              5.1 GB |                      4.6 GB |                5.6 GB |
|      SERVER-TOTAL-NETWORK-TX-DATA-SUM |              3.8 GB |                      3.6 GB |                4.4 GB |
|           CLIENT-TOTAL-NETWORK-RX-SUM |              252 MB |                      357 MB |                206 MB |
|           CLIENT-TOTAL-NETWORK-TX-SUM |              1.5 GB |                      1.4 GB |                1.5 GB |
|                  SERVER-MAX-CPU-USAGE |            446.83 % |                   1122.00 % |              426.33 % |
|               SERVER-MAX-MEMORY-USAGE |              1.1 GB |                       15 GB |                4.6 GB |
|                  CLIENT-MAX-CPU-USAGE |            606.00 % |                    314.00 % |              215.00 % |
|               CLIENT-MAX-MEMORY-USAGE |               96 MB |                      2.4 GB |                 86 MB |
|                    CLIENT-ERROR-COUNT |                   0 |                       2,652 |                     0 |
|  SERVER-AVG-READS-COMPLETED-DELTA-SUM |                   0 |                         237 |                     2 |
|    SERVER-AVG-SECTORS-READS-DELTA-SUM |                   0 |                           0 |                     0 |
| SERVER-AVG-WRITES-COMPLETED-DELTA-SUM |             108,067 |                     157,034 |               675,072 |
|  SERVER-AVG-SECTORS-WRITTEN-DELTA-SUM |          20,449,360 |                  16,480,488 |           106,836,768 |
|           SERVER-AVG-DISK-SPACE-USAGE |              2.6 GB |                      6.9 GB |                2.9 GB |
+---------------------------------------+---------------------+-----------------------------+-----------------------+


zookeeper__r3_5_3_beta errors:
"zk: connection closed" (count 2,264)
"zk: could not connect to a server" (count 388)

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-LATENCY-MS

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-LATENCY-MS-BY-KEY

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-LATENCY-MS-BY-KEY-ERROR-POINTS

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-THROUGHPUT

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-VOLUNTARY-CTXT-SWITCHES

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-NON-VOLUNTARY-CTXT-SWITCHES

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-CPU

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/MAX-CPU

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-VMRSS-MB

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-VMRSS-MB-BY-KEY

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-VMRSS-MB-BY-KEY-ERROR-POINTS

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-READS-COMPLETED-DELTA

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-SECTORS-READ-DELTA

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-WRITES-COMPLETED-DELTA

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-SECTORS-WRITTEN-DELTA

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-READ-BYTES-NUM-DELTA

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-WRITE-BYTES-NUM-DELTA

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-RECEIVE-BYTES-NUM-DELTA

2018Q1-02-etcd-zookeeper-consul/write-1M-keys-best-throughput/AVG-TRANSMIT-BYTES-NUM-DELTA

Expand ▾ Collapse ▴

Documentation

Index ¶

Variables
func MakeTag(desc string) string
type Config
- func ReadConfig(fpath string, analyze bool) (*Config, error)
type CumulativeKeyNumAndOtherData
type CumulativeKeyNumAndOtherDataByUnixSecond
type CumulativeKeyNumAndOtherDataSlice
- func FindRangesData(data []CumulativeKeyNumAndOtherData, unit int64, totalRequests int64) CumulativeKeyNumAndOtherDataSlice
type CumulativeKeyNumToAvgLatency
type CumulativeKeyNumToAvgLatencySlice
- func FindRangesLatency(data report.TimeSeries, unit int64, totalRequests int64) CumulativeKeyNumToAvgLatencySlice
type ReqHandler

Constants ¶

This section is empty.

Variables ¶

View Source

var DiskSpaceUsageSummaryColumns = []string{
	"INDEX",
	"DATABASE-ENDPOINT",
	"DISK-SPACE-USAGE",
	"DISK-SPACE-USAGE-BYTES-NUM",
}

DiskSpaceUsageSummaryColumns defines summary columns.

Functions ¶

func MakeTag ¶

func MakeTag(desc string) string

MakeTag converts database description to database tag.

Types ¶

type Config ¶

type Config struct {
	TestTitle       string `yaml:"test_title"`
	TestDescription string `yaml:"test_description"`

	dbtesterpb.ConfigClientMachineInitial `yaml:"config_client_machine_initial"`

	AllDatabaseIDList                           []string                                              `yaml:"all_database_id_list"`
	DatabaseIDToConfigClientMachineAgentControl map[string]dbtesterpb.ConfigClientMachineAgentControl `yaml:"datatbase_id_to_config_client_machine_agent_control"`
	DatabaseIDToConfigAnalyzeMachineInitial     map[string]dbtesterpb.ConfigAnalyzeMachineInitial     `yaml:"datatbase_id_to_config_analyze_machine_initial"`

	dbtesterpb.ConfigAnalyzeMachineAllAggregatedOutput `yaml:"analyze_all_aggregated_output"`
	AnalyzePlotPathPrefix                              string                                `yaml:"analyze_plot_path_prefix"`
	AnalyzePlotList                                    []dbtesterpb.ConfigAnalyzeMachinePlot `yaml:"analyze_plot_list"`
	dbtesterpb.ConfigAnalyzeMachineREADME              `yaml:"analyze_readme"`
	// contains filtered or unexported fields
}

Config configures dbtester control clients.

func ReadConfig ¶

func ReadConfig(fpath string, analyze bool) (*Config, error)

ReadConfig reads control configuration file.

func (*Config) BroadcaseRequest ¶

func (cfg *Config) BroadcaseRequest(databaseID string, op dbtesterpb.Operation) (map[int]dbtesterpb.Response, error)

BroadcaseRequest sends request to all endpoints.

func (*Config) SaveDiskSpaceUsageSummary ¶

func (cfg *Config) SaveDiskSpaceUsageSummary(databaseID string, idxToResponse map[int]dbtesterpb.Response) error

SaveDiskSpaceUsageSummary saves data size summary.

func (*Config) Stress ¶

func (cfg *Config) Stress(databaseID string) error

Stress stresses the database.

func (*Config) ToRequest ¶

func (cfg *Config) ToRequest(databaseID string, op dbtesterpb.Operation, idx int) (req *dbtesterpb.Request, err error)

ToRequest converts configuration to 'dbtesterpb.Request'.

func (*Config) UploadToGoogle ¶

func (cfg *Config) UploadToGoogle(databaseID string, targetPath string) error

UploadToGoogle uploads target file to Google Cloud Storage.

func (*Config) WriteREADME ¶

func (cfg *Config) WriteREADME(summary string) error

WriteREADME writes README.

type CumulativeKeyNumAndOtherData ¶

type CumulativeKeyNumAndOtherData struct {
	UnixSecond int64
	Throughput int64

	CumulativeKeyNum int64

	MinMemoryMB float64
	AvgMemoryMB float64
	MaxMemoryMB float64

	AvgReadBytesDelta  float64
	AvgWriteBytesDelta float64
}

CumulativeKeyNumAndOtherData wraps the cumulative number of keys and according memory data. So the higher 'CumulativeKeyNum' is, the later the data points are in the time series.

type CumulativeKeyNumAndOtherDataByUnixSecond ¶

type CumulativeKeyNumAndOtherDataByUnixSecond []CumulativeKeyNumAndOtherData

CumulativeKeyNumAndOtherDataByUnixSecond is a slice of CumulativeKeyNumAndOtherData to sort by UnixSecond.

func (CumulativeKeyNumAndOtherDataByUnixSecond) Len ¶

func (t CumulativeKeyNumAndOtherDataByUnixSecond) Len() int

func (CumulativeKeyNumAndOtherDataByUnixSecond) Less ¶

func (t CumulativeKeyNumAndOtherDataByUnixSecond) Less(i, j int) bool

func (CumulativeKeyNumAndOtherDataByUnixSecond) Swap ¶

func (t CumulativeKeyNumAndOtherDataByUnixSecond) Swap(i, j int)

type CumulativeKeyNumAndOtherDataSlice ¶

type CumulativeKeyNumAndOtherDataSlice []CumulativeKeyNumAndOtherData

CumulativeKeyNumAndOtherDataSlice is a slice of CumulativeKeyNumAndOtherData to sort by CumulativeKeyNum.

func FindRangesData ¶

func FindRangesData(data []CumulativeKeyNumAndOtherData, unit int64, totalRequests int64) CumulativeKeyNumAndOtherDataSlice

FindRangesData sorts all data points by its timestamp. And then aggregate by the cumulative throughput, in order to map the number of keys to the average memory usage.

func (CumulativeKeyNumAndOtherDataSlice) Len ¶

func (t CumulativeKeyNumAndOtherDataSlice) Len() int

func (CumulativeKeyNumAndOtherDataSlice) Less ¶

func (t CumulativeKeyNumAndOtherDataSlice) Less(i, j int) bool

func (CumulativeKeyNumAndOtherDataSlice) Swap ¶

func (t CumulativeKeyNumAndOtherDataSlice) Swap(i, j int)

type CumulativeKeyNumToAvgLatency ¶

type CumulativeKeyNumToAvgLatency struct {
	CumulativeKeyNum int64

	MinLatency time.Duration
	AvgLatency time.Duration
	MaxLatency time.Duration
}

CumulativeKeyNumToAvgLatency wraps the cumulative number of keys and according latency data. So the higher 'CumulativeKeyNum' is, the later the data points are in the time series.

type CumulativeKeyNumToAvgLatencySlice ¶

type CumulativeKeyNumToAvgLatencySlice []CumulativeKeyNumToAvgLatency

CumulativeKeyNumToAvgLatencySlice is a slice of CumulativeKeyNumToAvgLatency to sort by CumulativeKeyNum.

func FindRangesLatency ¶

func FindRangesLatency(data report.TimeSeries, unit int64, totalRequests int64) CumulativeKeyNumToAvgLatencySlice

FindRangesLatency sorts all data points by its timestamp. And then aggregate by the cumulative throughput, in order to map the number of keys to the average latency.

type DataPoint struct {
	Timestamp  int64
	MinLatency time.Duration
	AvgLatency time.Duration
	MaxLatency time.Duration
	ThroughPut int64
}

If unit is 1000 and the average throughput per second is 30,000 and its average latency is 10ms, it will have 30 data points with latency 10ms.

func (CumulativeKeyNumToAvgLatencySlice) Len ¶

func (t CumulativeKeyNumToAvgLatencySlice) Len() int

func (CumulativeKeyNumToAvgLatencySlice) Less ¶

func (t CumulativeKeyNumToAvgLatencySlice) Less(i, j int) bool

func (CumulativeKeyNumToAvgLatencySlice) Swap ¶

func (t CumulativeKeyNumToAvgLatencySlice) Swap(i, j int)

type ReqHandler ¶

type ReqHandler func(ctx context.Context, req *request) error

ReqHandler wraps request handler.

Source Files

View all

Directories

Path	Synopsis
agent Package agent is a database agent in remote servers.	Package agent is a database agent in remote servers.
analyze Package analyze analyzes the bench test results, specific to dbtester tests.	Package analyze analyzes the bench test results, specific to dbtester tests.
cmd
dbtester dbtester is distributed database tester.	dbtester is distributed database tester.
control Package control controls the database agents and benchmark testers.	Package control controls the database agents and benchmark testers.
dbtesterpb Package dbtesterpb is a generated protocol buffer package.	Package dbtesterpb is a generated protocol buffer package.
pkg
fileinspect Package fileinspect implements various file inspecting utils.	Package fileinspect implements various file inspecting utils.
ntp Package ntp syncs system time with NTP server.	Package ntp syncs system time with NTP server.
remotestorage Package remotestorage wraps cloud storage API.	Package remotestorage wraps cloud storage API.
remotestorage/example

?	: This menu
/	: Search site
f or F	: Jump to