Page History

Apache GitHub Pull Request

https://github.com/apache/knox/pull/365(This document is still Work In Progress!)

Performance test tool design

The general goal is to have an extendable performance test framework that will drive the Knox Gateway configured by different use cases.

...

• easy to extend with new use-cases
• has the ability to execute long-running jobshas the ability to connect to it and query different metrics
• generate meaningful and easy-to-read reports of those metrics automatically (in configurable time periods)

...

As of today (17 Aug 2020), there is only one use-case implementation exists to address the above-written acquire/renew/use Knox Delegation token case. The related resources (Java classes, properties files) are located in the gateway-performance-test Maven module. Here is the list of the most relevant resources:

• src/main/java
  o/org.aapache.kknox.ggateway.pperformance.test

  
  • t.PerformanceTestRunner - this is the entry point of the tool. This class comes with a main method that reads the given configuration file and executes all enabled use-case runners
  • o.a.k.g.p.t.ResponseTimeCache - this class acts as a holder of response times and shared between the worker threads (which write into the cache) and the report generation threads (reading data from it)
  • o.a.k.g.p.t.reporting.GatewayMetricsReporter - this class generates the human-readable reports in JSON and YAML format in a fixed schedule marked by perf.test.report.generation.periodInSecs 
  • o.a.k.g.p.t.knoxtoken.KnoxTokenUseCaseRunner - this class is responsible for
    • start N worker threads that are acquiring Knox DTs parallel (marked by perf.test.usecase.knoxtoken.numOfThreads)
    • and 2 more threads to
      • renew an already acquired Knox DT
      • do an HDFS ls command using an already acquired Knox DT

  • o.a.k.g.p.t.knoxtoken.KnoxTokenWorkerThread - this represents the job which that actually acquires/renews/uses Knox DTs. The renew/use actions are running only on 1-1 thread and they wait 2 times more time between two subsequential calls then simply executing the acquire action. In other words, by default, a worker thread acquires a Knox DT between every 5 to 10 seconds (on N threads) whereas a worker thread which renews/uses a previously acquired Knox token waits between 10-20 seconds.

  • o.a.k.g.p.t.knoxtoken.KnoxTokenCache - stores the already acquired Knox DTs (if the number of DTs reaches 500 the cache is cleaned automatically)
• src/test/resources
  • performance.test.configuration.properties - contains the above-described configuration file
  • performanceTest-log4j.properties - the Log4j configuration of the tool. By default, it prints log messages on the STDOUT as well as writes them into target/logs/performanceTest.log

Knox gateway requisites

The performance test tool tries to connect to a Knox instance to

• acquire/renew/use Knox delegation tokens using its token API (/knoxtoken/api/v1/token)
• fetch useful metrics via JMX

...

<?xml version="1.0" encoding="UTF-8"?>
<topology>
   <name>tokenbased</name>
   <gateway>
      <provider>
         <role>federation</role>
         <name>JWTProvider</name>
         <enabled>true</enabled>
         <param>
            <name>knox.token.audiences</name>
            <value>tokenbased</value>
         </param>
      </provider>
   </gateway>
   <service>
      <role>WEBHDFS</role>
      <url>http://YOUR_HDFS_SERVICE_HOST:20101/webhdfs</url>
   </service>
</topology>

Since the the 'KnoxToken Use Case' tries to use an already acquired Knox DT to run an action I chose to do this as simple as possible: using KnoxShell's HDFSclass we issue an ls command with a KnoxShell session that uses a Knox DT. It is very important that the tokenbased topology comes with the WEBHDFS service for this purpose.

If you plan to have this topology with a different name (or you already have one that uses JWT and has WEBHDFS), please update the perf.test.usecase.knoxtoken.topology.tokenbased configuration accordingly.

How to run

Running the performance tool is as simple as running the following Maven command in the project root:

...

perf.test.usecase.knoxtoken.numOfThreads=10
perf.test.usecase.knoxtoken.testDurationInSecs=21600

Test results

The JSON/YAML test results are generated under target/testResults/[json|yaml]:

• heapGauges.YYYY-MM-DD.[json|yaml]. Sample:

  Code Block
  --- metrics:name=heap.init,type=gauges: Number: 268435456 Value: 268435456 metrics:name=heap.usage,type=gauges: Number: 0.04556474316352607 Value: 0.04556474316352607 metrics:name=heap.committed,type=gauges: Number: 537919488 Value: 537919488 metrics:name=heap.max,type=gauges: Number: 3817865216 Value: 3817865216 metrics:name=heap.used,type=gauges: Number: 174097040 Value: 174094952

  
  

• responseTimes.YYYY-MM-DD.[json|yaml]. Sample:

  Code Block
  --- acquireResponseTimes: _data: - 204 - 204 - 204 - 18 - 19 - 21 - 19 - 16 - 13 - 11 - 10 - 14 - 13 - 11 - 12 - 11 - 14 - 10 mode: - 11.0 - 204.0 min: 10.0 max: 204.0 mean: 45.777777777777786 geometricMean: 21.52096561302238 renewResponseTimes: _data: - 449 mode: - 449.0 min: 449.0 max: 449.0 mean: 449.0 geometricMean: 449.0000000000001

  
  

• timers.YYYY-MM-DD.[json|yaml]. Sample:

  Code Block

  --- metrics:name=client./gateway/sandbox/knoxtoken/api/.POST-requests,type=timers: Mean: 402.24274066722455 StdDev: 91.5058294023066 "75thPercentile": 418.193352 "98thPercentile": 628.076518 RateUnit: "events/second" "95thPercentile": 628.076518 "99thPercentile": 628.076518 Max: 628.076518 Count: 9 FiveMinuteRate: 0.00943518979245522 "50thPercentile": 404.177556 MeanRate: 0.0036809848722434983 Min: 115.866694 OneMinuteRate: 0.01908228359419756 DurationUnit: "milliseconds" "999thPercentile": 628.076518 FifteenMinuteRate: 0.017525836234274554 metrics:name=client./gateway/sandbox/knoxtoken/api/.GET-requests,type=timers: Mean: 11.77009795297069 StdDev: 8.295945542138911 "75thPercentile": 12.517118 "98thPercentile": 39.362919999999995 RateUnit: "events/second" "95thPercentile": 31.40917 "99thPercentile": 39.362919999999995 Max: 1183.7718479999999 Count: 77 FiveMinuteRate: 0.11133648462048056 "50thPercentile": 8.21257 MeanRate: 0.03134296740421139 Min: 6.442307 OneMinuteRate: 0.2513294281670072 DurationUnit: "milliseconds" "999thPercentile": 39.362919999999995 FifteenMinuteRate: 0.12650178822825908 metrics:name=client./gateway/tokenbased/webhdfs/v1.GET-requests,type=timers: Mean: 529.0935038293519 StdDev: 37.26100859914753 "75thPercentile": 540.324314 "98thPercentile": 682.462197 RateUnit: "events/second" "95thPercentile": 540.324314 "99thPercentile": 682.462197 Max: 4007.184354 Count: 8 FiveMinuteRate: 0.012584932473001562 "50thPercentile": 520.5765289999999 MeanRate: 0.003278659499845765 Min: 462.182908 OneMinuteRate: 0.033547665513349575 DurationUnit: "milliseconds" "999thPercentile": 682.462197 FifteenMinuteRate: 0.018536358285136553

  
  

• tokenStateStatistics.YYYY-MM-DD.[json|yaml]. Sample:

  Code Block
  --- metrics:name=TokenStateService,type=Statistics: KeystoreInteractions: removeAlias: 11 saveAlias: 25 getAlias: 41 GatewayCredentialsFileSize: 89299 NumberOfTokensAdded: 89 NumberOfTokensRenewed: 11