Uploaded image for project: 'Core Server'
  1. Core Server
  2. SERVER-50443

When j:false, performance degradation from 4.0 to 4.2 for 1-node replica sets

    • Type: Icon: Bug Bug
    • Resolution: Done
    • Priority: Icon: Major - P3 Major - P3
    • None
    • Affects Version/s: 4.2.8
    • Component/s: Performance
    • None
    • ALL
    • Hide

      Tests are run using https://github.com/idealo/mongodb-performance-test.git:

      java -jar ./latest-version/mongodb-performance-test.jar -m insert -o 10000000 -t 100 -db test -c perf -port 28888 -h mongoserver

      When running 4.0.6 standalone server with:
      {{docker run -ti -v /srv/mongo/test-mongo:/srv/mongodb/ --net host
      mongo:4.0.6 mongod --bind_ip_all --port=28888 --storageEngine=wiredTiger
      --dbpath=/srv/mongodb/ --journal --oplogSize=8192
      --enableMajorityReadConcern=0}}

      I got:
      mean rate = 174014.88 calls/second

      When I start as replica set (no other replicas added, just start with replSet and call rs.initiate():
      {{docker run -ti -v /srv/mongo/test-mongo:/srv/mongodb/ --net host
      mongo:4.0.6 mongod --bind_ip_all --port=28888 --storageEngine=wiredTiger
      --dbpath=/srv/mongodb/ --journal --oplogSize=8192
      --enableMajorityReadConcern=0 --replSet wtf}}

      I got:
      mean rate = 44243.62 calls/second

      Huge drop, but fine, RS is enabled. Now let’s try it with 4.2.8:

      {{docker run -ti -v /srv/mongo/test-mongo:/srv/mongodb/ --net
      host mongo:4.2.8 mongod --bind_ip_all --port=28888
      --dbpath=/srv/mongodb/ --enableMajorityReadConcern=0:}}

      mean rate = 167455.01 calls/second

      and same with replset enabled:
      {{docker run -ti -v /srv/mongo/test-mongo:/srv/mongodb/ --net host
      mongo:4.2.8 mongod --bind_ip_all --port=28888 --dbpath=/srv/mongodb/
      --enableMajorityReadConcern=0 --replSet wtf:}}

      mean rate = 10868.81 calls/second

      Huge drop to 10kcalls/second.

       

      I've made git bisect and identified these commits which drags performance down the most:

      b2384a85f62431d88ac01067ae4d1d3dbb4fede5 mean rate = 103095,21 calls/second

      4cee07d8a97bb0663e7bfbc3f2e1fbf539140adf is the first bad commit
      commit 4cee07d8a97bb0663e7bfbc3f2e1fbf539140adf
      Author: Judah Schvimer <judah@mongodb.com>
      Date: Mon May 21 18:52:55 2018 -0400

      SERVER-35113 Allow single voting primaries to advance stable timestamp even when last applied does not advance

      :040000 040000 948822d607a4ae02c4b18c9eac471e123fae892d 66d95e2b3a7fbdfda5aff07b58799dc8af359f3e M jstests
      :040000 040000 053e1a2ae25ef2e2599f2f89ad7a580a6509b2b3 cbf9ce44531ec2908fc8f5ee8cc0e569356dfffb M src

      drags down to mean rate = 22841,24 calls/second

       

      Second:

      >>>>>> bcaa4eb01c mean rate = 37827,93 calls/second

      e4b08de746ef4e472bfdd4790fdb2b89d5a62f1a is the first bad commit
      commit e4b08de746ef4e472bfdd4790fdb2b89d5a62f1a
      Author: Lingzhi Deng <lingzhi.deng@mongodb.com>
      Date: Mon Apr 29 12:09:48 2019 -0400

      SERVER-40462: Disallow transactions on shard servers with writeConcernMajorityJournalDefault = false

      :040000 040000 e98acae85e9e9e09dc773dd464be513376791c96 bda8f41ecf04f82c70c8c2bc42dcdc3375b9e656 M jstests
      :040000 040000 0819f92e0a939abe440863dcc060aa48b7fcb128 752ad757384381fc777490bbca2c4f570c7aeec4 M src

      >>>>>>>> bcde4c4fd3 mean rate = 26577,37 calls/second

       

      3rd:

      86efdd6b047927d3a37577a6033bacc709c7ae89 is the first bad commit
      commit 86efdd6b047927d3a37577a6033bacc709c7ae89
      Author: Ben Caimano <ben.caimano@mongodb.com>
      Date: Mon Dec 16 21:39:41 2019 +0000

      SERVER-43945 Expose out of order latch acquisitions in serverStatus

      This review does several related things:

      • Modifies the Mutex type to have one Identity object per invocation context
      • Adds a latchAnalysis field to serverStatus
      • Officially turns on CI-only failure for acq level violations
      • Changes a few acq level declarations to be more permissive

      :040000 040000 cc0ee8580816f86084bcdab28f8ae92ec9da2bf1 a77afd98d1fa56c79b71de4ec7835e4905a9c664 M jstests
      :040000 040000 c9d93da9281647969dea4d505776120c9262a4d9 c2c56acf209149bdede17b97e7c3aadc2c882d1d M src

      574b62017cad311690eba012222459ff5e41e8c5 mean rate = 11348,16 calls/second

      Show
      Tests are run using https://github.com/idealo/mongodb-performance-test.git: java -jar ./latest-version/mongodb-performance-test.jar -m insert -o 10000000 -t 100 -db test -c perf -port 28888 -h mongoserver When running 4.0.6 standalone server with: {{docker run -ti -v /srv/mongo/test-mongo:/srv/mongodb/ --net host mongo:4.0.6 mongod --bind_ip_all --port=28888 --storageEngine=wiredTiger --dbpath=/srv/mongodb/ --journal --oplogSize=8192 --enableMajorityReadConcern=0}} I got: mean rate = 174014.88 calls/second When I start as replica set (no other replicas added, just start with replSet and call rs.initiate(): {{docker run -ti -v /srv/mongo/test-mongo:/srv/mongodb/ --net host mongo:4.0.6 mongod --bind_ip_all --port=28888 --storageEngine=wiredTiger --dbpath=/srv/mongodb/ --journal --oplogSize=8192 --enableMajorityReadConcern=0 --replSet wtf}} I got: mean rate = 44243.62 calls/second Huge drop, but fine, RS is enabled. Now let’s try it with 4.2.8: {{docker run -ti -v /srv/mongo/test-mongo:/srv/mongodb/ --net host mongo:4.2.8 mongod --bind_ip_all --port=28888 --dbpath=/srv/mongodb/ --enableMajorityReadConcern=0:}} mean rate = 167455.01 calls/second and same with replset enabled: {{docker run -ti -v /srv/mongo/test-mongo:/srv/mongodb/ --net host mongo:4.2.8 mongod --bind_ip_all --port=28888 --dbpath=/srv/mongodb/ --enableMajorityReadConcern=0 --replSet wtf:}} mean rate = 10868.81 calls/second Huge drop to 10kcalls/second.   I've made git bisect and identified these commits which drags performance down the most: b2384a85f62431d88ac01067ae4d1d3dbb4fede5 mean rate = 103095,21 calls/second 4cee07d8a97bb0663e7bfbc3f2e1fbf539140adf is the first bad commit commit 4cee07d8a97bb0663e7bfbc3f2e1fbf539140adf Author: Judah Schvimer <judah@mongodb.com> Date: Mon May 21 18:52:55 2018 -0400 SERVER-35113 Allow single voting primaries to advance stable timestamp even when last applied does not advance :040000 040000 948822d607a4ae02c4b18c9eac471e123fae892d 66d95e2b3a7fbdfda5aff07b58799dc8af359f3e M jstests :040000 040000 053e1a2ae25ef2e2599f2f89ad7a580a6509b2b3 cbf9ce44531ec2908fc8f5ee8cc0e569356dfffb M src drags down to mean rate = 22841,24 calls/second   Second: >>>>>> bcaa4eb01c mean rate = 37827,93 calls/second e4b08de746ef4e472bfdd4790fdb2b89d5a62f1a is the first bad commit commit e4b08de746ef4e472bfdd4790fdb2b89d5a62f1a Author: Lingzhi Deng <lingzhi.deng@mongodb.com> Date: Mon Apr 29 12:09:48 2019 -0400 SERVER-40462 : Disallow transactions on shard servers with writeConcernMajorityJournalDefault = false :040000 040000 e98acae85e9e9e09dc773dd464be513376791c96 bda8f41ecf04f82c70c8c2bc42dcdc3375b9e656 M jstests :040000 040000 0819f92e0a939abe440863dcc060aa48b7fcb128 752ad757384381fc777490bbca2c4f570c7aeec4 M src >>>>>>>> bcde4c4fd3 mean rate = 26577,37 calls/second   3rd: 86efdd6b047927d3a37577a6033bacc709c7ae89 is the first bad commit commit 86efdd6b047927d3a37577a6033bacc709c7ae89 Author: Ben Caimano <ben.caimano@mongodb.com> Date: Mon Dec 16 21:39:41 2019 +0000 SERVER-43945 Expose out of order latch acquisitions in serverStatus This review does several related things: Modifies the Mutex type to have one Identity object per invocation context Adds a latchAnalysis field to serverStatus Officially turns on CI-only failure for acq level violations Changes a few acq level declarations to be more permissive :040000 040000 cc0ee8580816f86084bcdab28f8ae92ec9da2bf1 a77afd98d1fa56c79b71de4ec7835e4905a9c664 M jstests :040000 040000 c9d93da9281647969dea4d505776120c9262a4d9 c2c56acf209149bdede17b97e7c3aadc2c882d1d M src 574b62017cad311690eba012222459ff5e41e8c5 mean rate = 11348,16 calls/second
    • 0

      Hi all,

      I’m currently running mongodb 4.0.6 and playing with upgrade to 4.2.8. After some benchmarks I’ve noticed really strange and huge performance degradation.

      Can anyone give me some insight what’s going on?

      I've made

            Assignee:
            kelsey.schubert@mongodb.com Kelsey Schubert
            Reporter:
            nextsux@gmail.com Martin Mlynář
            Votes:
            0 Vote for this issue
            Watchers:
            9 Start watching this issue

              Created:
              Updated:
              Resolved: