summaryrefslogtreecommitdiffstats
path: root/docs/development/devtools/drools-s3p.rst
diff options
context:
space:
mode:
authorRam Krishna Verma <ram_krishna.verma@bell.ca>2022-04-27 14:17:38 +0000
committerGerrit Code Review <gerrit@onap.org>2022-04-27 14:17:38 +0000
commitd7907d8bf16aac75018e36968a3d74291b9e64bb (patch)
tree16eb0f990730069d40d2f5a008f08a16a003ab1a /docs/development/devtools/drools-s3p.rst
parent78430075624c287b4a25e7b0b39622f9bf3313a3 (diff)
parent9012afea38cc74e8c41b61a5f2a242d78057d2c5 (diff)
Merge "s3p drools documentation"
Diffstat (limited to 'docs/development/devtools/drools-s3p.rst')
-rw-r--r--docs/development/devtools/drools-s3p.rst168
1 files changed, 23 insertions, 145 deletions
diff --git a/docs/development/devtools/drools-s3p.rst b/docs/development/devtools/drools-s3p.rst
index 22c1b47d..571e09a3 100644
--- a/docs/development/devtools/drools-s3p.rst
+++ b/docs/development/devtools/drools-s3p.rst
@@ -10,32 +10,20 @@
Policy Drools PDP component
~~~~~~~~~~~~~~~~~~~~~~~~~~~
-Both the Performance and the Stability tests were executed against an ONAP installation in the policy-k8s tenant
-in the windriver lab, from an independent VM running the jmeter tool to inject the load.
+Both the Performance and the Stability tests were executed against an ONAP installation in the Policy tenant
+in the UNH lab, from the admin VM running the jmeter tool to inject the load.
General Setup
*************
-The installation runs the following components in a single VM:
-
-- AAF
-- AAI
-- DMAAP
-- POLICY
-
-The VM has the following hardware spec:
-
-- 126GB RAM
-- 12 VCPUs
-- 155GB Ephemeral Disk
-
-Jmeter is run from a different VM with the following configuration:
+Agent VMs in this lab have the following configuration:
- 16GB RAM
-- 8 VCPUs
-- 155GB Ephemeral Disk
+- 8 VCPU
-The drools-pdp container uses the JVM memory settings from a default OOM installation.
+Jmeter is run from the admin VM.
+
+The drools-pdp container uses the JVM memory and CPU settings from the default OOM installation.
Other ONAP components exercised during the stability tests were:
@@ -51,22 +39,6 @@ The following components are simulated during the tests.
- APPC responses for the vCPE and vFW use cases.
- AAI to answer queries for the use cases under test.
-SO, and AAI actors were simulated within the PDP-D JVM by enabling the
-feature-controlloop-utils before running the tests.
-
-PDP-D Setup
-***********
-
-The kubernetes charts were modified previous to the installation
-to add the following script that enables the controlloop-utils feature:
-
-.. code-block:: bash
-
- oom/kubernetes/policy/charts/drools/resources/configmaps/features.pre.sh:
-
- #!/bin/sh
- sh -c "features enable controlloop-utils"
-
Stability Test of Policy PDP-D
******************************
@@ -82,132 +54,38 @@ The tests focused on the following use cases:
For 72 hours the following 5 scenarios ran in parallel:
- vCPE success scenario
-- vCPE failure scenario (failure returned by simulated APPC recipient through DMaaP).
- vDNS success scenario.
-- vDNS failure scenario (failure by introducing in the DCAE ONSET a non-existent vserver-name reference).
- vFirewall success scenario.
+- vCPE failure scenario (simulates a failure scenario returned by simulated APPC recipient through DMaaP).
+- vDNS failure scenario (simulates a failure by introducing in the DCAE ONSET a non-existent vserver-name reference).
Five threads ran in parallel, one for each scenario, back to back with no pauses. The transactions were initiated
by each jmeter thread group. Each thread initiated a transaction, monitored the transaction, and
as soon as the transaction ending was detected, it initiated the next one.
-JMeter was run in a docker container with the following command:
-
-.. code-block:: bash
-
- docker run --interactive --tty --name jmeter --rm --volume $PWD:/jmeter -e VERBOSE_GC="" egaillardon/jmeter-plugins --nongui --testfile s3p.jmx --loglevel WARN
-
-The results were accessed by using the telemetry API to gather statistics:
-
-
-vCPE Success scenario
-=====================
-
-ControlLoop-vCPE-48f0c2c3-a172-4192-9ae3-052274181b6e:
-
-.. code-block:: bash
-
- # Times are in milliseconds
-
- Control Loop Name: ControlLoop-vCPE-48f0c2c3-a172-4192-9ae3-052274181b6e
- Number of Transactions Executed: 114007
- Number of Successful Transactions: 112727
- Number of Failure Transactions: 1280
- Average Execution Time: 434.9942021103967 ms.
-
-
-vCPE Failure scenario
-=====================
-
-ControlLoop-vCPE-Fail:
+The results are illustrated on the following graphs:
-.. code-block:: bash
-
- # Times are in milliseconds
-
- Control Loop Name: ControlLoop-vCPE-Fail
- Number of Transactions Executed: 114367
- Number of Successful Transactions: 114367 (failure transactions are expected)
- Number of Failure Transactions: 0 (success transactions are not expected)
- Average Execution Time: 433.61750330077734 ms.
-
-
-vDNS Success scenario
-=====================
-
-ControlLoop-vDNS-6f37f56d-a87d-4b85-b6a9-cc953cf779b3:
-
-.. code-block:: bash
-
- # Times are in milliseconds
-
- Control Loop Name: ControlLoop-vDNS-6f37f56d-a87d-4b85-b6a9-cc953cf779b3
- Number of Transactions Executed: 237512
- Number of Successful Transactions: 229532
- Number of Failure Transactions: 7980
- Average Execution Time: 268.028794334602 ms.
-
-
-vDNS Failure scenario
-=====================
-
-ControlLoop-vDNS-Fail:
-
-.. code-block:: bash
-
- # Times are in milliseconds
-
- Control Loop Name: ControlLoop-vDNS-Fail
- Number of Transactions Executed: 1957987
- Number of Successful Transactions: 1957987 (failure transactions are expected)
- Number of Failure Transactions: 0 (success transactions are not expected)
- Average Execution Time: 39.369322166081794
-
-
-vFirewall Success scenario
-==========================
-
-ControlLoop-vFirewall-d0a1dfc6-94f5-4fd4-a5b5-4630b438850a:
-
-.. code-block:: bash
-
- # Times are in milliseconds
-
- Control Loop Name: ControlLoop-vFirewall-d0a1dfc6-94f5-4fd4-a5b5-4630b438850a
- Number of Transactions Executed: 120308
- Number of Successful Transactions: 118895
- Number of Failure Transactions: 1413
- Average Execution Time: 394.8609236293513 ms.
+.. image:: images/s3p-drools-1.png
+.. image:: images/s3p-drools-2.png
+.. image:: images/s3p-drools-3.png
+.. image:: images/s3p-drools-4.png
Commentary
==========
-There has been a degradation of performance observed in this release
-when compared with the previous one.
-Approximately 1% of transactions were not completed as expected for
-some use cases. Average Execution Times are extended as well.
-The unexpected results seem to point in the direction of the
-interactions of the distributed locking feature with the database.
-These areas as well as the conditions for the test need to be investigated
-further.
+There is around 1% unexpected failures during the 72-hour run. This can also be seen in the
+final output of jmeter:
.. code-block:: bash
- # Common pattern in the audit.log for unexpected transaction completions
-
- a8d637fc-a2d5-49f9-868b-5b39f7befe25||ControlLoop-vFirewall-d0a1dfc6-94f5-4fd4-a5b5-4630b438850a|
- policy:usecases:[org.onap.policy.drools-applications.controlloop.common:controller-usecases:1.9.0:usecases]|
- 2021-10-12T19:48:02.052+00:00|2021-10-12T19:48:02.052+00:00|0|
- null:operational.modifyconfig.EVENT.MANAGER.FINAL:1.0.0|dev-policy-drools-pdp-0|
- ERROR|400|Target Lock was lost|||VNF.generic-vnf.vnf-name||dev-policy-drools-pdp-0||
- dev-policy-drools-pdp-0|microservice.stringmatcher|
- {vserver.prov-status=ACTIVE, vserver.is-closed-loop-disabled=false,
- generic-vnf.vnf-name=fw0002vm002fw002, vserver.vserver-name=OzVServer}||||
- INFO|Session org.onap.policy.drools-applications.controlloop.common:controller-usecases:1.9.0:usecases|
-
- # The "Target Lock was lost" is a common message error in the unexpected results.
+ summary = 37705505 in 72:00:56 = 145.4/s Avg: 30 Min: 0 Max: 20345 Err: 360852 (0.96%)
+The 1% errors were found to be related to the nature of the run, where each one of the 5 use case
+threads run without pauses starting one after the other a new round of their assigned control loop.
+It has been found that at times, the release time of the lock (which requires DB operations) outruns
+the initiation of the next control loop (using the same resource), therefore the newly initiated control
+loop fails. In reality, this scenario with the same resource being used back to back in consecutive control
+loop rounds will be unlikely.
-END-OF-DOCUMENT