PERF-27 - Getting issue details... STATUS

Overview

In this api, we are checking the performance of placing a holding on a discovery service interface (EDS) running in the Fameflower release - 

 We tested it with 1, 5, 8, and 20 virtual users for 30 minutes.  

Backend:

• mod-patron-4.1.0
• mod-circulation-18.0.9
• mod-circulation-storage-11.0.0
• mod-inventory-storage-19.1.2
• mod-inventory-14.1.3
• mod-feesfines-15.7.2
• okapi-2.38.0
• mod-authtoken-2.4.0
• mod-permissions-5.9.0

Environment:

• 55 back-end modules deployed in 110 ECS services
• 3 okapi ECS services
• 8 m5.large  EC2 instances
• 1  db.r5.xlarge AWS RDS instance
• INFO logging level

High Level Summary

1. The placing a hold request takes about 1 seconds: POST /patron/account/{id}/instance/{instanceId}/hold. 
2. POST /patron/account/{id}/instance/{instanceId}/hold and POST /circulation/requests/instances show almost the same response time. In addition, mod-circulation's native item-level request code is 24.5% faster than mod-circulation's instance-level request code. Seems like the main issue is the mod-circulation instance-level request API, POST /circulation/requests/instances which is called by mod-patron subsequently.
3. JVM profiling shows the slowest operations, totalling more CPU time than other calls: 
   
   1. logging with INFO level - Logging level could be reduced to WARNING or ERROR, but at the cost of having less data to work with should there be a need to troubleshoot. The most CPU consuming methods are org.folio.circulation.support.Result.next and  org.folio.circulation.support.Result.after from org.folio.circulation.support.SingleRecordFetcher, and org.folio.circulation.support.SingleRecordFetcher.fetch method
   2. JSON de/serialization - fasterxml.jackson.databind.ObjectMapper.readValue method of mod-circulation-storage and mod-inventory-storage consumed high CPU resources as there were a lot of JSON decoding, this implementation could be reviewed for optimisation possibilities. The most CPU consuming methods are org.folio.rest.persist.PgUtil.streamGet from org.folio.rest.impl.InstanceStorageAPI.getInstanceStorageInstancesByInstanceId (mod-inventory-storage), org.folio.rest.impl.HoldingsStorageAPI.getHoldingsStorageHoldings(mod-inventory-storage) and org.folio.rest.persist.PostgresClient.doGet from  org.folio.rest.impl.LoansAPI.getLoanStorageLoans  and org.folio.rest.impl.RequestsAPI.getRequestStorageRequests(mod-circulation-storage)
   3. request's validation - need to consider using a more efficient request's validation instead of org.hibernate.validator.internal.engine.ValidatorImpl.validate method because it's one of the slowest operations. The slowest method is org.folio.rest.RestVerticle.route and org.folio.rest.RestVerticle.parseParams
4. mod-circulation-storage and mod-feesfines generated missing indexes warnings, see Recommended Improvements for the JIRAs that were created/covered by this testing effort.

Test Runs

Results

1. An instance-level request to mod-patron 
   

   POST /patron/account/{id}/instance/{instanceId}/hold

   {
       "item": {
   	    "instanceId": "c2063f7c-245a-4c24-b643-67fea831ebcc"
   	},
       "requestDate": "2020-06-01T18:51:24.000+0000",
       "pickupLocationId": "7068e104-aa14-4f30-a8bf-71f71cc15e07"
   }

   Overall time for the placing holds api in seconds 

   	Average	75th %tile
   1 user	933.01 ms	1.10 s
   5 users	1.01 s	1.12 s
   8 users	1.03 s	1.18 s
   20 users	1.33 s	1.47 s

2. mod-patron subsequently calls mod-circulation to place instance-level requests

    POST /circulation/requests/instances

   {
     "requestDate": "2017-07-29T22:25:37Z",
     "requesterId": "21932a85-bd00-446b-9565-46e0c1a5490b",
     "instanceId": "195efae1-588f-47bd-a181-13a2eb437701",
     "requestExpirationDate": "2017-07-25T22:25:37Z",
     "pickupServicePointId": "8359f2bc-b83e-48e1-8a5b-ca1a74e840de"
   }

   Overall time in seconds 

   	Average	75th %tile
   1 user	864.35 ms	1.01 s
   5 users	1.09 s	1.14 s
   8 users	1.15 s	1.18 s
   20 users	1.58 s	1.53 s

3.  An Item-level request 

   POST /circulation/requests

   {
     "requestDate": "2020-06-18T09:23:11.957Z",
     "requesterId": "e96618a9-04ee-4fea-aa60-306a8f4dd89b",
     "requestType": "Hold",
     "fulfilmentPreference": "Hold Shelf",
     "itemId": "0000023d-a95f-4396-8d37-42dda48cdf3d",
     "requestExpirationDate": "2020-06-18T09:23:11.957Z",
     "pickupServicePointId": "d80a641e-50b9-4bd2-99a2-a4a12c8515cb"
   }

   Overall time in seconds 

   	Average	75th %tile
   1 user	807.90 ms	819.00 ms
   8 users	855.92 ms	855.00 ms

                    The comparison table between an instance-level request and an Item-level request to mod-circulation (done 23.07.2020). The item-level request is 24.5% faster than an instance-level request

JVM Profiling

• Overall slow methods (between the modules profiled: mod-patron, mod-circulation, mod-circulation-storage, mod-inventory, mod-inventory-storage)

To see more drill down these methods at: http://ec2-54-165-58-65.compute-1.amazonaws.com/grafana/d/b3iP-8hiz/java-profiling-folio-new?orgId=1&from=1591958944552&to=1591961253729&var-service_tag=mod-patron&var-service_tag=mod-inventory-storage&var-service_tag=mod-inventory&var-service_tag=mod-circulation-storage&var-service_tag=mod-circulation&var-aggregation=10s

• Only slow mod-circulation methods:
  

To see more drill down these methods at: http://ec2-54-165-58-65.compute-1.amazonaws.com/grafana/d/b3iP-8hiz/java-profiling-folio-new?orgId=1&from=1591958944552&to=1591961253729&var-service_tag=mod-circulation&var-aggregation=10s

When drilling down org.apache.logging.log4j.spi.AbstractLogger.logIfEnabled, we get the following tree. To see more click here: