Overview

Test goal is to assess performance of the harvesting of 10M inventory records with metadataPrefix set to marc21, find possible issues, bottlenecks.

Ticket: https://folio-org.atlassian.net/browse/PERF-145

Summary

• Test lasted for 4 days and 6 hours.

• Average response time per request degrades from 1s to 20 s at the end of the test.

• Test finished after response time reached 20 seconds and 500 error was received.

• Query plans comparison showed that execution time of the query depends directly on the OFFSET parameter.

• No data consumptions issues were found.

Recommendations & Jiras

Issues to be reported:

• SLA issue: process time exceeds 6 hours requirement

• Response time degradation issue

• Internal server error recieved during the test

Ticket: https://folio-org.atlassian.net/browse/MODOAIPMH-464

Test Runs 

Results

Total requests sent during the test: 35 642 requests

Inventory records received: about 3,5M records (expected number – 10M records)

Response Times 

Grafana dashboard:

http://carrier-io.int.folio.ebsco.com/grafana/d/elIt9zCnz/jmeter-performance-test-copy?orgId=1&from=1671704123876&to=1672121479913&var-percentile=95&var-test_type=longevity&var-test=oai_testm_mg_v3&var-env=int&var-grouping=1s&var-low_limit=250&var-high_limit=750&var-db_name=jmeter&var-sampler_type=REQUEST

Open

Open

Memory Utilization

Open

CPU Utilization 

Open

Open

Error messages

00:31:03 [503635/oai] [fs09000000] [2ac43d6a-3945-4d99-87ee-7a8281238572] [mod_oai_pmh] ERROR ractGetRecordsHelper Exception getting ListRecords.

00:31:03 [503635/oai] [fs09000000] [2ac43d6a-3945-4d99-87ee-7a8281238572] [mod_oai_pmh] ERROR ractGetRecordsHelper ListRecords response from SRS status code: Internal Server Error: 500.

00:31:03 [] [] [] [] ERROR OaiPmhHandler        Error in the response from repository: status code - 500, response status message - Internal Server Error Internal Server Error

00:31:03 [165801/source-storage] [fs09000000] [2ac43d6a-3945-4d99-87ee-7a8281238572] [mod_source_record_storage] ERROR ExceptionHelper      null

io.netty.channel.StacklessClosedChannelException: null

RDS CPU Utilization 

Open

Open

Database load

Top SQL queries list

Open

Top SQL query

with "cte" as (select count(*) from "records_lb" where (("records_lb"."state" = $1::"record_state" or "records_lb"."state" = $2::"record_state" or "records_lb"."leader_record_status" = $3 or "records_lb"."leader_record_status" = $4 or "records_lb"."leader_record_status" = $5) and "records_lb"."updated_date" <= cast($6 as timestamp with time zone) and "records_lb"."record_type" = $7::"record_type" and "records_lb"."leader_record_status" is not null)) select "records_lb"."id", "records_lb"."snapshot_id", "records_lb"."matched_id", "records_lb"."generation", "records_lb"."record_type", "records_lb"."external_id", "records_lb"."state", "records_lb"."leader_record_status", "records_lb"."order", "records_lb"."suppress_discovery", "records_lb"."created_by_user_id", "records_lb"."created_date", "records_lb"."updated_by_user_id", "records_lb"."updated_date", "records_lb"."external_hrid", "marc_records_lb"."content", "count" from "records_lb" left outer join "marc_records_lb" on "records_lb"."id" = "marc_records_lb"."id" right outer join (select * from "cte") as "alias_80949780" on 1 = 1 where (("records_lb"."state" = $8::"record_state" or "records_lb"."state" = $9::"record_state" or "records_lb"."leader_record_status" = $10 or "records_lb"."leader_record_status" = $11 or "records_lb"."leader_record_status" = $12) and "records_lb"."updated_date" <= cast($13 as timestamp with time zone) and "records_lb"."record_type" = $14::"record_type" and "records_lb"."leader_record_status" is not null) order by "records_lb"."id" asc limit $15 offset $16

Query plan comparison

Left side: date - 2023-01-07 11:00:01, offset – 2 220 000

Right side: date - 2023-01-09 22:41:01, offset – 3 526 100

Open

Execution time - 13s (offset – 2 220 000), 21s (offset – 3 526 100).

Appendix

Infrastructure

PTF -environment ncp2 [

• 12 m6i.2xlarge EC2 instances located in us-east

• 2 instances of db.r6.xlarge database instances, one reader and one writer 

• MSK ptf-kakfa-3

  • 4 m5.2xlarge brokers in 2 zones

  • Apache Kafka version 2.8.0

  • EBS storage volume per broker 300 GiB

  • auto.create.topics.enable=true

  • log.retention.minutes=480

  • default.replication.factor=3

Modules memory and CPU parameters

Methodology/Approach

Check that metadataPrefix is set to marc21 in script (oai_testm_mg_v3).

Run test from Jenkins or directly in VM.