Overview

This document contains the results of testing Data Export (MARC BIB) on Quesnelia release with Data Export tests for 1k, 100k, 500k. Three csv files were prepared to run Data Export with Default instances export job profile and srs - holdings and items job profiles.

Ticket: PERF-822 - Getting issue details... STATUS

Summary

• DE jobs perform dramatically better in Quesnelia release if to compare with Poppy. No issues with token. All jobs with file's volume 1000, 100k, 500k records completed successfully.
• The improvement varies from file size or job profile - from 4 to 9 times better duration. Additional test was conducted with job profile prepared by script to check consistency of results.
• The Average CPU utilization for mod-data-export depends on file size and job profile. Exporting 100k records- 63% in Default and 92% in custom job profile. Exporting 500k - Default instances export job profile- 434%, srs - holdings and items- 296%
• Average Memory consumption for mod-data-export was close to 100%. Almost the same as in Poppy release.
• Average DB utilization - 17% with 100k and 33% with 500k. DB connections - 1360 instead of 200 in Poppy. 

Recommendations & Jiras

• Consider increasing of cpu for mod-data-export module to smoothen spikes with big size files.

Test Results

This table contains durations for jobs with 3 job profiles. 

Comparison

This table contains durations comparison between Poppy and Quesnelia releases

Service CPU Utilization

Mod-data-export-b

For Default instances export job profile with 100k file used 92% , during exporting 500k file - 434%.

For job profile "Export for Data Import updates" (created by script) - exporting with 100k - 33%, 500k - 202%. 

For srs - holdings and items job profile - 100k - 63%, 500k - 296%.

Memory Utilization

This graph contains DE related modules.

Mod-data-export-b use 97%, mod-inventory - 55%. The rest was under 50%

All modules memory consumption graph. Here we see a lot of modules that go with more than 50% consumption. 

DB CPU Utilization

DB utilized with Default instances export job profile 100k file - 17%, 500k file - 25%. For Export for Data Import updates(created by script) job profile - 100k file - 17%, 500k file - 38%. 

For srs - holdings and items job profile - 100k file - 17%, 500k file - 33%. 

DB Connections

DB connections - 1260 in Average. High peaks go with 1360.

DB Load

Default instances export job profile

Export for Data Import updates(test)1 job profile

srs - holdings and items job profile

SQL queries

Default instances export job profile

Export for Data Import updates(test)1 job profile

srs - holdings and items job profile

Top-SQL statement: 

Default instances export job profile

autovacuum: VACUUM fs09000000_mod_data_export.job_executions_export_ids

autovacuum: VACUUM fs09000000_mod_data_export.job_executions_export_ids

select ie1_0.id,ie1_0.holdings_record_id,ie1_0.jsonb from v_item ie1_0 where ie1_0.holdings_record_id in ($1)

select hre1_0.id,hre1_0.instance_id,hre1_0.jsonb from v_holdings_record hre1_0 where hre1_0.instance_id=$1

INSERT INTO job_executions_export_ids (job_execution_id, instance_id) VALUES ($1, $2) ON CONFLICT DO NOTHING

Errors / Additional information

Note: inconsistency of data or duplicated records in the file with instances may lead to job failures. Zip file generates on UI side if number of records more than 100.000. 

Methodology/Approach

3 files were prepared with query: SELECT id FROM [tenant_id]_mod_inventory_storage.instance where jsonb->>'source'='MARC' LIMIT 1000|100000|500000;

All tests were carried out sequentially with each job profile on main tenant fs09000000. 

To get status and time range for export jobs the query used: 

select exported as filesize,completed_date - started_date as duration,job_profile_name,status as status,started_date,completed_date
FROM [tenant]_mod_data_export.job_executions
where job_profile_name = 'srs - holdings and items'
ORDER BY completed_date DESC

select exported as filesize,completed_date - started_date as duration,job_profile_name,status as status,started_date,completed_date
FROM [tenant]_mod_data_export.job_executions
where job_profile_name = 'Default instances export job profile'
ORDER BY completed_date DESC

select exported as filesize,completed_date - started_date as duration,job_profile_name,status as status,started_date,completed_date
FROM [tenant]_mod_data_export.job_executions
where job_profile_name = 'Export for Data Import updates(test)1'
ORDER BY completed_date DESC

Infrastructure

PTF -environment qcp1

• 10 m6i.2xlarge EC2 instances located in US East (N. Virginia)us-east-1

• 2 database  instances, writer/reader

  Name	Memory GIB	vCPUs	max_connections
  db.r6g.xlarge	32 GiB	4 vCPUs	2731

  Data set for fs09000000

  • Instances - 25606331
  • Items       - 26779913
  • Holdings - 25576735
• MSK tenant
  
  • 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

QCP1 modules

Data Export related modules