Recommended Maximum File Sizes and Configuration
- Ann-Marie Breaux (Deactivated)
- Kateryna Senchenko
- Aliaksandr Fedasiuk
Owned by Ann-Marie Breaux (Deactivated)
Please work with your systems office or hosting to provider to ensure these configurations are in place. After more performance, reliability, and stability work, these recommendations will be re-evaluated.
For additional information, please see the following:
MARC Bibs: Performance Testing Data Import (Nolana)
Folijet - Morning Glory Snapshot Performance testing
MARC Bibs: Performance Rancher testing (Lotus)
MARC Bibs: Performance Testing Data Import (Juniper/Kiwi)
MARC Bibs: Performance Testing Data Import (Iris)
MARC Bibs: Profiles used for PTF Testing
Folijet - Orchid Performance Testing Results
Maximum File Size (MG) no background activity, no concurrent import jobs, single tenant
- CREATE (SRS MARC, Instances, Holdings, Items): 100,000 MARC Bib records (details)
- UPDATE (SRS MARC, Instances, Holdings, Items): 50,000 MARC Bib records
- For results with background activities (Check-in/Check-out), concurrent jobs and multi-tenant set up please refer to results from PTF
Import Statistics, with recommended values no background activity, no concurrent import jobs, single tenant
| Orchid | Morning Glory | Lotus | Juniper and Kiwi | |||||||
| 2 instances | 2 instances | 1 instance | 2 instances | 2 instances | ||||||
| MARC Records | CREATE Duration | UPDATE Duration | CREATE Duration | UPDATE Duration | CREATE Duration | UPDATE Duration | CREATE Duration | UPDATE Duration | CREATE Duration | UPDATE Duration |
|---|---|---|---|---|---|---|---|---|---|---|
| 5000 | 5 min | 7min | 11min | 13min | 16min | 8min | 13min | 15 minutes | 20-30 minutes | |
| 10,000 | 14 min | 16min | 22min | 31min | 40min | 19min | 25min | |||
| 30,000 | 46 min | 1h 31min | 1h 4 min | 45min | 1h 36min | |||||
| 50,000 | 54 min | 59min | 1h42min | 1h 34min | 2h 17min | 1h 21min | 1h 51min | 2h 30min+ | 11+ hours | |
| 100,000 | 3h37min (ongoing investigation) | 2h20min | 2h49min | 3h 21min | 3h 10min | 4h 10min | ||||
Maximum File Sizes (Juniper and Kiwi)
- CREATE Import (SRS MARC, Instances, Holdings, Items): 50,000 MARC records max
- UPDATE Import: 5,000 MARC records max
Import Statistics, with background activity
- With 5-users check-in/out background activities
And concurrent imports by different tenants
- Check-in/out time increases by 50%-100% depending on number of concurrent users.
- Data import takes 2x longer to complete.
| MARC Records | CREATE Duration | UPDATE Duration |
|---|---|---|
| 1,000 | 10 minutes | 10 minutes |
| 5,000 | 30 minutes | 40-60 minutes (depending on complexity of profile) |
| 25,000 | 2-3 hours | 8+ hours |
| 50,000 | 5+ hours | 22+ hours |
Import Statistics, no background activity, no concurrent import jobs
| MARC Records | CREATE Duration | UPDATE Duration |
|---|---|---|
| 1,000 | 5 minutes | 5 minutes |
| 5,000 | 15 minutes | 20-30 minutes (depending on complexity of profile) |
| 25,000 | 60-80 minutes | 4+ hours |
| 50,000 | 150+ minutes | 11+ hours |
Note: For Lotus 500,000 Create Duration -15h 37min (details)
Key Settings and Configurations
mod-data-import
Properties related to file upload that should be set at mod-configuration are described in the doc https://github.com/folio-org/mod-data-import#module-properties-to-set-up-at-mod-configuration
System property that can be adjusted | Default value |
|---|---|
| file.processing.marc.raw.buffer.chunk.size | 50 |
| file.processing.marc.json.buffer.chunk.size | 50 |
| file.processing.marc.xml.buffer.chunk.size | 10 |
| file.processing.edifact.buffer.chunk.size | 10 |
For releases prior to Kiwi it is recommended to set the file.processing.buffer.chunk.size property to 5 in order to prevent mod-source-record-storage from crashing with OOM during an Update import of 5,000 records. The property can be set to 1 to allow Update import of 10,000 records.
mod-source-record-manager
System property that can be adjusted | Default value | Comment | |
|---|---|---|---|
| srm.kafka.RawMarcChunkConsumer.instancesNumber | 1 | ||
| srm.kafka.StoredMarcChunkConsumer.instancesNumber | 1 | ||
| srm.kafka.DataImportConsumersVerticle.instancesNumber | 1 | ||
| srm.kafka.DataImportJournalConsumersVerticle.instancesNumber | 1 | ||
| srm.kafka.RawChunksKafkaHandler.maxDistributionNum | 100 | ||
| srm.kafka.CreatedRecordsKafkaHandler.maxDistributionNum | 100 | ||
| srm.kafka.DataImportConsumer.loadLimit | 5 | ||
| security.protocol | PLAINTEXT | ||
| ssl.protocol | TLSv1.2 | ||
| ssl.key.password | - | ||
| ssl.keystore.location | - | ||
| ssl.keystore.password | - | ||
| ssl.keystore.type | JKS | ||
| ssl.truststore.location | - | ||
| ssl.truststore.password | - | ||
| ssl.truststore.type | JKS | ||
| di.flow.control.max.simultaneous.records | 50 | Defines how many records can be processed by the system simultaneously | Morning Glory |
| di.flow.control.records.threshold | 25 | Defines how many records from previous batch must be processed before throwing new records to pipeline | Morning Glory |
| di.flow.control.enable | true | Allows for single record imports to be processed while larger file imports are being processed | Morning Glory |
di.flow.control.reset.state.interval | PT5M | Time between resetting state of flow control. By default it triggered each 5 mins. | Morning Glory |
| kafka.producer.batch.size | 16*1024 | Producer will attempt to batch records together into fewer requests whenever multiple records are being sent to the same partition. This helps performance on both the client and the server. This configuration controls the default batch size in bytes | Nolana |
| kafka.producer.linger.ms | 0 | Producer groups together any records that arrive in between request transmissions into a single batched request. Normally this occurs only under load when records arrive faster than they can be sent out. However in some circumstances the client may want to reduce the number of requests even under moderate load. This setting accomplishes this by adding a small amount of artificial delay that is, rather than immediately sending out a record the producer will wait for up to the given delay to allow other records to be sent so that the sends can be batched together | Nolana |
| kafka.producer.request.timeout.ms | 30000 | The configuration controls the maximum amount of time the client will wait for the response of a request. If the response is not received before the timeout elapses the client will resend the request if necessary or fail the request if retries are exhausted | Nolana |
| kafka.producer.delivery.timeout.ms | 120000 | An upper bound on the time to report success or failure after a call to send() returns. This limits the total time that a record will be delayed prior to sending, the time to await acknowledgement from the broker (if expected), and the time allowed for retriable send failures. kafka.producer.delivery.timeout.ms must be equal to or larger than kafka.producer.linger.ms + kafka.producer.request.timeout.ms | Nolana |
| kafka.producer.retry.backoff.ms | 100 | Time to wait before attempting to retry a failed request to a given topic partition | Nolana |
mod-source-record-storage
System property that can be adjusted | Default value |
|---|---|
| srs.kafka.ParsedMarcChunkConsumer.instancesNumber | 1 |
| srs.kafka.DataImportConsumer.instancesNumber | 1 |
| srs.kafka.ParsedRecordChunksKafkaHandler.maxDistributionNum | 100 |
| srs.kafka.DataImportConsumer.loadLimit | 5 |
| srs.kafka.DataImportConsumerVerticle.maxDistributionNum | 100 |
| srs.kafka.ParsedMarcChunkConsumer.loadLimit | 5 |
| security.protocol | PLAINTEXT |
| ssl.protocol | TLSv1.2 |
| ssl.key.password | - |
| ssl.keystore.location | - |
| ssl.keystore.password | - |
| ssl.keystore.type | - |
| ssl.truststore.location | JKS |
| ssl.truststore.password | - |
| ssl.truststore.type | JKS |
mod-inventory
System property that can be adjusted | Default value |
|---|---|
| inventory.kafka.DataImportConsumerVerticle.instancesNumber | 3 |
| inventory.kafka.MarcBibInstanceHridSetConsumerVerticle.instancesNumber | 3 |
| inventory.kafka.DataImportConsumer.loadLimit | 5 |
| inventory.kafka.DataImportConsumerVerticle.maxDistributionNumber | 100 |
| inventory.kafka.MarcBibInstanceHridSetConsumer.loadLimit | 5 |
| security.protocol | PLAINTEXT |
| ssl.protocol | TLSv1.2 |
| ssl.key.password | - |
| ssl.keystore.location | - |
| ssl.keystore.password | - |
| ssl.keystore.type | JKS |
| ssl.truststore.location | - |
| ssl.truststore.password | - |
| ssl.truststore.type | JKS |
mod-invoice
System property that can be adjusted | Default value |
|---|---|
| mod.invoice.kafka.DataImportConsumerVerticle.instancesNumber | 1 |
| mod.invoice.kafka.DataImportConsumer.loadLimit | 5 |
| mod.invoice.kafka.DataImportConsumerVerticle.maxDistributionNumber | 100 |
| dataimport.consumer.verticle.mandatory | false should be set to true in order to fail the module at start-up if data import Kafka consumer creation failed |
| security.protocol | PLAINTEXT |
| ssl.protocol | TLSv1.2 |
| ssl.key.password | - |
| ssl.keystore.location | - |
| ssl.keystore.password | - |
| ssl.keystore.type | JKS |
| ssl.truststore.location | - |
| ssl.truststore.password | - |
| ssl.truststore.type | JKS |
More information on configuring modules involved in Data Import process can be found at this link.
- Kafka (MSK):
- auto.create.topics.enable = true
- log.retention.minutes = 70-300
- Broker’s disk space: 300 GB
- 4 brokers, replication factor = 3, DI topics partition = 2
- Version 2.7 is 30% faster than version 1.6
JVM Settings
| module | running instances | CPU | memory | memoryReservation | maxMetaspaceSize | Xmx |
|---|---|---|---|---|---|---|
| mod-data-import | 1 | 256 | 2048 | 1844 | 512m | 1292m |
| mod-source-record-manager | 2 | 1024 | 2048 | 1844 | 800m | 1024m |
| mod-source-record-storage | 2 | 1024 | 1536 | 1440 | 512m | 1024m |
| mod-di-converter-storage | 2 | 128 | 1024 | 896 | 128m | 768m |
| mod-inventory | 2 | 1024 | 2880 | 2592 | 512m | 1814m |
| mod-inventory-storage | 2 | 1024 | 2208 | 1952 | 512m | 1440m |
Key improvement Delivered for Orchid:
- Added indexes for 035 and 010 MARC fields
- Tested with increased connection pool (DB_MAXPOOLSIZE) for mod-source-record-manager, mod-source-record-storage - default is set to 15, on high loads (>5000 records for udpate) should be increased to 30 for mod-source-record-manager and mod-source-record-storage.
Also provided new "DB_CONNECTION_TIMEOUT" env variable to 40 for mod-source-record-storage
{"name":"DB_CONNECTION_TIMEOUT",
"value":"40"}
Key Improvements Delivered for Morning Glory:
- Implemented Flow control mechanism that allows OCLC single record imports to be prioritized over imports of large files
- Alleviated eventloop blocking during batch save of records
- Reduced conversion of parsed content into a marc4j record
- Improved performance of sql query for retrieving log entries
- Fixed race conditions during mapping parameter initialization
- Removed JobExecutionCache
- Optimised functionality to create Instance records
Key Improvements Delivered for Kiwi/Lotus:
- Performance
- Improve speed and number of records for CREATEs and UPDATEs
- Reduce Kafka message size: 6 Data Import topics messages >200KB/ea
- Use less CPU for (de)serialization
- Significantly improve speed of loading UI Landing page
- Improve and optimize slow DB queries
- Remove events_cache topic and replace it by DB deduplication solutions (causes spikes in mod-inventory and the brokers, leads to instability and unpredictable outcomes)
- Improve resiliency and error handling to prevent import job from getting stuck
Key Improvements Delivered as of Iris Hotfix 3 and Juniper:
- No more accidental creation of duplicate records
- Ability to consistently make repeated updates on existing records
- CPU usage of multiple instances of the Data Import landing page now consumes 10% CPU, instead of maxing out at 100%
- When idle, mod-source-record-manager consumes 50% CPU instead of 90%
- Vertical scaling improves performance, especially with Iris HF2 Data Import modules