Investigate LDP1 Hosting Support

• Organize conversation with hosting vendors to identify current offerings, concerns
• Organize conversation with implementers about reporting platforms and hosting plans (blocked by Recruit from new institutions not represented)

Summary of conversation in Sys Ops & Management SIG

• LDP1 Deployment and Operations (Slides from Wayne Schneider at Index Data) 
• Official notes from Sys Ops meeting
• Angela's notes from meeting
• FOLIO Sys Ops & Mgt SIG (21-Oct-2022 09:58 Eastern U.S. time) | FOLIO Meeting Recordings (openlibraryfoundation.org)

Best practices mentioned

• general tips
  • Follow the LDP1 Administrator Guide
  • The production database should be a standalone Postgres instance with dedicated storage. Only use it for FOLIO reporting.
  • Use ldp_add_column.conf to make sure tables in LDP1 contain certain columns whether or not there are data for those columns. This helps make sure that automated derived table queries run correctly, even if the data are absent.
• periodic extraction of data 
  • LDP1, ldpmarc and folio-analytics data extraction jobs need to be run on a schedule and in sequence. Possible scheduling services include cron, Jenkins, Kubernets cronjob.
  • For order of extraction jobs, might try LDP1, then ldpmarc, then folio-analytics derived tables. (moved here from upgrades section)
  • Once a day tends to be pretty good.
  • While it updates the data, you wouldn't want to run large queries against the reporting database because the data update uses a lot of system resources.
  • Daily incremental update for ldpmarc is quite fast (e.g., 10 minutes for University of Chicago).
• performance
  • Run with network proximity between the FOLIO and LDP databases 
• backups / disaster recovery:
  • Q: What do you do for a postgres backup solution (for a production LDP), and what cadence to you do that on ?
    • A: some are running on Amazon AWS, RDS Postgres services for hosting, including disaster recovery (7 days of snapshots)
    • A: some have local snapshots. 7 days is likely enough, because LDP keeps history of the records internally.
• cloning
  • Q: When you clone FOLIO production environment, do you clone the LDP data over or do you build the LDP data back from the FOLIO data that you have cloned ?
         - A (Texas A&M): We don't clone the data over, but we re-build it from scratch. Or, if I upgrade, I just upgrade the LDP . In that way it preserves the history.
         - A (Wayne): If we refresh the staging environment for a tenant, we will re-build LDP from scratch.
• upgrades / testing
  • Q:  error checking and logging. Do testing and upgrades go together ? 
  • Run two instances: staging and production.
  • LDP1 and ldpmarc releases are both fairly independent of FOLIO releases, new versions come out on their own cadence
  • When there is a new version of LDP1 or ldpmarc, implement that in staging.
  • If it's more than a point upgrade, might invite users to test.
  • Follow that with upgrade of production.
  • An upgrade in LDP1 is really simple. You invoke the LDP server with an upgrade-database command. Then LDP knows what verison it is talking to.      
  • For ldpmarc, no upgrade process is necessary. If needed, the new version will automatically perform a full update (rather than incremental update).
  • FOLIO Analytics is tested with specific FOLIO flower releases, so upgrade those together; include LDP1 in standard FOLIO testing. Apart from that, invite user to test only when folio-analytics tables have changed.
  • Some have a special Jenkins job which creates a thing called the LDP engine. This engine includes all 3 components. They use a Docker image for all their setups for the Flower Release. When a new version of LDP comes, they build a new Docker image with the latest versions. Then they do a smoke test. If everything works fine, they create recommendations for the upcoming Flower Release. They use these until the next upcoming flower release.
• logging
  • Some use container log aggregation (CloudWatch, or Elastic), so the logs don't go away when the container goes away.
  • Some use FluentD/Rancher for logs, which then get pushed to Splunk; might need to log to standard error/out instead of to a file.
  • Probably a good idea to sent an alert for when the jobs fail, but could also check all manually each morning.
• security
  • Set up your network securely. Set up access to the reporting database securely. Secure (or pseudonomize) personally identifable information.
  • Apart from that, there are no security concerns !
  • Single-Sign On to database access is not supported. You have to use regular postgresql security. But that is more an integration problem than a security problem.
  • Make sure the LDP IP address is only accessible from certain subnets.
  • Make sure Kubernetes network is namespace isolated.
  • Set up users with read-only access, and use those accounts for connections via tools like Cloud Beaver. Then embed those tools in a VM that uses standard university permissions systems (e.g., SSO via Shibboleth). The different read-only user accounts can be granted permissions to just certain types of data, and different staff can be granted permission to use just the VM that matches the permissions they should have.
  • SSL is used for LDP database access unless otherwise configured.
• concerns
  • Data transfer time gets longer as data in FOLIO increases.
  • You will need more resources in LDP if your production database grows.
  • Silent failures. You do not have a good and detailed logging. You never know what has happened.  (But there is good support by Nassib!)
  • Documentation in LDP1 repository is mostly oriented to developers. It is not easy to find information if you are not familiar with the system.
  • For ldpmarc, incremental update is quick, but full update can be pretty long.  (This is less of an issue since ldpmarc 1.6 which includes significant performance improvements)
  • There is no automatic recovery. When process fails, it needs to be re-run manually.