2023.06.0 Release

Docker tag: quay.io/pganalyze/enterprise:v2023.06.0

Notes

• This release requires a database migration when coming from older releases, be sure to follow the recommended upgrade steps
  • Important: The included migration on the queries table (20230410170608_add_tables_to_queries) may take considerable time on the pganalyze database when monitoring many database servers. We recommend doing the upgrade during off-hours.
  • When upgrading, please note that the application will run the following SQL statement - when permissions are restricted, you may need to run this ahead of the upgrade:

    CREATE EXTENSION IF NOT EXISTS btree_gin WITH SCHEMA public;

Features

• This release contains preparations for upcoming autovacuum tuning recommendations, as well as the constraint programming based Index Advisor - more on this soon!
• Wait event monitoring:
  • Use higher data resolution when viewing a timeframe that's 1 hour or less
  • Reduce data resolution when viewing longer timeframes to improve clickability
  • Allow zooming the graph through click+drag, like other graphs in the application
• Add additional AWS settings for bundled collector
  • Adds support for configuring the Cluster ID, the Assume Role, as well as allows enabling IAM authentication when using the bundled collector
• The bundled collector has been updated to v0.50.0 (from v0.49.2 in prior release)
  • This is primarily a bugfix release in the context of an Enterprise Server installation using the bundled collector to monitor a self-managed or Amazon RDS/Aurora database server
  • In case you have installed the collector separately, we recommend upgrading the collector package/container and reviewing the full changelog
• Issue e-mail notifications: Include pganalyze organization name in email body, for easier filtering when using multiple organizations
• Add additional telemetry data to check pganalyze snapshot health via Prometheus/Grafana
  • See instructions for how to use the Prometheus endpoint
• Allow setting SIDEKIQ_CONCURRENCY=0 to turn off workers in combined Enterprise image
  • You can utilize this setting in order to separate the workers from the container servicing the pganalyze web UI, avoiding slow web requests caused by worker processing
  • Note such a setup requires having configured object storage as well as an external Redis server (set via REDIS_URL)
  • See all settings

Performance

• Query analysis: Speed up analysis by avoiding to process schema dependency tracking
• Improve performance of the getIssues API endpoint by avoiding N+1 queries
• Disable use of prepared statements for web application
  • In practice this ends up causing more overhead than benefit for the application, and we've seen better results without prepared statements
• Cache table names associated to each query directly on queries table
  • This speeds up background tasks working on a per-table basis (e.g. Index Advisor), as well as per-table queries list in the web application
• Background workers:
  • Don't run check-up logic for servers that have not received data in the last 24 hours
  • Speed up truncation of queries (truncated queries are used for display in tables)
  • Batch check-up logic together for 10 database servers at a time

Bugfixes

• Object storage: Fix DNS lookup errors from Rust-based snapshot workers
  • This may have shown as an I/O error that caused no snapshot data to be processed, despite the collector successfully sending data to the installation
  • If we have previously asked you to utilize an IP address for your MinIO endpoint as a workaround, you can now switch back to using a DNS hostname
• Query analysis:
  • Don't retry query analysis after "relation does not exist" errors
  • Avoid "Unexpected rtekind" errors for functions and VALUES
  • Fix planning for hash indexes (this allows Index Advisor to run on tables that have hash indexes, but does not yet recommend hash indexes)
• Remove query length limit for parsing
  • Previously query texts exceeding 20,000 characters were not parsed during the initial intake, causing them to not have referenced tables, or an associated statement type
• getIssues API endpoint: Fix legacy displayName field to actually return the name, not the UUID
  • Note that the displayName field is deprecated and replaced by references.name (see the current getIssues API docs)
• Fix parsing of EXPLAIN plans using the "Presorted Key" property when using auto_explain.log_format = text"
• Query sample deduplication: Ensure query sample with parameters is stored if it occurs at the same time as an auto_explain sample without parameters
• Add critical priority queue to ensure statistics data partitions are created even when queue is heavily backlogged

Security

• Update to Ruby 3.0
• Update nokogiri to resolve GHSA-pxvg-2qj5-37jq
• Upgrade svg-react-loader and xml2js to address GHSA-776f-qx25-q3cc