backstage release v3.31.0

Added

• (agent-platform) Search from three items, sessions organization filter, hide single-option filters by @Rotfuks — The Sessions overview gets an Organization filter. Search and filters appear on any Agent Platform list from three items, dropdowns offer type-ahead from three options, and a filter with only one possible choice is hidden.
• (agent-platform) Place environment and organization together on the new agent form by @Rotfuks — On the New agent form, Environment and Owner (formerly Organization) sit together as plain fields, Owner starting on Platform. With nothing to choose, both become information lines under More options. Agent pages and the Agents list say Owner too.
• (muster) Enter a connector’s client ID and secret in the wizard by @Rotfuks — When adding a connector whose people sign in with their own account, the wizard takes the app’s client ID and secret: New Secret has muster store them in a Kubernetes Secret owned by the connector, Existing Secret names one that holds them. No more kubectl beforehand.
• (muster, agent-platform) Show connector logos by @Rotfuks — Connectors show their logo on the Connectors list, a connector’s page and the agent overview, instead of a letter tile. Needs a muster that reports the connector’s icon; until that is rolled out, the letter tile stays.
• (ui-react) Move FileView to ui-react, and show the helm command and build log in CodeBlock by @marians — On the review step of the New agent wizard, the helm command under Install manually instead is a copyable command, not a file. A failed toolchain build’s log shows under Log of the failed job, opens at its end and scrolls inside its box.
• (several plugins) Tags on agents, skills, connectors and workflows by @Rotfuks — Agents, skills, connectors and workflows take tags when created or edited, and Customize filters and searches by them (Tags). The Connectors tab is one list now: the infrastructure and agent-platform tags replace its Your connectors, Infrastructure and Internal sub-tabs. Merged into #3315, which brought it to main; editing tags needs an agent-manager and muster that support them.

Fixed

• (agent-platform) Give refused raw TCP connections one remedy that no setting can allow by @marians — When a session shows a refused raw TCP connection (such as github.com:22), the notice no longer suggests asking the platform team or editing the agent. It says No egress setting can allow it. and suggests HTTPS where the host offers it.
• (agent-platform) Tell objects already being removed from new ones after reopening the Delete cluster dialog by @marians — Reopened while a cluster is being removed, the Delete cluster dialog lists what is already terminating as Still being removed, and offers Finish removal only when there is something new to remove.
• (backend) Load the Postgres search engine only where search runs on Postgres by @marians — For operators: a backend on SQLite (the chart’s default) no longer loads the Postgres search module, so its warning Postgres search engine is not supported, skipping registration of search-backend-module-pg stops reaching Sentry at every start.
• (backend-common) Ignore the empty TechDocs index warning from the Postgres search engine too by @marians — For operators: Index for techdocs was not replaced: indexer received 0 documents, logged on almost every TechDocs indexing run with the Postgres search engine, no longer reaches Sentry, like the was not created wording before it.
• (gs-node) Report a Mimir that cannot answer once, with a fixed title, and say when older data is temporarily unavailable by @marians — When the metrics store cannot read data older than a few hours, the Usage pages, a deployment’s resource usage and a cluster’s Gateways tab say Older data is temporarily unavailable instead of a generic error. For operators: each failure reaches Sentry once, under a fixed title.
• (muster) Report Muster connection failures as one Sentry issue per failure, not per installation by @marians — For operators: when the portal cannot reach Muster, Sentry groups the failures by kind (Muster MCP request failed: <reason>) with the installation in the event’s data, instead of one issue per installation, and reports each failed request once.
• (muster) Hold the connector’s Calls line in place with a skeleton while its usage is read by @marians — On a connector’s Details tab, the Calls, last 30 days line holds its place with a placeholder while the figure loads, so the lines below no longer jump. Without a figure it says why, such as Could not be read.

Full Changelog: https://github.com/giantswarm/backstage/compare/v3.30.0...v3.31.0