cds-ai/docs/cirrus-data-cloud/run-learnings.md

Cirrus Data Cloud Run Learnings

This file stores run-specific examples only when a run produced a new learning relevant to future tasks.

Entry Rule

• Add an entry only when the run changed workflow behavior, uncovered a new failure pattern, or confirmed a new required check.
• Do not add routine successful runs with no new learning.

Run Learning: Operation 14208

• Learning: wait-for-vm-registration helper registration can be the longest early-stage step.
• Action for future runs: if step 6/7 is slow, verify helper VM existence in vCenter before remediation.

Run Learning: Operation 14213

• Learning: completion response was sent before destination delete prompt, operation archive, and offline-host cleanup.
• Action for future runs: completion must be gated on delete prompt handling, archive, and cleanup verification.

Run Learning: Operation 14214

• Learning: stale helper/source entries can remain and require explicit offline-host cleanup reruns.
• Action for future runs: rerun cleanup until stale entries are actually removed.

Run Learning: Operation 14215

• Learning: helper creation can fail with vSphere ReconfigVM errors and recover via controlled retries.
• Action for future runs:
• remove leftover helper artifacts before retry
• avoid manual helper power actions during active task execution
• keep waiting while heartbeats/progress still advance

Run Learning: Operation 14216

• Learning: destination login validation and post-run cleanup were missed before completion reporting.
• Action for future runs: always perform destination login validation + archive + cleanup automatically before declaring completion.

Run Learning: Operation 14218

• Learning: source/helper entries can remain connected with stale last_checkin after migration.
• Action for future runs: enforce heartbeat-timeout waits and rerun cleanup until source/helper entries are removed.

Run Learning: Operation 14221

• Learning: source/helper CDC entries for the current request can be removed cleanly by timeout-based cleanup loop after archive, and final 4-entity status listing is effective for closure.
• Action for future runs:
• always provide final source/destination/access/helper listing across CDC and vCenter
• keep destination delete as explicit user-confirmed step only

Run Learning: Operation 14223

• Learning: on RHEL 10, CMC reinstall via installer script can fail when repo metadata is unavailable; local RPM install + explicit CDC endpoint config + manual register can recover the source in-place.
• Action for future runs:
• if Linux installer fails on repo metadata, check cached mtdi-daemon and galaxy-migrate RPMs and install directly
• enforce galaxy_complete_endpoint before manual register
• proceed with migrateops only after source host is confirmed connected in CDC