Skip to content

Fix per-node bootflash sizing in mixed-version fabrics - #494

Merged
monrog2 merged 3 commits into
v4.3.1-devfrom
codex/issue-493-per-node-bootflash
Nov 10, 2026
Merged

monrog2 merged 3 commits into
v4.3.1-devfrom
codex/issue-493-per-node-bootflash

Conversation

@monrog2

@monrog2 monrog2 commented Nov 6, 2026

Copy link
Copy Markdown
Collaborator

Switches already running split-image releases were sized using the fabric's lowest switch version, causing false bootflash failures in mixed-version fabrics. Use each node's fabricNode.version and firmwareRunning.ksFile, joined by (pod, node), to select the current architecture. Nodes at/after 6.0(2a) require twice their matching target image size (or one copy when downloaded), without requiring their installed image in the Firmware Repository.

The six-node #493 reproducer now requires 5389.46 MiB and passes with ~6489–6502 MiB available. Missing node version/architecture is listed for manual evaluation; confirmed space failures remain failures. Existing pre-boundary/crossing formulas and target-image availability checks are retained. #490's separate capacity/predownload redesign is intentionally excluded.

Fixes #493.

Validation (local Python 3.12, no live integration):

  • Bootflash suite: 32 passed, including the exact reproducer, both architectures, absent installed image, pod-aware joins, downloaded requirements, and missing inventory.
  • Established AciVersion, CheckManager, Result, and bootflash suites together: 160 passed, 8 pre-existing deprecation warnings. Focused suite rerun after final null-inventory handling: 32 passed.
  • git diff --check: passed.

Environment notes: default Python 3.15 is incompatible with the existing threading test harness; Python 3.12 was used. An initial full-suite attempt stopped at collection because jinja2 was unavailable; the proportionate suites above completed successfully.

@monrog2

monrog2 commented Nov 10, 2026

Copy link
Copy Markdown
Collaborator Author

Integration test summary — switch_bootflash_usage_check

Focused integration testing ran against PR #494 commit a8590df1d5ea1189912c266361d9ceb0e036a673 in GitLab pipeline 296.

Seven scenarios were tested: the fabrics’ configured versions plus six current/target version overrides spanning pre- and post-6.0(2) releases.

Results:

  • The actual-version run evaluated 36 connected fabrics: 14 passed, 5 returned valid insufficient-space failures, and 17 required manual review because target versions or firmware metadata were unavailable.
  • No connected-fabric execution produced a script error.
  • Live results showed different required-space values for nodes within the same fabric—for example, 5166.94 MB versus 5452.61 MB. The previous implementation calculated one fabric-wide requirement, so this confirms the new per-node version and running-image architecture logic was exercised.
  • The exact issue switch_bootflash_usage_check should be more granular than lowest switch version #493 regression test ran successfully under both Python 2.7 and Python 3.8.

The overall pipeline failed because some required fabrics could not be reached, not because of a failure in the bootflash check. The override scenarios were also limited by missing target-image metadata, so the run confirms that the fix was deployed and exercised but does not constitute an exact live reproduction of issue #493’s original 5.3(2e) → 6.1(6g) topology.

@monrog2
monrog2 merged commit cd4cdcb into v4.3.1-dev Nov 10, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

switch_bootflash_usage_check should be more granular than lowest switch version

1 participant