CNV Automation – Week of Feb 18-25, 2026

10K VM Scale Testing

Identified Ceph single-image contention as the root cause of clone deadlocks – all 10K clones target one RBD parent image, and Ceph serializes per-image operations, causing 600+ goroutines to pile up. Applied CSI provisioner throttling to limit concurrent clone calls from 600+ to 10. Test #6 is running stable at ~60% complete (~5,957 VMs running, ~460 DVs/hour) – first successful sustained 10K VM creation run.

kube-burner Cleanup Fix

Found why kube-burner was destroying all VMs when it hit a timeout – a YAML config flag was blocking on old namespace deletion, and a code bug treated normal transient states as fatal errors. Fixed both and contributed 5 PRs across 2 upstream repos:

• cnv-scenarios PR #23 - Output errors/warnings fix (merged) - CNV-79877
• cnv-scenarios PR #27 - Generic namespace cleanup (merged)
• kube-burner PR #1147 - verifyCondition retry fix (submitted upstream) - CNV-79878
• cnv-scenarios PR #28 - Storage defaults (on fork) - CNV-79907
• cnv-scenarios PR #29 - kube-burner 1.15.0 compat (on fork) - CNV-79908

HealthCrew Dashboard (CNV-79980)

• *CNV scenarios report system* – Generates visual HTML reports for kube-burner test runs so results are easy to read without digging through logs.
• *Test progression engine* – Automatically detects problems during a run (24 stop conditions) and suggests the next test configuration to try (~30 rules), replacing manual trial and error.
• *User templates* – Engineers can save, load, and share test configurations so they don't have to reconfigure 30+ parameters every time.

Storage Comparison Report

Built a side-by-side comparison of 5 storage backends (NVMe LSO, Ceph RBD, CephFS, IBM GPFS, NetApp GPFS) – shows which storage performs best for CNV workloads across sequential reads, writes, and random IOPS. Sourced from CNV-75751, Regression 4.21, and Fusion Access GA test results.

DFBUGS-394 Verification

Wrote the step-by-step procedure to verify that ODF storage pods now have higher priority so they won't be evicted during cluster stress – ready for execution on ODF 4.21.