Client v2

[saadqbal]

Note

Medium Risk
Medium risk due to substantial Helm chart template/RBAC/secret-name changes across multiple deployment targets plus new GitHub Actions release automation, which could break existing installs or publishing if misconfigured.

Overview
Adds a new unified Helm chart under tracebloc/ (v2.0.0) with platform-specific CI values, migration docs, and production install documentation, and wires in GitHub Actions to lint/template/schema-test the chart on PRs and package/publish it to gh-pages and GitHub Releases on tags.

Updates the legacy per-platform charts (aks/, bm/, eks/, oc/) with standard labels, dedicated ServiceAccounts, stricter secret handling (required client password), optional node selectors, PVC helm.sh/resource-policy: keep, MySQL split into dedicated resources with probes, tighter StorageClass permissions, and more scoped RBAC rules; adds DEPRECATED.md notices and OpenShift guidance.

Introduces one-command local Kubernetes setup scripts (scripts/install.sh + modular libs, and Windows install.ps1/install-k8s.ps1) that install Docker/k3d/kubectl/Helm, detect GPUs, and optionally install/configure NVIDIA/AMD components and device plugins.

^{Written by Cursor Bugbot for commit c72a8a0. This will update automatically on new commits. Configure here.}

[cursor]

GPU resources default changed, breaks non-GPU clusters

High Severity

When env.GPU_REQUESTS and env.GPU_LIMITS are not provided (the default, since env is {}), the template falls back to "nvidia.com/gpu=1" for both. All four legacy charts defaulted to empty strings for these values, meaning no GPU was requested. The unified chart now causes every spawned job to request an NVIDIA GPU by default. On clusters without GPU nodes — which is the common case — spawned jobs will be stuck in Pending forever because the scheduler can't satisfy the nvidia.com/gpu resource request.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Autofix Details

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: Installer fails on first run resolving nonexistent directory
  • validate_config now resolves HOST_DATA_DIR via its canonicalized parent when the target directory does not yet exist, so first-run defaults like ~/.tracebloc no longer fail.

Or push these changes by commenting:

@cursor push 919298949e

Preview (919298949e)

diff --git a/scripts/lib/common.sh b/scripts/lib/common.sh--- a/scripts/lib/common.sh+++ b/scripts/lib/common.sh@@ -200,7 +200,15 @@
# HOST_DATA_DIR must be under $HOME and must not be a system path (security)
local dir="$HOST_DATA_DIR"
[[ "$dir" != /* ]] && dir="$HOME/$dir"
- dir="$(cd -P "$dir" 2>/dev/null && pwd)" || true+ if [[ -d "$dir" ]]; then+ dir="$(cd -P "$dir" 2>/dev/null && pwd)" || true+ else+ local parent base+ parent="$(dirname "$dir")"+ base="$(basename "$dir")"+ parent="$(cd -P "$parent" 2>/dev/null && pwd)" || true+ [[ -n "$parent" ]] && dir="${parent}/${base}" || dir=""+ fi
[[ -z "$dir" ]] && error "HOST_DATA_DIR could not be resolved: $HOST_DATA_DIR"
case "$dir" in
/) error "HOST_DATA_DIR cannot be root (/)"

_{This Bugbot Autofix run was free. To enable autofix for future PRs, go to the Cursor dashboard.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}