fix: sync sticky disk implementation with stickydisk action

[adityamaru]

Problem

Users are experiencing connection errors when using setup-bazel:

Warning: Failed to mount sticky disk for /home/runner/.cache/bazel-repo: ConnectError: [unavailable] connect ECONNREFUSED 192.168.127.1:5557

Root Cause

The setup-bazel action has drifted from the stickydisk action, and is using the wrong environment variable name (VM_ID instead of BLACKSMITH_VM_ID), which prevents proper VM identification.

Changes

Critical Fix

• Fix environment variable name: Changed VM_ID to BLACKSMITH_VM_ID in 4 locations
  • This is the primary cause of the connection refused errors

Reliability Improvements

• Increase timeout: 15s → 45s for sticky disk operations
• Add sync: Flush pending writes before collecting filesystem usage
• Add drop_caches: Drop kernel caches before unmount to prevent 'device is busy' errors
• Add lost+found cleanup: Remove lost+found directory to prevent EACCES errors with build tools (pnpm, yarn, npm, docker buildx)
• Add debug logging: Log the port being used in createStickyDiskClient
• Improve mount point creation: Use sudo for system directories like /nix, /mnt

Testing

These changes align setup-bazel with the tested stickydisk action implementation, which has been working successfully in production.

Related

Syncs with: https://github.com/useblacksmith/stickydisk

---

Note

Switch to BLACKSMITH_VM_ID and add reliability improvements (longer timeouts, safe formatting, mount ownership, sync/drop_caches, and logging) for sticky disk operations.

• Sticky Disk API:
  • Replace VM_ID with BLACKSMITH_VM_ID in all service calls (getStickyDisk, commitStickyDisk).
• Mount/Format:
  • Extend abort timeout from 15s to 45s when fetching disk.
  • On first format, remove lost+found after mkfs.ext4 via temporary mount to avoid permission issues.
  • Create mount point with sudo and ensure ownership via chown before and after mount.
• Unmount/Commit Reliability:
  • Flush writes with sync before measuring usage.
  • Drop caches (echo 3 > /proc/sys/vm/drop_caches) prior to unmount to reduce "device is busy" errors.
  • Report filesystem usage if available; otherwise proceed without it.
• Observability:
  • Log gRPC client port in createStickyDiskClient.

^{Written by Cursor Bugbot for commit 2e5fa36. This will update automatically on new commits. Configure here.}