Droid 0.57.15 hangs on VPS when updater is enabled (IPv6 stalls; workaround in use)

[ain3sh]

Droid updater startup hang on Linux VPS with partial IPv6 egress

Summary

On a Hetzner Ubuntu 24.04 VPS, droid hangs at startup when auto-update is enabled. The hang is reproducible for both droid and droid --version.

The same Droid version works on WSL2. Current stable workaround on VPS is:

export FACTORY_DROID_AUTO_UPDATE_ENABLED=false

With this workaround, interactive Droid starts normally.

Impact

• Interactive CLI is blocked on affected host when updater is enabled.
• Non-interactive version checks also hang (--version).
• User can work only with updater disabled.

Environment

• Droid: 0.57.15
• Failing host: Hetzner VPS, Ubuntu 24.04, kernel 6.8.0-100-generic
• Working host: WSL2 Ubuntu 24.04, kernel 6.6.87.2-microsoft-standard-WSL2

Reproduction

# Hangs on VPS
env -u FACTORY_DROID_AUTO_UPDATE_ENABLED ~/.local/bin/droid --version

# Works immediately on VPS
env FACTORY_DROID_AUTO_UPDATE_ENABLED=false ~/.local/bin/droid --version

Expected

Updater check should fail fast and CLI should still start/exit promptly.

Actual

Updater-enabled startup enters a long-running wait loop and never completes in observed runs (>40s, and continued until externally terminated).

Key evidence

1. VPS behavior split by updater flag

   • See vps_droid_behavior.txt (noenv_ec:124, dis_ec:0, dis_out:0.57.15)

2. Endpoint connectivity differs by IP family on VPS

   • vps_downloads_connectivity.txt: IPv6 connects to downloads.factory.ai:443 time out; IPv4 succeeds quickly
   • vps_curl_ipv4_ipv6.txt: curl -4 succeeds, curl -6 times out

3. Hanging process socket state on VPS

   • vps_hung_socket_snapshot.txt: multiple IPv6 sockets in SYN-SENT to CloudFront v6 addresses while IPv4 sockets are established

4. WSL contrast

   • wsl_downloads_connectivity.txt: IPv6 fails immediately with ENETUNREACH and IPv4 succeeds; Droid does not hang there

5. Source-address clue on VPS

   • vps_ipv6_source_binding.txt: IPv6 requests fail from 2a01:4ff:f2:4002::2 but succeed from 2a01:4ff:f0:733e::1

6. Strace shape

   • vps_strace_excerpt_tail.txt: repeated epoll_pwait2 + periodic timerfd reads during hang (no forward progress to version output)

Working workaround (currently in use)

FACTORY_DROID_AUTO_UPDATE_ENABLED=false exported in shell profile. Confirmed working in Termius session (droid --version and full interactive droid).

Likely fix direction

• Ensure updater check never blocks startup/exit path.
• Use strict per-address connection timeouts and robust Happy Eyeballs/IPv4 fallback when IPv6 handshakes stall.
• Treat stalled IPv6 paths as non-fatal and continue.

[ain3sh]

Attached diagnostic bundle for this report: droid_hang_bundle_20260217_131635.tar.gz

Contents include:

• BUG_REPORT_GITHUB.md (concise full write-up)
• ARTIFACT_GUIDE.md (quick triage map)
• VPS/WSL connectivity comparisons
• Hung socket snapshot (SYN-SENT IPv6)
• Strace excerpts showing updater-enabled wait loop
• Repro outputs confirming current working workaround (FACTORY_DROID_AUTO_UPDATE_ENABLED=false)