Summary
The viewer repeatedly fails to launch on eglfs / linuxfb boards and exhausts its attempt budget. This is the highest-volume live issue in Sentry and is not addressed by PR #3240, whose watchdog early-returns on non-Wayland boards.
Sentry: ANTHIAS-D (443 events, WebviewLaunchError: AnthiasViewer did not start after 30 attempts) and ANTHIAS-19 (28 events, the 3-attempt variant). Both still firing; ANTHIAS-D incremented during the 2026-08-07 triage sweep itself.
Affected hardware
device_type pi3, pi3-64, pi4-64, pi2 (and one pi5 on the 19 group); kernel_machine both aarch64 and armv7l; both balena=true and false. These are the eglfs and linuxfb display stacks — #3240's recovery is Wayland-only, so it cannot fire here.
Notes for whoever picks this up
- The differing attempt counts across sibling issues (3 vs 30) are the
max_attempts parameter changing over time, not two distinct defects. load_browser raises after range(1, max_attempts + 1) is exhausted.
- Two sibling signatures — ANTHIAS-F (325 events) and ANTHIAS-1C (42) — were resolved in the 2026-08-07 sweep because they genuinely stopped occurring a month ago. ANTHIAS-D and ANTHIAS-19 were deliberately left open because they have not.
- The failure modes recorded in the messages differ and are worth separating before fixing: "did not emit
AnthiasViewer started" versus "exited before emitting". The second is a crash, the first is a hang or a missed handshake.
- Related-looking but separately resolved: the GDBus
NoReply cluster (ANTHIAS-42/3/3H/48, ~877 events) stopped at the same 2026-07-08 boundary. If this turns out to share a root cause with those, that boundary is a useful clue about what changed.
Suggested first step
Reproduce on the armhf testbed (Pi 2) and a 64-bit eglfs board (Pi 3-64 / Pi 4), which are the two stacks carrying the volume, and capture the viewer log around the failed handshake to establish which of the two failure modes dominates.
Summary
The viewer repeatedly fails to launch on eglfs / linuxfb boards and exhausts its attempt budget. This is the highest-volume live issue in Sentry and is not addressed by PR #3240, whose watchdog early-returns on non-Wayland boards.
Sentry: ANTHIAS-D (443 events,
WebviewLaunchError: AnthiasViewer did not start after 30 attempts) and ANTHIAS-19 (28 events, the 3-attempt variant). Both still firing; ANTHIAS-D incremented during the 2026-08-07 triage sweep itself.Affected hardware
device_typepi3, pi3-64, pi4-64, pi2 (and one pi5 on the 19 group);kernel_machinebothaarch64andarmv7l; bothbalena=trueandfalse. These are the eglfs and linuxfb display stacks — #3240's recovery is Wayland-only, so it cannot fire here.Notes for whoever picks this up
max_attemptsparameter changing over time, not two distinct defects.load_browserraises afterrange(1, max_attempts + 1)is exhausted.AnthiasViewer started" versus "exited before emitting". The second is a crash, the first is a hang or a missed handshake.NoReplycluster (ANTHIAS-42/3/3H/48, ~877 events) stopped at the same 2026-07-08 boundary. If this turns out to share a root cause with those, that boundary is a useful clue about what changed.Suggested first step
Reproduce on the armhf testbed (Pi 2) and a 64-bit eglfs board (Pi 3-64 / Pi 4), which are the two stacks carrying the volume, and capture the viewer log around the failed handshake to establish which of the two failure modes dominates.