Y-axis labels on the RTT chart were truncated above 99ms because the
left margin was a fixed guess; it's now sized to the widest tick
label. The x-axis only showed start/end timestamps; added evenly
spaced intermediate tickmarks and labels. Plot area enlarged 15%.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Only wipe real (non-RTT) plugin data on an actual client reboot (boot
flag), not on every ordinary OVERDUE/DOWN -> UP recovery. A transient
network blip no longer erases CPU/memory/etc. history.
Also split the shared time-series chart into separate line/area segments
wherever the gap between samples is much larger than the typical spacing,
so missing data (host overdue, or history that simply hasn't accumulated
across a drop) renders as a visual gap instead of an interpolated line.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DEimzMv4Q5EjFg3hoiZ69T
Since RTT history keys make host.plugin_data non-empty after the very
first heartbeat, the "no hosts" empty state can now only trigger when
zero hosts have ever connected, not when hosts exist but haven't sent
plugin metrics yet (that in-between state no longer exists). Update
the copy to describe the actual current trigger condition.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DEimzMv4Q5EjFg3hoiZ69T
host.plugin_data now always holds rtt_ipv4/rtt_ipv6 history after the
first heartbeat, which broke two pre-existing behaviors that assumed
plugin_data reflects only real client-collected data:
- The request_update ACK gate in handle_datagram() never fired again
after a host's first heartbeat, so stale OS/agent-version info was
never refreshed after a reconnect (only a client restart fixed it).
- plugin_data.clear() on every UP transition wiped rtt_* history too,
so a flaky host could never accumulate a useful RTT graph.
Add _is_rtt_key()/_has_real_plugin_data() helpers so both spots treat
rtt_* keys as synthetic: the gate now looks only at real plugin data,
and the recovery clear only drops real plugin keys, preserving RTT
history across reconnects.
Also fixes two pre-existing E127 continuation-indent flake8 issues in
tests/test_udp_rtt_history.py.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DEimzMv4Q5EjFg3hoiZ69T
Add fetchHostInfo() call to the 30-second auto-refresh setInterval loop
for expanded host cards. This ensures the Connectivity table and RTT
charts update in real-time without requiring the user to collapse and
re-expand the host card.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DEimzMv4Q5EjFg3hoiZ69T
Pulls the CPU chart's SVG-drawing logic into a parameterized
renderTimeSeriesChart(elId, pts, opts) helper (yDomain, colorFor,
clipId, unitSuffix), keeping renderCpuChart's signature and behavior
unchanged. Task 5 will reuse this for the RTT chart.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DEimzMv4Q5EjFg3hoiZ69T
Capture RTT on every heartbeat using the existing plugin_data history
mechanism, recording under rtt_ipv4 or rtt_ipv6 keys. Maintains 100-sample
retention per host and address family.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DEimzMv4Q5EjFg3hoiZ69T
flap._state persists at module level keyed by (host, service) and only
clears via a RECOVER-triggered quiet window. A host dropped mid-flap with
no RECOVER ever received leaves ok_since permanently None, so the
flapping flag could never clear on its own.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
docs/PLUGIN_DEVELOPMENT.md referenced a nonexistent hbd/plugin.py and
hbd/plugins/ directory; the real module is hbd/client/plugin.py with
plugins under hbd/client/plugins/, as README.md and the actual code
already use. Anyone following the doc would hit ModuleNotFoundError.
foot.html was never included by any template (confirmed via git grep
and template audit) despite being actively maintained; wired it into
about.html. menu.html was fully commented out and unreferenced, so
removed it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CMD packets arrive as unauthenticated UDP datagrams, yet every hbc client
executed the shell command they carry without any opt-in. Add an
allow_remote_command config key, default false: when off, the command is
logged and refused with "Refused: allow_remote_command is false" (visible in
the server event log under the command service), and subprocess is never
reached. When on, the client warns at startup that it will execute CMD
packets.
Applied to all four clients that handle CMD — hbc, hbc_windows.py,
hbc_mini.py, and the C hbc_mini — since gating only one leaves the others
wide open. The C client reads the same key from ~/.hbc.json and needs a
rebuild to pick it up. UPD (self-update) is unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A (host, service) pair is flapping once it exceeds flap_count warning or
critical notifications within flap_interval minutes. The notification that
trips the state carries "Now flapping!! No more messages!" and every later
one is dropped, including RECOVER. The state ends silently flap_interval
minutes after a RECOVER, provided no further alert arrived meanwhile.
Hooked into notify.send_notification, the single choke point for channel
delivery, so only outbound notifications are suppressed — eventlog keeps
recording, leaving the journal and /log with the full history of the flap.
Threshold alerts key on their metric path, so a flapping disk check cannot
silence a CPU alert; connectivity, boot and shutdown events key on the host
itself. State lives at module level in flap.py and is never pickled.
Flapping pairs surface in Host.stateinfo() and render as an amber badge on
the live dashboard. Config: flap_count (5), flap_interval (10 minutes),
0 in either disables; both editable on the settings page.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replaces the 11-column table with severity-sorted record rows: host,
combined status chip, latency, alert chips, time-in-state, and the most
recent alert message. Per-family address/state/latency/last-change move
to a hover card on the status chip and permanently to a new Connectivity
table in the Host Overview info section (host info API now includes
connections). Last-alert is seeded from /api/0/alerts and kept fresh
from the event stream; the event log below uses the same row idiom with
its filters intact. WS reconnect state shows in the toolbar instead of
a modal.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NfPpSpccTWBfZg1FTveyaU
Extract the settings redesign's tokens and components into shared
static/hbd-ui.css + hbd-ui.js and dedupe settings.html against them.
About becomes yaml-key sections with kv rows; Alerts gets stat tiles,
chip filters, and alert record rows (same fetch/ack logic); Host
Overview keeps its DOM and live-update JS but is re-skinned to the
token system with a page toolbar. Live Dashboard intentionally
untouched pending its own redesign.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NfPpSpccTWBfZg1FTveyaU
Sticky offsets (toolbar, rail, mobile chips, anchor margins) derive from
the measured nav/toolbar heights via CSS variables, since the fixed nav's
height varies with viewport width.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NfPpSpccTWBfZg1FTveyaU
One idiom for every section: mono identifier, badges, fact chips, inline
accordion editor. Scalar config sections merge into one grouped 'server'
section. Sidebar becomes YAML-key nav (mobile: horizontal chips); sticky
toolbar carries pending-changes state and Publish. Staging, publish,
channel CRUD, and permission gating are unchanged — same payload shapes
and endpoints.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NfPpSpccTWBfZg1FTveyaU
The users-section loop reused 'username' as its loop variable, overwriting
the requesting user's name so host/threshold filtering compared against the
last user in the config. Rename to 'uname' and add a regression test with a
user that is not last in the users dict.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NfPpSpccTWBfZg1FTveyaU
Setting args.boot = False after sending the boot message was dead code: the
SIGHUP restart re-execs via os.execv(sys.argv[0], sys.argv) with the original
argv, so the re-launched process re-parsed -b and announced a boot again. The
exiting process also sent a spurious shutdown (send_shutdown armed by -b), so
each config reload looked like a host reboot to the server.
Strip -b/--boot from the argv passed to execv, and skip the shutdown message
in cleanup() when dorestart is set. A real SIGTERM/SIGINT shutdown still sends
shutdown as before.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
andreas
Andreas WredeandClaude Sonnet 5