{% extends "base.html" %} {% block title %}Source health — {{ site_name }}{% endblock %} {% block content %}

Source health

Every {{ status.interval_hours }} hours the canary asks each active source about a known indicator and records the answer. A source that changes its answer, stops answering or slows down shows up here before it changes a verdict.

{{ status.counts.ok }}answering
{{ status.counts.warn }}warning
{{ status.counts.bad }}failing
{{ status.counts.skip }}not asked
{{ status.counts.unchecked }}unchecked
{{ status.counts.off }}not configured
{% if status.status == 'ok' %}
Every active source answered as expected{% if status.checked_at %}, last checked {{ rel_time(status.checked_at) }}{% endif %}.
{% if status.stale_count %}
{{ status.stale_count }} active source{{ '' if status.stale_count == 1 else 's' }} {{ 'has' if status.stale_count == 1 else 'have' }} no canary result from the last {{ status.stale_after_hours }} hours, so {{ 'it is' if status.stale_count == 1 else 'they are' }} unproven rather than healthy. A pager watching ?strict=1 fails on that until the next canary run.
{% endif %} {% elif status.status == 'unknown' %}
No check has been recorded yet, so nothing here is a claim about any source. Run scripts/source_canary.py once, or wait for the timer.
{% elif status.status == 'bad' %}
{{ status.counts.bad }} active source{{ '' if status.counts.bad == 1 else 's' }} failed the last check. Lookups still return every other source's answer; the failures are listed first below.
{% else %}
{{ status.counts.warn }} source{{ '' if status.counts.warn == 1 else 's' }} answered differently than expected or slowed down. Nothing is down, but a changed answer is what a broken parser looks like from outside. {% if status.stale_count %} {{ status.stale_count }} of them are stale: no canary result in over {{ status.stale_after_hours }} hours, so the timer may not be running. A pager watching ?strict=1 treats that as a failure. {% endif %}
{% endif %}
{% for s in status.sources %} {% endfor %}
SourceSupportsHealthLast answerLatencyChecked
{% if s.homepage %}{{ s.display }} {% else %}{{ s.display }}{% endif %}
{{ s.name }}
{{ s.kinds|join(', ') }} {{ s.state }} {% if s.status %}{{ s.status }}{% if s.verdict and s.verdict != 'unknown' %} · {{ s.verdict }}{% endif %} {% else %}–{% endif %} {% if s.note %}
{{ s.note }}
{% endif %}
{% if s.latency_ms %}{{ s.latency_ms }} ms{% if s.drift and s.drift > 1.2 %} {{ '%.1f'|format(s.drift) }}× baseline{% endif %} {% else %}–{% endif %} {% if s.checked_at %}{{ rel_time(s.checked_at) }}{% else %}never{% endif %}

What the canary asks

Four indicators whose answer should not change, each asked of every source that supports its type. A run costs one request per active source, so it bypasses the cache on purpose: a cached answer would hide exactly the breakage it is looking for.

sha256  275a021bbfb6489e54d471899f7db9d1663fc695ec2fe2a2c4538aabf651fd0f  EICAR test file
url     http://malware.testing.google.test/testing/malware/             Safe Browsing sample
domain  example.com                                                       stable, benign
ip      8.8.8.8                                                           stable, benign

The first healthy answer becomes the expected one. From then on a source that answers "not found" where it used to answer "found" is a warning rather than a success — that is the shape of both silent breakages fixed in 2.6.0 and 2.7.0, neither of which returned an error.

Machine-readable: /api/v1/status returns the same data. ?strict=1 answers 503 when a source is failing or when a canary result is stale (nothing recorded for {{ status.stale_after_hours }} hours, so the timer may be dead); ?strict=warn is the loud setting and also fails on a source that is merely slow or that changed its answer. A stale row is flagged in this payload so a monitor can tell "slow" from "unproven".

{% endblock %}