Shield Security Events Feed
Shield → Security Events is a project-scoped, filterable feed of the individual audit events Shield's edges emit — every block, detect-mode and shadow-mode finding, every enforced DLP redaction, plus a sample of allowed traffic. Events flow from each edge sidecar to the central ClickHouse table (elchi_shield_audit) and are served to the UI through the backend (/api/v3/shield/events). The tab is gated to Admin and Owner roles.
Where the Overview dashboard shows rates and trends, this feed answers the forensic question: which request, on which edge, was blocked by which engine and rule, and why.
What an event contains — and what it never contains
Each row carries the event metadata only: timestamp, action, severity, engine, rule id, policy id, method, host, path, response status code, the edge identity (instance, node id, listener), request id (Envoy x-request-id), processing phase and direction, the active config version, and a short reason string.
Edges running elchi-shield 1.2.0 or newer add the scoring trail of the Coraza WAF: for a CRS decision the rule_id is only the blocking-evaluation rule (949110 inbound, 959100 outbound — "anomaly score exceeded"), so the event also lists which rules filled the score — each with its id, message template, severity, the points it added, how many variables it matched, its paranoia level and rule file — plus the request's anomaly score and the threshold in force. The trail is rule metadata only; the matched data (the payload) is never exported. Allowed requests carry it too: a request that scored 3 against a threshold of 5 is recorded as allow with its rule and 3/5.
Edges running elchi-shield 1.3 or newer also name what each rule matched — the variable, never its value: ARGS:id, ARGS:json.user.password, REQUEST_HEADERS:user-agent. A name that could itself carry data is cut back: a *_NAMES collection keeps only the collection (there the "key" is the matched content, e.g. a payload sent as a parameter name), and a key that is not name-shaped (letters, digits, _ . - [ ], at most 64 characters, at least one letter) or that looks like an email, a card number or a token becomes * (ARGS:*). At most 8 names per rule.
Add exception… on a matched rule turns a false positive into a fix: pick what the rule should skip (one of the variables it matched, or the whole rule) and where (paths starting with the event's path, exactly that path, or everywhere). The policy whose WAF decided the event opens with the rule exception added to that WAF — the route's, the domain's or the project defaults' — unsaved, for you to review and save. If the policy changed since the event (the route or the WAF is gone), the dialog says so instead.
Audit events never contain header or body values — no tokens, cookies, payloads, or query strings. The path is stored query-stripped, and engine reason strings are structured descriptions, never request content. What lands in ClickHouse is safe to retain and share: enough to attribute and investigate a finding, nothing that leaks the request itself.
Two things to remember when reading counts:
- Findings are always audited — every block, detect and shadow event is recorded; redact events (enforced DLP redactions) are capped per edge at 50/s by default, as a redacting route rewrites every matching response.
- Allowed traffic is sampled (per-policy
sampling_rate, 5% by default, plus an optional global cap), soallowrows are a sample, not a total. Use the Overview dashboard or Prometheus metrics for exact request tallies. - Rows expire from ClickHouse by TTL (7 days by default) — see Metrics, Audit & Health for the schema and retention.
Filters
The filter bar drives both the feed and the summary, server-side:
- Time range — quick relative ranges from Last 15 minutes to Last 30 days, or a custom range with times. A relative range is live: Refresh (and the Live auto-refresh) slides the window forward to "now". Manually editing the range pins it, and refreshes re-query the same fixed window.
- Engine — one of the engines that produced findings (the dropdown is built from real data in the window; e.g.
coraza,jwt,bot,ratelimit,ipreputation,dlp,body_size,body_decode,anomaly, …). - Action —
block,detect,shadow,redact(a DLP redaction in block mode: the request was allowed with the named kind masked; shield v1.3.2+), orallow. - Severity —
critical,high,medium,low,info. - Edge (node) — a specific edge's node id.
- Findings only — on by default: keeps blocks, detections and shadow findings and — on edges that record the scoring trail — the allowed requests that scored under the threshold (they are findings too); the clean, sampled allow stream is hidden. Switch it off to see that stream.
- Rule — set by clicking a rule in the Top rules card (or the
rule=query parameter): events involving that rule, i.e. rows whoserule_idit is or whose scoring trail contains it — so942100finds the CRS blocks it scored in, whoserule_idis949110. - Domain — set by a rule count in the WAF Studio (or the
pol=query parameter, the domain's hosts joined by,followed by|): the events the domain's policies decided, its routes included. Remove it with its chip. - Quick search — a client-side filter on host / path / request id, applied to the currently loaded page only (paging is collapsed while a search is active).
Draft filter edits are applied with Apply (the button lights up when the draft differs from what's active); Clear returns to the defaults.
Summary, chart, and facets
Above the feed:
- Summary cards — Total, Blocked, Detected, Shadow and Redacted counts for the filtered window, computed server-side from
engine × action × severitygroups over the same filter as the feed (so the total is exact, not an estimate). - Activity over time — a stacked area chart with one series per action, bucketed over the window.
- Top engines — the engines producing the most events in the window.
- Top rules — the rules behind the window's decisions, ranked by the events they took part in: for the WAF the CRS rules that filled the anomaly score (from the scoring trail), for the other engines their rule id. Click a rule to filter the feed on it. Empty until the edges run shield 1.2.0+.
The engine and edge dropdowns are facets: they list the distinct values actually present in the selected time window, so you never filter on an engine that produced nothing.
The feed
The table pages 50 events at a time, newest first: time, action, severity, engine, the request (method host+path), status code, reason (with the rule id in the tooltip; on a trail-carrying event a chip such as 3 rules · 25/5 — rules that scored, score over threshold), and the edge. Expanding a row reveals the remaining fields — request id, policy id, rule id, node id, listener, instance, phase, direction, config version, and the full reason — each copyable for cross-referencing with edge logs (the request id matches the request_id in Shield's per-request decision log).
The expanded row also shows Matched rules when the edge recorded them:
-
a header
score N / threshold M;under thresholdon an allowed request that scored, orblocked directly by rule <id> (no anomaly score)when a rule denied on its own (for example200002, Coraza's "Failed to parse request body" — adenythat writes no score, so the total can be0/5on a block); -
one line per rule: id, severity, the points it added (
+10 (×2)when the rule matched two variables — the CRS scores once per matched variable, so the lines add up to the score), paranoia level, message and file; adenybadge instead of points for a rule that blocked directly; -
no CRS rule matchedon a clean, sampled allow;not recorded by this edge (shield < 1.2.0)for rows written by older edges. -
Live — a toggle that auto-refreshes every 15 seconds while the tab is visible, sliding a relative window forward each tick.
-
CSV — exports the filtered result set (up to 500 rows) client-side, with the trail as
matched_rule_ids(;-separated),anomaly_scoreandanomaly_thresholdcolumns. Only the redacted fields are exported, and cells are neutralized against spreadsheet formula injection.
When to use it vs the Overview dashboard
Reach for Security Events when you have a what happened question: verifying that a new policy blocks the right things, tracing a user-reported 403 by request id, reviewing what a shadow-mode policy would have blocked before enforcing it, or pulling evidence for an incident. Reach for the Overview dashboard when you have a how much / how healthy question — rates, latency, and edge health. The two are designed as a pair: spot the spike on Overview, explain it here.
Video tutorial
Short, hands-on videos showing this feature working for real. English audio, English and Turkish subtitles. All video tutorials →