elasticsearch-audit（弹性搜索审计）

Elasticsearch Audit Logging

Enable and configure security audit logging for Elasticsearch via the cluster settings API. Audit logs record security events such as authentication attempts, access grants and denials, role changes, and API key operations — essential for compliance and incident investigation.

For Kibana audit logging (saved object access, login/logout, space operations), see kibana-audit. For authentication and API key management, see elasticsearch-authn. For roles and user management, see elasticsearch-authz. For diagnosing security errors, see elasticsearch-security-troubleshooting.

For detailed API endpoints and event types, see references/api-reference.md.

Deployment note: Audit logging configuration differs across deployment types. See Deployment Compatibility for details.

Jobs to Be Done

• Enable or disable security audit logging on a cluster
• Select which security events to record (authentication, access, config changes)
• Create filter policies to reduce audit log noise
• Query audit logs for failed authentication attempts
• Investigate unauthorized access or privilege escalation incidents
• Set up compliance-focused audit configuration
• Detect brute-force login patterns from audit data
• Configure audit output to an index for programmatic querying

Prerequisites

Prompt the user for any missing values.

Enable Audit Logging

Enable audit logging dynamically without a restart:

curl -X PUT "${ELASTICSEARCH_URL}/_cluster/settings" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "persistent": {
      "xpack.security.audit.enabled": true
    }
  }'

To disable, set xpack.security.audit.enabled to false. Verify current state:

curl "${ELASTICSEARCH_URL}/_cluster/settings?include_defaults=true&flat_settings=true" \
  <auth_flags> | jq '.defaults | with_entries(select(.key | startswith("xpack.security.audit")))'

Audit Output

Audit events can be written to two outputs. Both can be active simultaneously.

Configure output via API

curl -X PUT "${ELASTICSEARCH_URL}/_cluster/settings" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "persistent": {
      "xpack.security.audit.enabled": true,
      "xpack.security.audit.outputs": ["index", "logfile"]
    }
  }'

The index output is required for programmatic querying of audit events. The logfile output is useful for shipping to external SIEM tools via Filebeat.

Note: On self-managed clusters, xpack.security.audit.outputs may require a static setting in elasticsearch.yml on older versions (pre-8.x). On 8.x+, prefer the cluster settings API.

Select Events to Record

Control which event types are included or excluded. By default, all events are recorded when audit is enabled.

Include specific events only

curl -X PUT "${ELASTICSEARCH_URL}/_cluster/settings" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "persistent": {
      "xpack.security.audit.logfile.events.include": [
        "authentication_failed",
        "access_denied",
        "access_granted",
        "anonymous_access_denied",
        "tampered_request",
        "run_as_denied",
        "connection_denied"
      ]
    }
  }'

Exclude noisy events

curl -X PUT "${ELASTICSEARCH_URL}/_cluster/settings" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "persistent": {
      "xpack.security.audit.logfile.events.exclude": [
        "access_granted"
      ]
    }
  }'

Excluding access_granted significantly reduces log volume on busy clusters — use this when only failures matter.

Event types reference

See references/api-reference.md for the complete event type list with field details.

Filter Policies

Filter policies let you suppress specific audit events by user, realm, role, or index without disabling the event type globally. Multiple policies can be active — an event is logged only if no policy filters it out.

Ignore system and internal users

curl -X PUT "${ELASTICSEARCH_URL}/_cluster/settings" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "persistent": {
      "xpack.security.audit.logfile.events.ignore_filters": {
        "system_users": {
          "users": ["_xpack_security", "_xpack", "elastic/fleet-server"],
          "realms": ["_service_account"]
        }
      }
    }
  }'

Ignore health-check traffic on specific indices

curl -X PUT "${ELASTICSEARCH_URL}/_cluster/settings" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "persistent": {
      "xpack.security.audit.logfile.events.ignore_filters": {
        "health_checks": {
          "users": ["monitoring-user"],
          "indices": [".monitoring-*"]
        }
      }
    }
  }'

Filter policy fields

An event is filtered out if it matches all specified fields within a single policy.

Remove a filter policy

Set the policy to null:

curl -X PUT "${ELASTICSEARCH_URL}/_cluster/settings" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "persistent": {
      "xpack.security.audit.logfile.events.ignore_filters.health_checks": null
    }
  }'

Query Audit Events

When the index output is enabled, audit events are stored in .security-audit-* indices and can be queried.

Search for failed authentication attempts

curl -X POST "${ELASTICSEARCH_URL}/.security-audit-*/_search" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "query": {
      "bool": {
        "filter": [
          { "term": { "event.action": "authentication_failed" } },
          { "range": { "@timestamp": { "gte": "now-24h" } } }
        ]
      }
    },
    "sort": [{ "@timestamp": { "order": "desc" } }],
    "size": 50
  }'

Search for access denied events on a specific index

curl -X POST "${ELASTICSEARCH_URL}/.security-audit-*/_search" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "query": {
      "bool": {
        "filter": [
          { "term": { "event.action": "access_denied" } },
          { "term": { "indices": "logs-*" } },
          { "range": { "@timestamp": { "gte": "now-7d" } } }
        ]
      }
    },
    "sort": [{ "@timestamp": { "order": "desc" } }],
    "size": 20
  }'

Search for security configuration changes

curl -X POST "${ELASTICSEARCH_URL}/.security-audit-*/_search" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "query": {
      "bool": {
        "filter": [
          { "term": { "event.action": "security_config_change" } },
          { "range": { "@timestamp": { "gte": "now-7d" } } }
        ]
      }
    },
    "sort": [{ "@timestamp": { "order": "desc" } }],
    "size": 50
  }'

This captures role creation/deletion, user changes, API key operations, and role mapping updates.

Count events by type and detect brute-force patterns

Use terms aggregations on event.action (with size: 0) to count events by type over a time window. To detect brute-force attempts, aggregate authentication_failed events by source.ip with min_doc_count: 5. See references/api-reference.md for full aggregation query examples.

Correlate with Kibana Audit Logs

Kibana has its own audit log covering application-layer events that Elasticsearch does not see (saved object CRUD, Kibana logins, space operations). When a user performs an action in Kibana, Kibana makes requests to Elasticsearch on the user's behalf. Both systems record the same trace.id (passed via the X-Opaque-Id header), which serves as the primary correlation key.

Prerequisite: Kibana audit must be enabled separately in kibana.yml. See the kibana-audit skill for setup instructions, event types, and Kibana-specific filter policies.

Find ES audit events triggered by a Kibana action

Given a trace.id from a Kibana audit event, search the ES audit index to see the underlying Elasticsearch operations:

curl -X POST "${ELASTICSEARCH_URL}/.security-audit-*/_search" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "query": {
      "bool": {
        "filter": [
          { "term": { "trace.id": "'"${TRACE_ID}"'" } },
          { "range": { "@timestamp": { "gte": "now-24h" } } }
        ]
      }
    },
    "sort": [{ "@timestamp": { "order": "asc" } }]
  }'

Correlate by user and time window

When trace.id is unavailable (e.g. direct API calls), fall back to user + time-window correlation:

curl -X POST "${ELASTICSEARCH_URL}/.security-audit-*/_search" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "query": {
      "bool": {
        "filter": [
          { "term": { "user.name": "'"${USERNAME}"'" } },
          { "range": { "@timestamp": { "gte": "now-5m" } } }
        ]
      }
    },
    "sort": [{ "@timestamp": { "order": "asc" } }]
  }'

Secondary correlation fields: user.name, source.ip, and @timestamp.

Unified querying

Ship Kibana audit logs to Elasticsearch via Filebeat (see kibana-audit for the Filebeat config) so that both .security-audit-* (ES) and kibana-audit-* (Kibana) indices can be searched together in a single multi-index query filtered by trace.id.

Examples

Enable audit logging for compliance

Request: "Enable audit logging and record all failed access and authentication events."

curl -X PUT "${ELASTICSEARCH_URL}/_cluster/settings" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "persistent": {
      "xpack.security.audit.enabled": true,
      "xpack.security.audit.logfile.events.include": [
        "authentication_failed",
        "access_denied",
        "anonymous_access_denied",
        "run_as_denied",
        "connection_denied",
        "tampered_request",
        "security_config_change"
      ]
    }
  }'

This captures all denial and security change events while excluding high-volume success events.

Investigate a suspected unauthorized access attempt

Request: "Someone may have tried to access the secrets-* index. Check the audit logs."

curl -X POST "${ELASTICSEARCH_URL}/.security-audit-*/_search" \
  <auth_flags> \
  -H "Content-Type: application/json" \
  -d '{
    "query": {
      "bool": {
        "filter": [
          { "terms": { "event.action": ["access_denied", "authentication_failed"] } },
          { "wildcard": { "indices": "secrets-*" } },
          { "range": { "@timestamp": { "gte": "now-48h" } } }
        ]
      }
    },
    "sort": [{ "@timestamp": { "order": "desc" } }],
    "size": 100
  }'

Review user.name, source.ip, and event.action in the results to identify the actor and pattern.

Reduce audit noise on a busy cluster

Request: "Audit logs are too large. Filter out monitoring traffic and successful reads."

Exclude access_granted from event types, then add a filter policy for monitoring users and indices. See Filter Policies for the full syntax.

Guidelines

Prefer index output for programmatic access

Enable the index output to make audit events queryable. The logfile output is better for shipping to external SIEM tools via Filebeat but cannot be queried through the Elasticsearch API.

Start restrictive, then widen

Begin with failure events only (authentication_failed, access_denied, security_config_change). Add success events only when needed — they generate high volume.

Use filter policies instead of disabling events

Suppress specific users or indices with filter policies rather than excluding entire event types.

Monitor audit index size

Set up an ILM policy to roll over and delete old .security-audit-* indices. A 30-90 day retention is typical.

Enable Kibana audit for full coverage

For application-layer events (saved object access, Kibana logins, space operations), enable Kibana audit logging as well. See the kibana-audit skill for setup. Use trace.id to correlate — see Correlate with Kibana Audit Logs above.

Avoid superuser credentials

Use a dedicated admin user or API key with manage privileges. Reserve elastic for emergency recovery only.

Deployment Compatibility

ECH notes: ES audit is configured via the cluster settings API. Logfile output is accessible through the Cloud console deployment logs. Index output works the same as self-managed.

Serverless notes:

• Audit logging is not user-configurable on Serverless. Security events are managed by Elastic as part of the platform.
• If a user asks about auditing on Serverless, direct them to the Elastic Cloud console or their account team.