Skip to main content

BowerSecurity_CL columns

Every Sentinel path writes to one table, BowerSecurity_CL. SentinelSchema.Default in src/Bower.Dcr defines the columns, and deploy/bicep/main.bicep and the Azure Monitor Agent rule (deploy/bicep/ama-syslog-dcr.bicep) follow it. Tests fail when any of them differ, and a test checks that this page lists the same columns in the same order.

The schema only ever grows: no column is renamed or retyped, and schemaVersion stays 1.0.0.

ColumnTypeFrom the envelope
TimeGenerateddatetimetimeGenerated
EventIdstringeventId
EventOriginalIdstringeventOriginalId
EventCategorystringeventCategory
EventTypestringeventType
EventActionstringeventAction
EventResultstringeventResult
EventSeveritystringeventSeverity
ApplicationNamestringapplication.name
ApplicationEnvironmentstringapplication.environment
ActorUserIdstringactor.userId
ActorUsernamestringactor.username
SourceIpAddressstringsource.ipAddress
CorrelationIdstringrequest.correlationId
PolicyIdstringsecurity.policyId
PolicyVersionstringsecurity.policyVersion
PolicyHashstringsecurity.policyHash
ValueScoreintsecurity.valueScore
CollectorIdstringcollector.id
PrivacyDetecteddynamicprivacy.detected
Labelsdynamiclabels
RawEnvelopedynamicthe whole envelope
TimeObserveddatetimetimeObserved
EventOutcomeReasonstringeventOutcomeReason
Targetdynamictarget
Changedynamicchange
Attributesdynamicattributes
Origindynamiccollector.origin, or the agent's host on the Azure Monitor Agent path

Origin​

Origin says where an event entered Bower, so a row can be traced to the system that wrote it during an investigation. The producer cannot set it. Bower rebuilds collector on every event, so a value an application puts there is replaced.

Origin.kindSet byFields
cloudwatch-logsthe collector, from the cloud forwarder's X-Bower-Origin headerlogGroup, logStream (the CloudWatch Logs group and stream the line was written to, for example the instance ID)
ama-syslogthe collection rule, from the row Azure Monitor Agent madehost (the agent's Computer)
(empty)any other path: the producer is already identified by its ingest token
  • cloudwatch-logs. The collector honours the header only from producers listed in BOWER_ORIGIN_PRODUCERS, which should be the bower-cloud agent's tokens-file producer ID. A header from any other producer is ignored. The log group and stream identify the server that wrote the line only as far as your IAM policies allow: limit logs:PutLogEvents to each server's own stream (see the CloudWatch agent quick start).
  • ama-syslog. The host is the machine the agent ran on, not a value in the message. A local account can still send a forged message on that host (see the trust boundary), but the row names the host it came from.

Live on 10-11-2026 (#249), Azure Monitor Agent 1.45.1 on Ubuntu 24.04: a row sent through ama-syslog had Origin {"kind":"ama-syslog","host":"<the VM>"}. A row forged with logger by nobody, whose envelope claimed a CloudWatch Logs origin, had the same Origin: the agent's real host, not the claim. A record uploaded through the Logs Ingestion rule kept its collector.origin (cloudwatch-logs, log group and stream) in Origin.

Example query, rows per server in the last day:

BowerSecurity_CL
| where TimeGenerated > ago(1d)
| summarize Rows = count() by Kind = tostring(Origin.kind),
Server = coalesce(tostring(Origin.host), tostring(Origin.logStream))

The two Azure paths, compared​

Live on 11-11-2026 (#249), one VM sent the same envelope once through ama-syslog and once through azure-logs-ingestion (managed identity) into the same table. Every typed column was equal, including TimeGenerated to the millisecond, the empty strings in EventOriginalId and ActorUserId, nulls in the dynamic columns, and accented text in EventOutcomeReason. The differences:

Columnama-syslogazure-logs-ingestion
Origin{"kind":"ama-syslog","host":"<the VM>"}empty
_ResourceIdthe VM's resource IDempty
RawEnvelopeonly the keys the envelope hasalso eventOriginalId ("") and labels, privacy, target, change, attributes (null)

The RawEnvelope difference is the Logs Ingestion stream declaring every column: pack_all() packs declared columns that are absent too. Read optional envelope fields with isempty() or isnull() rather than testing for the key, and the query works on both paths.