v1.4.0 Release Notes
This release upgrades the Logster application only. The classifier model and the Linux sensors stay as they are. Windows endpoints need a configuration change only if you want PowerShell detection (see PowerShell script detection below).
What's new
Connected-component scoring (optional)
A time window often holds several unrelated chains of activity. In connected-component mode, Logster analyses each chain separately, so a single suspicious command is judged on its own instead of alongside unrelated benign activity. The graph view now highlights the chain that decided an attack (Show deciding component).
The mode is off by default because it makes roughly three times as many classifier calls. You can turn it on with one Helm command:
helm upgrade logster ./logster-<ver>.tgz \
-n logster \
--reset-then-reuse-values \
--set inference.scoringMode=connected_component \
--timeout 20m
To return to the default, run the same command with
--set inference.scoringMode= (empty value). With Helm older than 3.14,
replace --reset-then-reuse-values with -f logster-values-backup.yaml.
Verdict cache
When activity the model has already judged benign appears again unchanged, Logster reuses that verdict instead of re-sending it to the classifier. Only benign verdicts are reused, an attack on an endpoint clears its cache, and every reused verdict is re-checked at least every two hours. The graph view marks reused activity (Show reused nodes).
The verdict cache is on by default. To turn it off:
helm upgrade logster ./logster-<ver>.tgz \
-n logster \
--reset-then-reuse-values \
--set inference.verdictCacheMode=off \
--timeout 20m
Subtree whitelisting
A rule on a process can now also exclude everything the process starts: child processes, scripts, files, DNS lookups and connections. Tick "also exclude everything it starts" when you create the rule.
Note
The subtree option requires an exact rule. Pattern rules cannot use it, because a pattern could also match an attacker's command line that only mentions your script.
See Whitelisting for the full guide.
PowerShell script detection
Logster now analyses PowerShell Script Block Logging (Event ID 4104) alongside Sysmon. Each script appears in the graph as a script node, linked to the process that ran it, and whitelist rules can match script content.
To enable it on Windows endpoints, turn on Script Block Logging and, if
PowerShell 7 (pwsh) is installed, its separate policy. Then restart the
Splunk Universal Forwarder. Endpoints with PowerShell 7 also need the
updated logster_forwarder app, which adds the
PowerShellCore/Operational input.
The full steps are in Enable PowerShell Script Block Logging.
Reliability
- Vector memory — Vector now caps how much it reads ahead from Kafka, keeping its memory use steady while it catches up on a large backlog.
- Whitelist matching budget — Graph whitelist matching has a time limit, so a complex Strict or Shape entry can no longer stall the analysis.