What are datasets?

Learn how to import, delete, and interact with custom or third-party datasets in Cortex Cloud.

Prerequisite

Dataset Management requires View/Edit RBAC permissions for Data Management (under Configurations → Data Management), which are the same permissions required for Parsing Rules, Data Model Rules, and Event Forwarding.

Cortex Cloud runs every Cortex Query Language (XQL) query against a dataset. A dataset is a collection of column:value sets. If you do not specify a dataset in your query, Cortex Cloud runs the query against the default datasets configured, which is by default xdr_data. The xdr_data dataset contains all of the endpoint and network data that Cortex Cloud collects. You can always change the default datasets using the set to default option. You can also upload datasets as a CSV, TSV, or JSON file that contains the data you are interested in querying. These uploaded datasets are called lookup datasets.

It's also possible to create dataset views, which provide a virtual representation of data from one or more datasets, based on the Cortex Query Language (XQL) query defined. Dataset views enhance data efficiency and security. For example, by segregating data for specific user needs or access privileges through the Role-based access control (RBAC) settings. For more information, see Dataset views.

To query other datasets, you have the following options:

• Set a dataset as default, which enables you to query the datasets without specifying them in the query.
• Name a specific dataset at the beginning of your query with the dataset stage command.

Dataset types

The type of dataset is based on the method used to upload the data. The possible types include:

• Correlation: A dataset containing data saved from a correlation rule.
• Lookup: A dataset containing key-value pairs that can be used as a reference to correlate to events. For example, a user list with corresponding access privileges. You can import or create a lookup dataset, and then reference the values for a certain key, run queries and take action. For more information, see Lookup datasets.
• Raw: Every dataset where PANW data is ingested out-of-the-box or third-party data is ingested using a configured dedicated collector. The schema for the raw dataset is automatically generated based on the log data collected by Cortex Cloud and the data format sent, such as JSON, CEF, and LEEF.
• Snapshot: A dataset that contains only the last successful snapshot of the data, such as Workday or ServiceNow CMDB tables.
• System: Cortex Cloud datasets that are created out-of-the-box.
• User: If saved by a query using the target command, the Type can be either User or Lookup.

Dataset parsers

To ensure accurate visibility and threat detection within Cortex Cloud, the platform utilizes a variety of specialized dataset parsers. The selection and effectiveness of these parsers are directly dependent on the specific data sources ingested, encompassing both the source data (the original log format generated by the device) and the integration configuration. By identifying whether logs arrive as structured JSON, standardized security formats like CEF and LEEF, or unstructured raw text, Cortex Cloud can properly parse the information for analysis.

Supported parser types

• CEF (Common Event Format): A parser for logs formatted in the ArcSight CEF standard. CEF logs contain a pipe-delimited header with vendor, product, version, event class, name, and severity fields, followed by key-value pair extensions. Commonly used by vendors like Check Point, Fortinet, and Zscaler.
• Cisco ASA: A parser for specific Cisco Adaptive Security Appliance (ASA) syslog messages. Supports logs like connection-related messages (Built/Teardown) and AnyConnect VPN events. Only a subset of Cisco ASA message types is supported.
• Corelight: A parser for network traffic logs generated by Corelight sensors (based on Zeek/Bro). Processes structured JSON logs containing network connection metadata such as connection records, DNS queries, and HTTP transactions.
• Filebeat: A parser for logs collected and forwarded using Elastic Filebeat agents. Extracts the log payload from the Filebeat JSON envelope, handling metadata fields such as timestamps, agent information, and host details.
• JSON: A general-purpose parser for logs sent in standard JSON format. This is the most common format for third-party data sources that send structured data, including cloud services, SaaS applications, and API-based integrations.
• LEEF (Log Event Extended Format): A parser for the IBM QRadar LEEF standard. LEEF logs contain a tab-delimited or custom-delimited header with version, vendor, product, product version, and event ID fields, followed by key-value pair attributes. Typically used by IBM security products and QRadar-integrated vendors.

  Note

  CEF and LEEF logs are often wrapped in a Syslog envelope (RFC 3164/5424). Cortex Cloud automatically detects and strips the Syslog header to extract the reporting device IP and hostname before parsing the inner CEEF or LEEF payload.
• Raw Text: A parser for unstructured, plain-text log data that does not conform to any specific structured format. The raw text is ingested as-is and can be processed using parsing rules. This is also the default parser used when the log format cannot be identified or is not explicitly defined.
• WEC (Windows Event Collection): A parser for Windows Event logs collected via Windows Event Forwarding (WEF/WEC). Parses XML-formatted Windows Event logs and converts them into a structured format. Supports Windows Security, System, and Application event logs.
• Windows DNS Debug: A parser for Microsoft Windows DNS Server debug log files. Extracts DNS query and response details including query type, remote IP, protocol, response code, and question name from the Windows DNS debug log format.
• Winlogbeat: A parser for Windows Event logs collected using Elastic Winlogbeat agents. Similar to the WEC parser but handles the Winlogbeat-specific JSON envelope format, extracting Windows Event data along with associated metadata.

Datasets in XQL

Important

By default, forensic datasets are not included in XQL query results, unless the dataset query is explicitly defined to use a forensic dataset.

Cortex Query Language (XQL) supports using different languages for dataset and field names. In addition, when setting up your XQL query, it is important to keep in mind the following:

• The dataset formats supported are dependent on the data retention offerings available in Cortex Cloud according to whether you want to query hot storage or cold storage.

  • Hot Storage queries are performed on a dataset using the format dataset = <dataset name>. This is the default option.

    dataset = xdr_data

  • Cold Storage queries are performed using the format cold_dataset = <dataset name>.

    cold_dataset = xdr_data

• Dataset refresh times: While most out-of-the-box system datasets are ingested in near real-time, the following datasets have specific refresh schedules.

  • endpoints: Refreshed every hour.
  • pan_dss_raw: Refreshed daily.
  • Forensics datasets: Data collection behavior depends on your Agent Settings profile.

    • Default: Data is collected as a one-time snapshot and does not update.
    • Scheduled: If you specify a collection interval, the value represents the number of hours between updates, such as an interval of 24 equals once per day.
    • Minimum: The shortest allowable interval is 12 hours.
• Query against a dataset by selecting it with the dataset command when you create an XQL query. For more information, see Create XQL query.
• After your query runs, you can always save your query results as a dataset. You can use the target stage command to save query results as a dataset.
• Schema changes to datasets may not be reflected in the autocomplete suggestions and definitions as you type in real time the XQL query and can appear with a slight delay.

Managing datasets and dataset views

You can manage your datasets and dataset views in Cortex Cloud from the Settings → Configurations → Data Management → Dataset Management page.

Below are some of the main tasks available for all dataset types by right-clicking a particular dataset or dataset view listed in either the Datasets or Dataset Views table. Only tasks that need further explanation are explained below. Datasets and dataset views can only be deleted if there are no other dependencies. For example, if a Correlation Rule is based on a dataset or dataset view or dataset view, you wouldn't be able to delete the dataset or dataset view until you removed the dataset view from the XQL query of the Correlation Rule.

Note

For more information on tasks specific to lookup datasets, see Lookup datasets.